deep_ensemble_with_uncertainty_and_spec_fe_eval.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "177fdb80",
   "metadata": {},
   "outputs": [],
   "source": [
    "import codecs\n",
    "import copy\n",
    "import gc\n",
    "import json\n",
    "import logging\n",
    "import os\n",
    "import pickle\n",
    "import random\n",
    "import time\n",
    "from typing import List, Tuple, Union"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "4a1922ce",
   "metadata": {},
   "outputs": [],
   "source": [
    "os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "8920a50b",
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "from matplotlib.colors import Normalize\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from sklearn.compose import ColumnTransformer\n",
    "from sklearn.decomposition import PCA\n",
    "from sklearn.feature_selection import VarianceThreshold\n",
    "from sklearn.impute import SimpleImputer\n",
    "from sklearn.metrics import mean_absolute_error, mean_squared_error\n",
    "from sklearn.metrics import median_absolute_error, r2_score\n",
    "from sklearn.model_selection import StratifiedKFold\n",
    "from sklearn.pipeline import Pipeline\n",
    "from sklearn.preprocessing import StandardScaler, MinMaxScaler\n",
    "from sklearn.preprocessing import Binarizer, KBinsDiscretizer\n",
    "import tensorflow as tf\n",
    "import tensorflow_addons as tfa\n",
    "import tensorflow_probability as tfp\n",
    "import umap"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "ab715c57",
   "metadata": {},
   "outputs": [],
   "source": [
    "from assessment import calc_uncertainty_regection_curve, f_beta_metrics\n",
    "from uncertainty import ensemble_uncertainties_regression"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "95df40bb",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CatBoost is installed\n"
     ]
    }
   ],
   "source": [
    "try:\n",
    "    from catboost import CatBoostRegressor\n",
    "    CATBOOST_EXISTS = True\n",
    "    print('CatBoost is installed')\n",
    "except:\n",
    "    CatBoostRegressor = None\n",
    "    CATBOOST_EXISTS = False\n",
    "    print('CatBoost is NOT installed')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "740cad7a",
   "metadata": {},
   "outputs": [],
   "source": [
    "def build_preprocessor(X: np.ndarray, colnames: List[str]) -> Pipeline:\n",
    "    X_ = Pipeline(steps=[\n",
    "        (\n",
    "            'imputer', SimpleImputer(\n",
    "                missing_values=np.nan, strategy='constant',\n",
    "                fill_value=-1.0\n",
    "            )\n",
    "        ),\n",
    "        (\n",
    "            'scaler',\n",
    "            MinMaxScaler()\n",
    "        )\n",
    "    ]).fit_transform(X)\n",
    "    X_ = np.rint(X_ * 100000.0).astype(np.int32)\n",
    "    binary_features = dict()\n",
    "    categorical_features = dict()\n",
    "    removed_features = []\n",
    "    for col_idx in range(X.shape[1]):\n",
    "        values = set(X_[:, col_idx].tolist())\n",
    "        print(f'Column {col_idx} \"{colnames[col_idx]}\" has ' \\\n",
    "              f'{len(values)} unique values.')\n",
    "        if len(values) > 1:\n",
    "            if len(values) < 3:\n",
    "                binary_features[col_idx] = np.min(X[:, col_idx])\n",
    "            else:\n",
    "                categorical_features[col_idx] = len(values)\n",
    "        else:\n",
    "            removed_features.append(col_idx)\n",
    "        del values\n",
    "    del X_\n",
    "    all_features = set(range(X.shape[1]))\n",
    "    useful_features = sorted(list(all_features - set(removed_features)))\n",
    "    if len(useful_features) == 0:\n",
    "        raise ValueError('Training inputs are bad. All features are removed.')\n",
    "    print(f'There are {X.shape[1]} features.')\n",
    "    if len(removed_features) > 0:\n",
    "        print(f'These features will be removed: ' \\\n",
    "              f'{[colnames[col_idx] for col_idx in removed_features]}.')\n",
    "    transformers = []\n",
    "    if (len(categorical_features) > 0) and (len(binary_features) > 0):\n",
    "        print(f'There are {len(categorical_features)} categorical ' \\\n",
    "              f'features and {len(binary_features)} binary features.')\n",
    "    elif len(categorical_features) > 0:\n",
    "        print(f'There are {len(categorical_features)} categorical features.')\n",
    "    else:\n",
    "        print(f'There are {len(binary_features)} binary features.')\n",
    "    for col_idx in categorical_features:\n",
    "        n_unique_values = categorical_features[col_idx]\n",
    "        transformers.append(\n",
    "            (\n",
    "                colnames[col_idx],\n",
    "                KBinsDiscretizer(\n",
    "                    n_bins=min(max(n_unique_values // 3, 3), 256),\n",
    "                    encode='ordinal',\n",
    "                    strategy=('quantile' if n_unique_values > 50 else 'kmeans')\n",
    "                ),\n",
    "                (col_idx,)\n",
    "            )\n",
    "        )\n",
    "    for col_idx in binary_features:\n",
    "        transformers.append(\n",
    "            (\n",
    "                colnames[col_idx],\n",
    "                Binarizer(threshold=0.0),\n",
    "                (col_idx,)\n",
    "            )\n",
    "        )\n",
    "    preprocessor = Pipeline(steps=[\n",
    "        (\n",
    "            'imputer', SimpleImputer(\n",
    "                missing_values=np.nan, strategy='constant',\n",
    "                fill_value=-1.0\n",
    "            )\n",
    "        ),\n",
    "        (\n",
    "            'minmax_scaler',\n",
    "            MinMaxScaler()\n",
    "        ),\n",
    "        (\n",
    "            'composite_transformer', ColumnTransformer(\n",
    "                transformers=transformers,\n",
    "                sparse_threshold=0.0,\n",
    "                n_jobs=1\n",
    "            )\n",
    "        ),\n",
    "        (\n",
    "            'selector',\n",
    "            VarianceThreshold()\n",
    "        ),\n",
    "        (\n",
    "            'standard_scaler',\n",
    "            StandardScaler(with_mean=True, with_std=True)\n",
    "        ),\n",
    "        (\n",
    "            'pca',\n",
    "            PCA(random_state=42)\n",
    "        )\n",
    "    ])\n",
    "    return preprocessor.fit(X)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "7c14606d",
   "metadata": {},
   "outputs": [],
   "source": [
    "def reduce_dimensions_of_data(features: np.ndarray) -> np.ndarray:\n",
    "    preprocessed_features = Pipeline(\n",
    "        steps=[\n",
    "            ('scaler', StandardScaler()),\n",
    "            ('pca', PCA(n_components=features.shape[1] // 3,\n",
    "                        random_state=42))\n",
    "        ]\n",
    "    ).fit_transform(features)\n",
    "    print('Features are preprocessed.')\n",
    "    reduced_features = umap.UMAP(\n",
    "        low_memory=False,\n",
    "        n_jobs=-1,\n",
    "        random_state=42,\n",
    "        verbose=True\n",
    "    ).fit_transform(preprocessed_features)\n",
    "    print('Feature space is reduced.')\n",
    "    del preprocessed_features\n",
    "    return reduced_features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "4320c5db",
   "metadata": {},
   "outputs": [],
   "source": [
    "def show_temperature(features: np.ndarray, targets: np.ndarray,\n",
    "                     title: str='', figure_id: int=0):\n",
    "    if features.shape[0] != targets.shape[0]:\n",
    "        err_msg = f'Features do not correspond to targets! ' \\\n",
    "                  f'{features.shape[0]} != {targets.shape[0]}'\n",
    "        raise ValueError(err_msg)\n",
    "    if len(features.shape) != 2:\n",
    "        err_msg = f'Features are wrong! Expected 2-D array, got ' \\\n",
    "                  f'{len(features.shape)}-D one.'\n",
    "        raise ValueError(err_msg)\n",
    "    if features.shape[1] != 2:\n",
    "        err_msg = f'Features are wrong! Expected number of ' \\\n",
    "                  f'columns is 2, got {features.shape[1]}.'\n",
    "        raise ValueError(err_msg)\n",
    "    if len(targets.shape) != 1:\n",
    "        err_msg = f'Targets are wrong! Expected 1-D array, got ' \\\n",
    "                  f'{len(targets.shape)}-D one.'\n",
    "        raise ValueError(err_msg)\n",
    "    sorted_targets = sorted(targets.tolist())\n",
    "    n_percentile2 = max(int(round(0.01 * len(sorted_targets))), 1)\n",
    "    min_target = sorted_targets[n_percentile2]\n",
    "    max_target = sorted_targets[-n_percentile2]\n",
    "    del sorted_targets\n",
    "    clipped_targets = np.empty(targets.shape, dtype=np.float64)\n",
    "    for sample_idx in range(targets.shape[0]):\n",
    "        if targets[sample_idx] < min_target:\n",
    "            clipped_targets[sample_idx] = min_target\n",
    "        elif targets[sample_idx] > max_target:\n",
    "            clipped_targets[sample_idx] = max_target\n",
    "        else:\n",
    "            clipped_targets[sample_idx] = targets[sample_idx]\n",
    "    temperature_colors = clipped_targets.tolist()\n",
    "    temperature_norm = Normalize(vmin=np.min(temperature_colors),\n",
    "                                 vmax=np.max(temperature_colors))\n",
    "    fig = plt.figure(figure_id, figsize=(11, 11))\n",
    "    plt.scatter(x=features[:, 0], y=features[:, 1],\n",
    "                marker='o', cmap=plt.cm.get_cmap(\"jet\"), c=temperature_colors,\n",
    "                norm=temperature_norm)\n",
    "    if len(title) > 0:\n",
    "        plt.title(f'UMAP projections of weather data {title} (temperature)')\n",
    "    else:\n",
    "        plt.title(f'UMAP projections of weather data (temperature)')\n",
    "    plt.colorbar()\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "11cd550f",
   "metadata": {},
   "outputs": [],
   "source": [
    "def filter_dataset(y: np.ndarray) -> List[int]:\n",
    "    all_values = sorted(y.tolist())\n",
    "    n = len(all_values)\n",
    "    if n <= 10000:\n",
    "        err_msg = f'y is wrong! Expected length of y is greater than 10000, ' \\\n",
    "                  f'but got {n}.'\n",
    "        raise ValueError(err_msg)\n",
    "    y001 = all_values[int(round((n - 1) * 0.001))]\n",
    "    y999 = all_values[int(round((n - 1) * 0.999))]\n",
    "    del all_values\n",
    "    filtered_indices = list(filter(\n",
    "        lambda idx: (y[idx] > y001) and (y[idx] < y999),\n",
    "        range(n)\n",
    "    ))\n",
    "    return filtered_indices"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "2615b61b",
   "metadata": {},
   "outputs": [],
   "source": [
    "def build_neural_network(input_size: int, layer_size: int, n_layers: int,\n",
    "                         dropout_rate: float, scale_coeff: float,\n",
    "                         nn_name: str) -> tf.keras.Model:\n",
    "    feature_vector = tf.keras.layers.Input(\n",
    "        shape=(input_size,), dtype=tf.float32,\n",
    "        name=f'{nn_name}_feature_vector'\n",
    "    )\n",
    "    outputs = []\n",
    "    hidden_layer = tf.keras.layers.AlphaDropout(\n",
    "        rate=dropout_rate,\n",
    "        seed=random.randint(0, 2147483647),\n",
    "        name=f'{nn_name}_dropout1'\n",
    "    )(feature_vector)\n",
    "    for layer_idx in range(1, (2 * n_layers) // 3 + 1):\n",
    "        try:\n",
    "            kernel_initializer = tf.keras.initializers.LecunNormal(\n",
    "                seed=random.randint(0, 2147483647)\n",
    "            )\n",
    "        except:\n",
    "            kernel_initializer = tf.compat.v1.keras.initializers.lecun_normal(\n",
    "                seed=random.randint(0, 2147483647)\n",
    "            )\n",
    "        hidden_layer = tf.keras.layers.Dense(\n",
    "            units=layer_size,\n",
    "            activation='selu',\n",
    "            kernel_initializer=kernel_initializer,\n",
    "            bias_initializer='zeros',\n",
    "            name=f'{nn_name}_dense{layer_idx}'\n",
    "        )(hidden_layer)\n",
    "        hidden_layer = tf.keras.layers.AlphaDropout(\n",
    "            rate=dropout_rate,\n",
    "            seed=random.randint(0, 2147483647),\n",
    "            name=f'{nn_name}_dropout{layer_idx + 1}'\n",
    "        )(hidden_layer)\n",
    "    try:\n",
    "        kernel_initializer = tf.keras.initializers.LecunNormal(\n",
    "            seed=random.randint(0, 2147483647)\n",
    "        )\n",
    "    except:\n",
    "        kernel_initializer = tf.compat.v1.keras.initializers.lecun_normal(\n",
    "            seed=random.randint(0, 2147483647)\n",
    "        )\n",
    "    projection_layer = tf.keras.layers.Dense(\n",
    "        units=50,\n",
    "        activation=None,\n",
    "        use_bias=False,\n",
    "        kernel_initializer=kernel_initializer,\n",
    "        name=f'{nn_name}_projection'\n",
    "    )(hidden_layer)\n",
    "    for layer_idx in range((2 * n_layers) // 3 + 1, n_layers + 1):\n",
    "        try:\n",
    "            kernel_initializer = tf.keras.initializers.LecunNormal(\n",
    "                seed=random.randint(0, 2147483647)\n",
    "            )\n",
    "        except:\n",
    "            kernel_initializer = tf.compat.v1.keras.initializers.lecun_normal(\n",
    "                seed=random.randint(0, 2147483647)\n",
    "            )\n",
    "        hidden_layer = tf.keras.layers.Dense(\n",
    "            units=layer_size,\n",
    "            activation='selu',\n",
    "            kernel_initializer=kernel_initializer,\n",
    "            bias_initializer='zeros',\n",
    "            name=f'{nn_name}_dense{layer_idx}'\n",
    "        )(hidden_layer)\n",
    "        hidden_layer = tf.keras.layers.AlphaDropout(\n",
    "            rate=dropout_rate,\n",
    "            seed=random.randint(0, 2147483647),\n",
    "            name=f'{nn_name}_dropout{layer_idx + 1}'\n",
    "        )(hidden_layer)\n",
    "    try:\n",
    "        kernel_initializer = tf.keras.initializers.LecunNormal(\n",
    "            seed=random.randint(0, 2147483647)\n",
    "        )\n",
    "    except:\n",
    "        kernel_initializer = tf.compat.v1.keras.initializers.lecun_normal(\n",
    "            seed=random.randint(0, 2147483647)\n",
    "        )\n",
    "    output_layer = tf.keras.layers.Dense(\n",
    "        units=2,\n",
    "        activation=None,\n",
    "        use_bias=False,\n",
    "        kernel_initializer=kernel_initializer,\n",
    "        name=f'{nn_name}_output'\n",
    "    )(hidden_layer)\n",
    "    bayesian_layer = tfp.layers.DistributionLambda(\n",
    "        lambda t: tfp.distributions.Normal(\n",
    "            loc=t[..., :1],\n",
    "            scale=1e-6 + tf.math.softplus((1.0 / scale_coeff) * t[..., 1:])\n",
    "        ),\n",
    "        name=f'{nn_name}_distribution'\n",
    "    )(output_layer)\n",
    "    neural_network = tf.keras.Model(\n",
    "        inputs=feature_vector,\n",
    "        outputs=[bayesian_layer, projection_layer],\n",
    "        name=nn_name\n",
    "    )\n",
    "    neural_network.build(input_shape=(None, input_size))\n",
    "    return neural_network"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "489b033f",
   "metadata": {},
   "outputs": [],
   "source": [
    "def show_training_process(history: tf.keras.callbacks.History, metric_name: str,\n",
    "                          figure_id: int=1, comment: str=''):\n",
    "    val_metric_name = 'val_' + metric_name\n",
    "    if metric_name not in history.history:\n",
    "        err_msg = f'The metric \"{metric_name}\" is not found! Available metrics are: ' \\\n",
    "                  f'{list(history.history.keys())}.'\n",
    "        raise ValueError(err_msg)\n",
    "    plt.figure(figure_id, figsize=(5, 5))\n",
    "    interesting_metric = history.history[metric_name]\n",
    "    plt.plot(list(range(len(interesting_metric))), interesting_metric,\n",
    "             label=f'Training {metric_name}')\n",
    "    if val_metric_name in history.history:\n",
    "        interesting_val_metric = history.history[val_metric_name]\n",
    "        assert len(interesting_metric) == len(interesting_val_metric)\n",
    "        plt.plot(list(range(len(interesting_val_metric))),\n",
    "                 interesting_val_metric,\n",
    "                 label=f'Validation {metric_name}')\n",
    "    plt.xlabel('Epochs')\n",
    "    plt.ylabel(metric_name)\n",
    "    if len(comment) > 0:\n",
    "        plt.title(f'Training process of {comment}')\n",
    "    else:\n",
    "        plt.title('Training process')\n",
    "    plt.legend(loc='best')\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "ba630050",
   "metadata": {},
   "outputs": [],
   "source": [
    "def predict_with_single_nn(input_data: np.ndarray, model_for_prediction: tf.keras.Model,\n",
    "                           batch_size: int, output_scaler: StandardScaler) \\\n",
    "        -> Tuple[np.ndarray, np.ndarray]:\n",
    "    if len(input_data.shape) != 2:\n",
    "        err_msg = f'The `input_data` argument is wrong! Expected 2-D array, ' \\\n",
    "                  f'got {len(input_data.shape)}-D one!'\n",
    "        raise ValueError(err_msg)\n",
    "    n_batches = int(np.ceil(input_data.shape[0] / float(batch_size)))\n",
    "    pred_mean = []\n",
    "    pred_std = []\n",
    "    for batch_idx in range(n_batches):\n",
    "        batch_start = batch_idx * batch_size\n",
    "        batch_end = min(input_data.shape[0], batch_start + batch_size)\n",
    "        instant_predictions = model_for_prediction(input_data[batch_start:batch_end])[0]\n",
    "        if not isinstance(instant_predictions, tfp.distributions.Distribution):\n",
    "            err_msg = f'Minibatch {batch_idx}: predictions are wrong! ' \\\n",
    "                      f'Expected tfp.distributions.Distribution, ' \\\n",
    "                      f'got {type(instant_predictions)}.'\n",
    "            raise ValueError(err_msg)\n",
    "        instant_mean = instant_predictions.mean()\n",
    "        instant_std = instant_predictions.stddev()\n",
    "        del instant_predictions\n",
    "        if not isinstance(instant_mean, np.ndarray):\n",
    "            instant_mean = instant_mean.numpy()\n",
    "        if not isinstance(instant_std, np.ndarray):\n",
    "            instant_std = instant_std.numpy()\n",
    "        instant_mean = instant_mean.astype(np.float64).flatten()\n",
    "        instant_std = instant_std.astype(np.float64).flatten()\n",
    "        pred_mean.append(instant_mean)\n",
    "        pred_std.append(instant_std)\n",
    "        del instant_mean, instant_std\n",
    "    pred_mean = np.concatenate(pred_mean)\n",
    "    pred_std = np.concatenate(pred_std)\n",
    "    pred_mean = output_scaler.inverse_transform(\n",
    "        pred_mean.reshape((input_data.shape[0], 1))\n",
    "    ).flatten()\n",
    "    pred_std *= output_scaler.scale_[0]\n",
    "    return pred_mean, pred_std * pred_std"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "7ec86be4",
   "metadata": {},
   "outputs": [],
   "source": [
    "def evaluate_single_nn(pred_means: np.ndarray, pred_vars: np.ndarray,\n",
    "                       true_outputs: np.ndarray) -> float:\n",
    "    if len(pred_means.shape) != 1:\n",
    "        err_msg = f'The `pred_means` argument is wrong! Expected 1-D array, ' \\\n",
    "                  f'got {len(pred_means.shape)}-D one.'\n",
    "        raise ValueError(err_msg)\n",
    "    if len(pred_vars.shape) != 1:\n",
    "        err_msg = f'The `pred_vars` argument is wrong! Expected 1-D array, ' \\\n",
    "                  f'got {len(pred_vars.shape)}-D one.'\n",
    "        raise ValueError(err_msg)\n",
    "    if len(true_outputs.shape) != 1:\n",
    "        err_msg = f'The `true_outputs` argument is wrong! Expected 1-D array, ' \\\n",
    "                  f'got {len(true_outputs.shape)}-D one.'\n",
    "        raise ValueError(err_msg)\n",
    "    n_test_samples = true_outputs.shape[0]\n",
    "    if n_test_samples < 5:\n",
    "        raise ValueError(f'Number of test samples = {n_test_samples} is too small!')\n",
    "    if n_test_samples != pred_means.shape[0]:\n",
    "        err_msg = f'The `pred_means` does not correspond to the `true_outputs`! ' \\\n",
    "                  f'{pred_means.shape[0]} != {n_test_samples}'\n",
    "        raise ValueError(err_msg)\n",
    "    if n_test_samples != pred_vars.shape[0]:\n",
    "        err_msg = f'The `pred_vars` does not correspond to the `true_outputs`! ' \\\n",
    "                  f'{pred_vars.shape[0]} != {n_test_samples}'\n",
    "        raise ValueError(err_msg)\n",
    "    \n",
    "    all_preds_ = np.empty((1, n_test_samples, 2), dtype=np.float32)\n",
    "    all_preds_[0, :, 0] = pred_means\n",
    "    all_preds_[0, :, 1] = pred_vars\n",
    "    all_uncertainty_ = ensemble_uncertainties_regression(all_preds_)\n",
    "    uncertainties = all_uncertainty_['tvar']\n",
    "    del all_preds_, all_uncertainty_\n",
    "    \n",
    "    errors = (pred_means - true_outputs) ** 2\n",
    "    rejection_mse_ = calc_uncertainty_regection_curve(errors, uncertainties)\n",
    "    return np.mean(rejection_mse_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "8401ac8a",
   "metadata": {},
   "outputs": [],
   "source": [
    "def predict_by_ensemble(input_data: np.ndarray,\n",
    "                        preprocessing: Pipeline,\n",
    "                        ensemble: List[tf.keras.Model],\n",
    "                        postprocessing: List[StandardScaler],\n",
    "                        minibatch: int) -> np.ndarray:\n",
    "    num_samples = input_data.shape[0]\n",
    "    ensemble_size = len(postprocessing)\n",
    "    if ensemble_size != len(ensemble):\n",
    "        err_msg = f'Ensemble of preprocessors does not correspond to ' \\\n",
    "                  f'ensemble of models! {ensemble_size} != {len(ensemble)}'\n",
    "        raise ValueError(err_msg)\n",
    "    predictions_of_ensemble = np.empty((ensemble_size, num_samples, 2),\n",
    "                                       dtype=np.float64)\n",
    "    X = preprocessing.transform(input_data).astype(np.float32)\n",
    "    for model_idx, (cur_model, post_) in enumerate(zip(ensemble, postprocessing)):\n",
    "        y_mean, y_var = predict_with_single_nn(\n",
    "            input_data=X,\n",
    "            model_for_prediction=cur_model,\n",
    "            output_scaler=post_,\n",
    "            batch_size=minibatch\n",
    "        )\n",
    "        predictions_of_ensemble[model_idx, :, 0] = y_mean\n",
    "        predictions_of_ensemble[model_idx, :, 1] = y_var\n",
    "    return predictions_of_ensemble"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "87069127",
   "metadata": {},
   "outputs": [],
   "source": [
    "random.seed(42)\n",
    "np.random.seed(42)\n",
    "tf.random.set_seed(42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "640baa60",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "data\\yandex-shifts\\weather True\n"
     ]
    }
   ],
   "source": [
    "data_dir = os.path.join('data', 'yandex-shifts', 'weather')\n",
    "print(f'{data_dir} {os.path.isdir(data_dir)}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "51bc248e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "models\\yandex-shifts\\weather True\n"
     ]
    }
   ],
   "source": [
    "model_dir = os.path.join('models', 'yandex-shifts', 'weather')\n",
    "print(f'{model_dir} {os.path.isdir(model_dir)}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "f18a0a43",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "models\\yandex-shifts\\weather-baseline-catboost True\n"
     ]
    }
   ],
   "source": [
    "baseline_model_dir = os.path.join('models', 'yandex-shifts',\n",
    "                                  'weather-baseline-catboost')\n",
    "print(f'{baseline_model_dir} {os.path.isdir(baseline_model_dir)}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "63b5e35b",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "data\\yandex-shifts\\weather\\dev_in.csv True\n"
     ]
    }
   ],
   "source": [
    "dev_in_name = os.path.join(data_dir, 'dev_in.csv')\n",
    "print(f'{dev_in_name} {os.path.isfile(dev_in_name)}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "3279f561",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "data\\yandex-shifts\\weather\\dev_out.csv True\n"
     ]
    }
   ],
   "source": [
    "dev_out_name = os.path.join(data_dir, 'dev_out.csv')\n",
    "print(f'{dev_out_name} {os.path.isfile(dev_out_name)}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "3eae59d1",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "data\\yandex-shifts\\weather\\eval.csv True\n"
     ]
    }
   ],
   "source": [
    "eval_name = os.path.join(data_dir, 'eval.csv')\n",
    "print(f'{eval_name} {os.path.isfile(eval_name)}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "726686a5",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\bond0\\Anaconda3\\envs\\env_for_tf20\\lib\\site-packages\\sklearn\\base.py:318: UserWarning: Trying to unpickle estimator SimpleImputer from version 0.23.2 when using version 0.22.1. This might lead to breaking code or invalid results. Use at your own risk.\n",
      "  UserWarning)\n",
      "C:\\Users\\bond0\\Anaconda3\\envs\\env_for_tf20\\lib\\site-packages\\sklearn\\base.py:318: UserWarning: Trying to unpickle estimator MinMaxScaler from version 0.23.2 when using version 0.22.1. This might lead to breaking code or invalid results. Use at your own risk.\n",
      "  UserWarning)\n",
      "C:\\Users\\bond0\\Anaconda3\\envs\\env_for_tf20\\lib\\site-packages\\sklearn\\base.py:318: UserWarning: Trying to unpickle estimator KBinsDiscretizer from version 0.23.2 when using version 0.22.1. This might lead to breaking code or invalid results. Use at your own risk.\n",
      "  UserWarning)\n",
      "C:\\Users\\bond0\\Anaconda3\\envs\\env_for_tf20\\lib\\site-packages\\sklearn\\base.py:318: UserWarning: Trying to unpickle estimator Binarizer from version 0.23.2 when using version 0.22.1. This might lead to breaking code or invalid results. Use at your own risk.\n",
      "  UserWarning)\n",
      "C:\\Users\\bond0\\Anaconda3\\envs\\env_for_tf20\\lib\\site-packages\\sklearn\\base.py:318: UserWarning: Trying to unpickle estimator ColumnTransformer from version 0.23.2 when using version 0.22.1. This might lead to breaking code or invalid results. Use at your own risk.\n",
      "  UserWarning)\n",
      "C:\\Users\\bond0\\Anaconda3\\envs\\env_for_tf20\\lib\\site-packages\\sklearn\\base.py:318: UserWarning: Trying to unpickle estimator VarianceThreshold from version 0.23.2 when using version 0.22.1. This might lead to breaking code or invalid results. Use at your own risk.\n",
      "  UserWarning)\n",
      "C:\\Users\\bond0\\Anaconda3\\envs\\env_for_tf20\\lib\\site-packages\\sklearn\\base.py:318: UserWarning: Trying to unpickle estimator StandardScaler from version 0.23.2 when using version 0.22.1. This might lead to breaking code or invalid results. Use at your own risk.\n",
      "  UserWarning)\n",
      "C:\\Users\\bond0\\Anaconda3\\envs\\env_for_tf20\\lib\\site-packages\\sklearn\\base.py:318: UserWarning: Trying to unpickle estimator PCA from version 0.23.2 when using version 0.22.1. This might lead to breaking code or invalid results. Use at your own risk.\n",
      "  UserWarning)\n",
      "C:\\Users\\bond0\\Anaconda3\\envs\\env_for_tf20\\lib\\site-packages\\sklearn\\base.py:318: UserWarning: Trying to unpickle estimator Pipeline from version 0.23.2 when using version 0.22.1. This might lead to breaking code or invalid results. Use at your own risk.\n",
      "  UserWarning)\n"
     ]
    }
   ],
   "source": [
    "preprocessing_name = os.path.join(model_dir, 'preprocessing_pipeline.pkl')\n",
    "with open(preprocessing_name, 'rb') as fp:\n",
    "    common_preprocessor = pickle.load(fp)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "753b988e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Pipeline(memory=None,\n",
      "         steps=[('imputer',\n",
      "                 SimpleImputer(add_indicator=False, copy=True, fill_value=-1.0,\n",
      "                               missing_values=nan, strategy='constant',\n",
      "                               verbose=0)),\n",
      "                ('minmax_scaler',\n",
      "                 MinMaxScaler(copy=True, feature_range=(0, 1))),\n",
      "                ('composite_transformer',\n",
      "                 ColumnTransformer(n_jobs=1, remainder='drop',\n",
      "                                   sparse_threshold=0.0,\n",
      "                                   transformer_weights=None,\n",
      "                                   transformers=[('topogra...\n",
      "                                                 ('cmc_0_2_2_500',\n",
      "                                                  KBinsDiscretizer(encode='ordinal',\n",
      "                                                                   n_bins=256,\n",
      "                                                                   strategy='quantile'),\n",
      "                                                  (29,)), ...],\n",
      "                                   verbose=False)),\n",
      "                ('selector', VarianceThreshold(threshold=0.0)),\n",
      "                ('standard_scaler',\n",
      "                 StandardScaler(copy=True, with_mean=True, with_std=True)),\n",
      "                ('pca',\n",
      "                 PCA(copy=True, iterated_power='auto', n_components=None,\n",
      "                     random_state=42, svd_solver='auto', tol=0.0,\n",
      "                     whiten=False))],\n",
      "         verbose=False)\n"
     ]
    }
   ],
   "source": [
    "print(common_preprocessor)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "9d952714",
   "metadata": {},
   "outputs": [],
   "source": [
    "config_name = os.path.join(model_dir, 'weather_snn_config.json')\n",
    "with codecs.open(config_name, mode='r', encoding='utf-8') as fp:\n",
    "    nn_ens_hyperparams = json.load(fp)\n",
    "num_features = nn_ens_hyperparams['input_size']\n",
    "ensemble_size = nn_ens_hyperparams['ensemble_size']\n",
    "nn_layer_size = nn_ens_hyperparams['layer_size']\n",
    "nn_depth = nn_ens_hyperparams['n_layers']\n",
    "nn_dropout = nn_ens_hyperparams['alpha_dropout_rate']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "01916bc9",
   "metadata": {},
   "outputs": [],
   "source": [
    "postprocessing_name = os.path.join(model_dir, 'postprocessing_scalers.pkl')\n",
    "with open(postprocessing_name, 'rb') as fp:\n",
    "    postprocessing_scalers = pickle.load(fp)\n",
    "if len(postprocessing_scalers) != ensemble_size:\n",
    "    err_msg_ = f'Postprocessors do not correspond to NN ensemble! ' \\\n",
    "               f'{len(postprocessing_scalers)} != {ensemble_size}'\n",
    "    raise ValueError(err_msg_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "c9d044e7",
   "metadata": {},
   "outputs": [],
   "source": [
    "deep_ensemble = []\n",
    "BATCH_SIZE = 4096\n",
    "new_figure_id = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "1af52dbe",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "========================================\n",
      " Self-Normalizing Network 1 for weather prediction\n",
      "========================================\n",
      "\n",
      "Model: \"weather_snn_1\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_1_feature_vector (I [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout1 (AlphaDr (None, 123)          0           weather_snn_1_feature_vector[0][0\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense1 (Dense)    (None, 512)          63488       weather_snn_1_dropout1[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout2 (AlphaDr (None, 512)          0           weather_snn_1_dense1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense2 (Dense)    (None, 512)          262656      weather_snn_1_dropout2[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout3 (AlphaDr (None, 512)          0           weather_snn_1_dense2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense3 (Dense)    (None, 512)          262656      weather_snn_1_dropout3[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout4 (AlphaDr (None, 512)          0           weather_snn_1_dense3[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense4 (Dense)    (None, 512)          262656      weather_snn_1_dropout4[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout5 (AlphaDr (None, 512)          0           weather_snn_1_dense4[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense5 (Dense)    (None, 512)          262656      weather_snn_1_dropout5[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout6 (AlphaDr (None, 512)          0           weather_snn_1_dense5[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense6 (Dense)    (None, 512)          262656      weather_snn_1_dropout6[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout7 (AlphaDr (None, 512)          0           weather_snn_1_dense6[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense7 (Dense)    (None, 512)          262656      weather_snn_1_dropout7[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout8 (AlphaDr (None, 512)          0           weather_snn_1_dense7[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense8 (Dense)    (None, 512)          262656      weather_snn_1_dropout8[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout9 (AlphaDr (None, 512)          0           weather_snn_1_dense8[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense9 (Dense)    (None, 512)          262656      weather_snn_1_dropout9[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout10 (AlphaD (None, 512)          0           weather_snn_1_dense9[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense10 (Dense)   (None, 512)          262656      weather_snn_1_dropout10[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout11 (AlphaD (None, 512)          0           weather_snn_1_dense10[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense11 (Dense)   (None, 512)          262656      weather_snn_1_dropout11[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout12 (AlphaD (None, 512)          0           weather_snn_1_dense11[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense12 (Dense)   (None, 512)          262656      weather_snn_1_dropout12[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout13 (AlphaD (None, 512)          0           weather_snn_1_dense12[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense13 (Dense)   (None, 512)          262656      weather_snn_1_dropout13[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout14 (AlphaD (None, 512)          0           weather_snn_1_dense13[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense14 (Dense)   (None, 512)          262656      weather_snn_1_dropout14[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout15 (AlphaD (None, 512)          0           weather_snn_1_dense14[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense15 (Dense)   (None, 512)          262656      weather_snn_1_dropout15[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout16 (AlphaD (None, 512)          0           weather_snn_1_dense15[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense16 (Dense)   (None, 512)          262656      weather_snn_1_dropout16[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout17 (AlphaD (None, 512)          0           weather_snn_1_dense16[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense17 (Dense)   (None, 512)          262656      weather_snn_1_dropout17[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout18 (AlphaD (None, 512)          0           weather_snn_1_dense17[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dense18 (Dense)   (None, 512)          262656      weather_snn_1_dropout18[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_dropout19 (AlphaD (None, 512)          0           weather_snn_1_dense18[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_output (Dense)    (None, 2)            1024        weather_snn_1_dropout19[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_distribution (Dis ((None, 1), (None, 1 0           weather_snn_1_output[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_1_projection (Dense (None, 50)           25600       weather_snn_1_dropout13[0][0]    \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 2 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_2\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_2_feature_vector (I [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout1 (AlphaDr (None, 123)          0           weather_snn_2_feature_vector[0][0\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense1 (Dense)    (None, 512)          63488       weather_snn_2_dropout1[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout2 (AlphaDr (None, 512)          0           weather_snn_2_dense1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense2 (Dense)    (None, 512)          262656      weather_snn_2_dropout2[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout3 (AlphaDr (None, 512)          0           weather_snn_2_dense2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense3 (Dense)    (None, 512)          262656      weather_snn_2_dropout3[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout4 (AlphaDr (None, 512)          0           weather_snn_2_dense3[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense4 (Dense)    (None, 512)          262656      weather_snn_2_dropout4[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout5 (AlphaDr (None, 512)          0           weather_snn_2_dense4[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense5 (Dense)    (None, 512)          262656      weather_snn_2_dropout5[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout6 (AlphaDr (None, 512)          0           weather_snn_2_dense5[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense6 (Dense)    (None, 512)          262656      weather_snn_2_dropout6[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout7 (AlphaDr (None, 512)          0           weather_snn_2_dense6[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense7 (Dense)    (None, 512)          262656      weather_snn_2_dropout7[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout8 (AlphaDr (None, 512)          0           weather_snn_2_dense7[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense8 (Dense)    (None, 512)          262656      weather_snn_2_dropout8[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout9 (AlphaDr (None, 512)          0           weather_snn_2_dense8[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense9 (Dense)    (None, 512)          262656      weather_snn_2_dropout9[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout10 (AlphaD (None, 512)          0           weather_snn_2_dense9[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense10 (Dense)   (None, 512)          262656      weather_snn_2_dropout10[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout11 (AlphaD (None, 512)          0           weather_snn_2_dense10[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense11 (Dense)   (None, 512)          262656      weather_snn_2_dropout11[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout12 (AlphaD (None, 512)          0           weather_snn_2_dense11[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense12 (Dense)   (None, 512)          262656      weather_snn_2_dropout12[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout13 (AlphaD (None, 512)          0           weather_snn_2_dense12[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense13 (Dense)   (None, 512)          262656      weather_snn_2_dropout13[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout14 (AlphaD (None, 512)          0           weather_snn_2_dense13[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense14 (Dense)   (None, 512)          262656      weather_snn_2_dropout14[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout15 (AlphaD (None, 512)          0           weather_snn_2_dense14[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense15 (Dense)   (None, 512)          262656      weather_snn_2_dropout15[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout16 (AlphaD (None, 512)          0           weather_snn_2_dense15[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense16 (Dense)   (None, 512)          262656      weather_snn_2_dropout16[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout17 (AlphaD (None, 512)          0           weather_snn_2_dense16[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense17 (Dense)   (None, 512)          262656      weather_snn_2_dropout17[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout18 (AlphaD (None, 512)          0           weather_snn_2_dense17[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dense18 (Dense)   (None, 512)          262656      weather_snn_2_dropout18[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_dropout19 (AlphaD (None, 512)          0           weather_snn_2_dense18[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_output (Dense)    (None, 2)            1024        weather_snn_2_dropout19[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_distribution (Dis ((None, 1), (None, 1 0           weather_snn_2_output[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_2_projection (Dense (None, 50)           25600       weather_snn_2_dropout13[0][0]    \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 3 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_3\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_3_feature_vector (I [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout1 (AlphaDr (None, 123)          0           weather_snn_3_feature_vector[0][0\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense1 (Dense)    (None, 512)          63488       weather_snn_3_dropout1[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout2 (AlphaDr (None, 512)          0           weather_snn_3_dense1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense2 (Dense)    (None, 512)          262656      weather_snn_3_dropout2[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout3 (AlphaDr (None, 512)          0           weather_snn_3_dense2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense3 (Dense)    (None, 512)          262656      weather_snn_3_dropout3[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout4 (AlphaDr (None, 512)          0           weather_snn_3_dense3[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense4 (Dense)    (None, 512)          262656      weather_snn_3_dropout4[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout5 (AlphaDr (None, 512)          0           weather_snn_3_dense4[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense5 (Dense)    (None, 512)          262656      weather_snn_3_dropout5[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout6 (AlphaDr (None, 512)          0           weather_snn_3_dense5[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense6 (Dense)    (None, 512)          262656      weather_snn_3_dropout6[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout7 (AlphaDr (None, 512)          0           weather_snn_3_dense6[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense7 (Dense)    (None, 512)          262656      weather_snn_3_dropout7[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout8 (AlphaDr (None, 512)          0           weather_snn_3_dense7[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense8 (Dense)    (None, 512)          262656      weather_snn_3_dropout8[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout9 (AlphaDr (None, 512)          0           weather_snn_3_dense8[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense9 (Dense)    (None, 512)          262656      weather_snn_3_dropout9[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout10 (AlphaD (None, 512)          0           weather_snn_3_dense9[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense10 (Dense)   (None, 512)          262656      weather_snn_3_dropout10[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout11 (AlphaD (None, 512)          0           weather_snn_3_dense10[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense11 (Dense)   (None, 512)          262656      weather_snn_3_dropout11[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout12 (AlphaD (None, 512)          0           weather_snn_3_dense11[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense12 (Dense)   (None, 512)          262656      weather_snn_3_dropout12[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout13 (AlphaD (None, 512)          0           weather_snn_3_dense12[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense13 (Dense)   (None, 512)          262656      weather_snn_3_dropout13[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout14 (AlphaD (None, 512)          0           weather_snn_3_dense13[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense14 (Dense)   (None, 512)          262656      weather_snn_3_dropout14[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout15 (AlphaD (None, 512)          0           weather_snn_3_dense14[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense15 (Dense)   (None, 512)          262656      weather_snn_3_dropout15[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout16 (AlphaD (None, 512)          0           weather_snn_3_dense15[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense16 (Dense)   (None, 512)          262656      weather_snn_3_dropout16[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout17 (AlphaD (None, 512)          0           weather_snn_3_dense16[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense17 (Dense)   (None, 512)          262656      weather_snn_3_dropout17[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout18 (AlphaD (None, 512)          0           weather_snn_3_dense17[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dense18 (Dense)   (None, 512)          262656      weather_snn_3_dropout18[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_dropout19 (AlphaD (None, 512)          0           weather_snn_3_dense18[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_output (Dense)    (None, 2)            1024        weather_snn_3_dropout19[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_distribution (Dis ((None, 1), (None, 1 0           weather_snn_3_output[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_3_projection (Dense (None, 50)           25600       weather_snn_3_dropout13[0][0]    \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 4 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_4\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_4_feature_vector (I [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout1 (AlphaDr (None, 123)          0           weather_snn_4_feature_vector[0][0\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense1 (Dense)    (None, 512)          63488       weather_snn_4_dropout1[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout2 (AlphaDr (None, 512)          0           weather_snn_4_dense1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense2 (Dense)    (None, 512)          262656      weather_snn_4_dropout2[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout3 (AlphaDr (None, 512)          0           weather_snn_4_dense2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense3 (Dense)    (None, 512)          262656      weather_snn_4_dropout3[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout4 (AlphaDr (None, 512)          0           weather_snn_4_dense3[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense4 (Dense)    (None, 512)          262656      weather_snn_4_dropout4[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout5 (AlphaDr (None, 512)          0           weather_snn_4_dense4[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense5 (Dense)    (None, 512)          262656      weather_snn_4_dropout5[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout6 (AlphaDr (None, 512)          0           weather_snn_4_dense5[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense6 (Dense)    (None, 512)          262656      weather_snn_4_dropout6[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout7 (AlphaDr (None, 512)          0           weather_snn_4_dense6[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense7 (Dense)    (None, 512)          262656      weather_snn_4_dropout7[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout8 (AlphaDr (None, 512)          0           weather_snn_4_dense7[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense8 (Dense)    (None, 512)          262656      weather_snn_4_dropout8[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout9 (AlphaDr (None, 512)          0           weather_snn_4_dense8[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense9 (Dense)    (None, 512)          262656      weather_snn_4_dropout9[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout10 (AlphaD (None, 512)          0           weather_snn_4_dense9[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense10 (Dense)   (None, 512)          262656      weather_snn_4_dropout10[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout11 (AlphaD (None, 512)          0           weather_snn_4_dense10[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense11 (Dense)   (None, 512)          262656      weather_snn_4_dropout11[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout12 (AlphaD (None, 512)          0           weather_snn_4_dense11[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense12 (Dense)   (None, 512)          262656      weather_snn_4_dropout12[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout13 (AlphaD (None, 512)          0           weather_snn_4_dense12[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense13 (Dense)   (None, 512)          262656      weather_snn_4_dropout13[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout14 (AlphaD (None, 512)          0           weather_snn_4_dense13[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense14 (Dense)   (None, 512)          262656      weather_snn_4_dropout14[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout15 (AlphaD (None, 512)          0           weather_snn_4_dense14[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense15 (Dense)   (None, 512)          262656      weather_snn_4_dropout15[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout16 (AlphaD (None, 512)          0           weather_snn_4_dense15[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense16 (Dense)   (None, 512)          262656      weather_snn_4_dropout16[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout17 (AlphaD (None, 512)          0           weather_snn_4_dense16[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense17 (Dense)   (None, 512)          262656      weather_snn_4_dropout17[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout18 (AlphaD (None, 512)          0           weather_snn_4_dense17[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dense18 (Dense)   (None, 512)          262656      weather_snn_4_dropout18[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_dropout19 (AlphaD (None, 512)          0           weather_snn_4_dense18[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_output (Dense)    (None, 2)            1024        weather_snn_4_dropout19[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_distribution (Dis ((None, 1), (None, 1 0           weather_snn_4_output[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_4_projection (Dense (None, 50)           25600       weather_snn_4_dropout13[0][0]    \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 5 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_5\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_5_feature_vector (I [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout1 (AlphaDr (None, 123)          0           weather_snn_5_feature_vector[0][0\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense1 (Dense)    (None, 512)          63488       weather_snn_5_dropout1[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout2 (AlphaDr (None, 512)          0           weather_snn_5_dense1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense2 (Dense)    (None, 512)          262656      weather_snn_5_dropout2[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout3 (AlphaDr (None, 512)          0           weather_snn_5_dense2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense3 (Dense)    (None, 512)          262656      weather_snn_5_dropout3[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout4 (AlphaDr (None, 512)          0           weather_snn_5_dense3[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense4 (Dense)    (None, 512)          262656      weather_snn_5_dropout4[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout5 (AlphaDr (None, 512)          0           weather_snn_5_dense4[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense5 (Dense)    (None, 512)          262656      weather_snn_5_dropout5[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout6 (AlphaDr (None, 512)          0           weather_snn_5_dense5[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense6 (Dense)    (None, 512)          262656      weather_snn_5_dropout6[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout7 (AlphaDr (None, 512)          0           weather_snn_5_dense6[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense7 (Dense)    (None, 512)          262656      weather_snn_5_dropout7[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout8 (AlphaDr (None, 512)          0           weather_snn_5_dense7[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense8 (Dense)    (None, 512)          262656      weather_snn_5_dropout8[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout9 (AlphaDr (None, 512)          0           weather_snn_5_dense8[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense9 (Dense)    (None, 512)          262656      weather_snn_5_dropout9[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout10 (AlphaD (None, 512)          0           weather_snn_5_dense9[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense10 (Dense)   (None, 512)          262656      weather_snn_5_dropout10[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout11 (AlphaD (None, 512)          0           weather_snn_5_dense10[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense11 (Dense)   (None, 512)          262656      weather_snn_5_dropout11[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout12 (AlphaD (None, 512)          0           weather_snn_5_dense11[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense12 (Dense)   (None, 512)          262656      weather_snn_5_dropout12[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout13 (AlphaD (None, 512)          0           weather_snn_5_dense12[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense13 (Dense)   (None, 512)          262656      weather_snn_5_dropout13[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout14 (AlphaD (None, 512)          0           weather_snn_5_dense13[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense14 (Dense)   (None, 512)          262656      weather_snn_5_dropout14[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout15 (AlphaD (None, 512)          0           weather_snn_5_dense14[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense15 (Dense)   (None, 512)          262656      weather_snn_5_dropout15[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout16 (AlphaD (None, 512)          0           weather_snn_5_dense15[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense16 (Dense)   (None, 512)          262656      weather_snn_5_dropout16[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout17 (AlphaD (None, 512)          0           weather_snn_5_dense16[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense17 (Dense)   (None, 512)          262656      weather_snn_5_dropout17[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout18 (AlphaD (None, 512)          0           weather_snn_5_dense17[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dense18 (Dense)   (None, 512)          262656      weather_snn_5_dropout18[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_dropout19 (AlphaD (None, 512)          0           weather_snn_5_dense18[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_output (Dense)    (None, 2)            1024        weather_snn_5_dropout19[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_distribution (Dis ((None, 1), (None, 1 0           weather_snn_5_output[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_5_projection (Dense (None, 50)           25600       weather_snn_5_dropout13[0][0]    \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 6 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_6\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_6_feature_vector (I [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout1 (AlphaDr (None, 123)          0           weather_snn_6_feature_vector[0][0\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense1 (Dense)    (None, 512)          63488       weather_snn_6_dropout1[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout2 (AlphaDr (None, 512)          0           weather_snn_6_dense1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense2 (Dense)    (None, 512)          262656      weather_snn_6_dropout2[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout3 (AlphaDr (None, 512)          0           weather_snn_6_dense2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense3 (Dense)    (None, 512)          262656      weather_snn_6_dropout3[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout4 (AlphaDr (None, 512)          0           weather_snn_6_dense3[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense4 (Dense)    (None, 512)          262656      weather_snn_6_dropout4[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout5 (AlphaDr (None, 512)          0           weather_snn_6_dense4[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense5 (Dense)    (None, 512)          262656      weather_snn_6_dropout5[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout6 (AlphaDr (None, 512)          0           weather_snn_6_dense5[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense6 (Dense)    (None, 512)          262656      weather_snn_6_dropout6[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout7 (AlphaDr (None, 512)          0           weather_snn_6_dense6[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense7 (Dense)    (None, 512)          262656      weather_snn_6_dropout7[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout8 (AlphaDr (None, 512)          0           weather_snn_6_dense7[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense8 (Dense)    (None, 512)          262656      weather_snn_6_dropout8[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout9 (AlphaDr (None, 512)          0           weather_snn_6_dense8[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense9 (Dense)    (None, 512)          262656      weather_snn_6_dropout9[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout10 (AlphaD (None, 512)          0           weather_snn_6_dense9[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense10 (Dense)   (None, 512)          262656      weather_snn_6_dropout10[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout11 (AlphaD (None, 512)          0           weather_snn_6_dense10[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense11 (Dense)   (None, 512)          262656      weather_snn_6_dropout11[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout12 (AlphaD (None, 512)          0           weather_snn_6_dense11[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense12 (Dense)   (None, 512)          262656      weather_snn_6_dropout12[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout13 (AlphaD (None, 512)          0           weather_snn_6_dense12[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense13 (Dense)   (None, 512)          262656      weather_snn_6_dropout13[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout14 (AlphaD (None, 512)          0           weather_snn_6_dense13[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense14 (Dense)   (None, 512)          262656      weather_snn_6_dropout14[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout15 (AlphaD (None, 512)          0           weather_snn_6_dense14[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense15 (Dense)   (None, 512)          262656      weather_snn_6_dropout15[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout16 (AlphaD (None, 512)          0           weather_snn_6_dense15[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense16 (Dense)   (None, 512)          262656      weather_snn_6_dropout16[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout17 (AlphaD (None, 512)          0           weather_snn_6_dense16[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense17 (Dense)   (None, 512)          262656      weather_snn_6_dropout17[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout18 (AlphaD (None, 512)          0           weather_snn_6_dense17[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dense18 (Dense)   (None, 512)          262656      weather_snn_6_dropout18[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_dropout19 (AlphaD (None, 512)          0           weather_snn_6_dense18[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_output (Dense)    (None, 2)            1024        weather_snn_6_dropout19[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_distribution (Dis ((None, 1), (None, 1 0           weather_snn_6_output[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_6_projection (Dense (None, 50)           25600       weather_snn_6_dropout13[0][0]    \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 7 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_7\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_7_feature_vector (I [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout1 (AlphaDr (None, 123)          0           weather_snn_7_feature_vector[0][0\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense1 (Dense)    (None, 512)          63488       weather_snn_7_dropout1[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout2 (AlphaDr (None, 512)          0           weather_snn_7_dense1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense2 (Dense)    (None, 512)          262656      weather_snn_7_dropout2[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout3 (AlphaDr (None, 512)          0           weather_snn_7_dense2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense3 (Dense)    (None, 512)          262656      weather_snn_7_dropout3[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout4 (AlphaDr (None, 512)          0           weather_snn_7_dense3[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense4 (Dense)    (None, 512)          262656      weather_snn_7_dropout4[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout5 (AlphaDr (None, 512)          0           weather_snn_7_dense4[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense5 (Dense)    (None, 512)          262656      weather_snn_7_dropout5[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout6 (AlphaDr (None, 512)          0           weather_snn_7_dense5[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense6 (Dense)    (None, 512)          262656      weather_snn_7_dropout6[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout7 (AlphaDr (None, 512)          0           weather_snn_7_dense6[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense7 (Dense)    (None, 512)          262656      weather_snn_7_dropout7[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout8 (AlphaDr (None, 512)          0           weather_snn_7_dense7[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense8 (Dense)    (None, 512)          262656      weather_snn_7_dropout8[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout9 (AlphaDr (None, 512)          0           weather_snn_7_dense8[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense9 (Dense)    (None, 512)          262656      weather_snn_7_dropout9[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout10 (AlphaD (None, 512)          0           weather_snn_7_dense9[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense10 (Dense)   (None, 512)          262656      weather_snn_7_dropout10[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout11 (AlphaD (None, 512)          0           weather_snn_7_dense10[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense11 (Dense)   (None, 512)          262656      weather_snn_7_dropout11[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout12 (AlphaD (None, 512)          0           weather_snn_7_dense11[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense12 (Dense)   (None, 512)          262656      weather_snn_7_dropout12[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout13 (AlphaD (None, 512)          0           weather_snn_7_dense12[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense13 (Dense)   (None, 512)          262656      weather_snn_7_dropout13[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout14 (AlphaD (None, 512)          0           weather_snn_7_dense13[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense14 (Dense)   (None, 512)          262656      weather_snn_7_dropout14[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout15 (AlphaD (None, 512)          0           weather_snn_7_dense14[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense15 (Dense)   (None, 512)          262656      weather_snn_7_dropout15[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout16 (AlphaD (None, 512)          0           weather_snn_7_dense15[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense16 (Dense)   (None, 512)          262656      weather_snn_7_dropout16[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout17 (AlphaD (None, 512)          0           weather_snn_7_dense16[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense17 (Dense)   (None, 512)          262656      weather_snn_7_dropout17[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout18 (AlphaD (None, 512)          0           weather_snn_7_dense17[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dense18 (Dense)   (None, 512)          262656      weather_snn_7_dropout18[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_dropout19 (AlphaD (None, 512)          0           weather_snn_7_dense18[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_output (Dense)    (None, 2)            1024        weather_snn_7_dropout19[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_distribution (Dis ((None, 1), (None, 1 0           weather_snn_7_output[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_7_projection (Dense (None, 50)           25600       weather_snn_7_dropout13[0][0]    \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 8 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_8\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_8_feature_vector (I [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout1 (AlphaDr (None, 123)          0           weather_snn_8_feature_vector[0][0\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense1 (Dense)    (None, 512)          63488       weather_snn_8_dropout1[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout2 (AlphaDr (None, 512)          0           weather_snn_8_dense1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense2 (Dense)    (None, 512)          262656      weather_snn_8_dropout2[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout3 (AlphaDr (None, 512)          0           weather_snn_8_dense2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense3 (Dense)    (None, 512)          262656      weather_snn_8_dropout3[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout4 (AlphaDr (None, 512)          0           weather_snn_8_dense3[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense4 (Dense)    (None, 512)          262656      weather_snn_8_dropout4[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout5 (AlphaDr (None, 512)          0           weather_snn_8_dense4[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense5 (Dense)    (None, 512)          262656      weather_snn_8_dropout5[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout6 (AlphaDr (None, 512)          0           weather_snn_8_dense5[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense6 (Dense)    (None, 512)          262656      weather_snn_8_dropout6[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout7 (AlphaDr (None, 512)          0           weather_snn_8_dense6[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense7 (Dense)    (None, 512)          262656      weather_snn_8_dropout7[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout8 (AlphaDr (None, 512)          0           weather_snn_8_dense7[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense8 (Dense)    (None, 512)          262656      weather_snn_8_dropout8[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout9 (AlphaDr (None, 512)          0           weather_snn_8_dense8[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense9 (Dense)    (None, 512)          262656      weather_snn_8_dropout9[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout10 (AlphaD (None, 512)          0           weather_snn_8_dense9[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense10 (Dense)   (None, 512)          262656      weather_snn_8_dropout10[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout11 (AlphaD (None, 512)          0           weather_snn_8_dense10[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense11 (Dense)   (None, 512)          262656      weather_snn_8_dropout11[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout12 (AlphaD (None, 512)          0           weather_snn_8_dense11[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense12 (Dense)   (None, 512)          262656      weather_snn_8_dropout12[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout13 (AlphaD (None, 512)          0           weather_snn_8_dense12[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense13 (Dense)   (None, 512)          262656      weather_snn_8_dropout13[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout14 (AlphaD (None, 512)          0           weather_snn_8_dense13[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense14 (Dense)   (None, 512)          262656      weather_snn_8_dropout14[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout15 (AlphaD (None, 512)          0           weather_snn_8_dense14[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense15 (Dense)   (None, 512)          262656      weather_snn_8_dropout15[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout16 (AlphaD (None, 512)          0           weather_snn_8_dense15[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense16 (Dense)   (None, 512)          262656      weather_snn_8_dropout16[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout17 (AlphaD (None, 512)          0           weather_snn_8_dense16[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense17 (Dense)   (None, 512)          262656      weather_snn_8_dropout17[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout18 (AlphaD (None, 512)          0           weather_snn_8_dense17[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dense18 (Dense)   (None, 512)          262656      weather_snn_8_dropout18[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_dropout19 (AlphaD (None, 512)          0           weather_snn_8_dense18[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_output (Dense)    (None, 2)            1024        weather_snn_8_dropout19[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_distribution (Dis ((None, 1), (None, 1 0           weather_snn_8_output[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_8_projection (Dense (None, 50)           25600       weather_snn_8_dropout13[0][0]    \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 9 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_9\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_9_feature_vector (I [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout1 (AlphaDr (None, 123)          0           weather_snn_9_feature_vector[0][0\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense1 (Dense)    (None, 512)          63488       weather_snn_9_dropout1[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout2 (AlphaDr (None, 512)          0           weather_snn_9_dense1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense2 (Dense)    (None, 512)          262656      weather_snn_9_dropout2[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout3 (AlphaDr (None, 512)          0           weather_snn_9_dense2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense3 (Dense)    (None, 512)          262656      weather_snn_9_dropout3[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout4 (AlphaDr (None, 512)          0           weather_snn_9_dense3[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense4 (Dense)    (None, 512)          262656      weather_snn_9_dropout4[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout5 (AlphaDr (None, 512)          0           weather_snn_9_dense4[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense5 (Dense)    (None, 512)          262656      weather_snn_9_dropout5[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout6 (AlphaDr (None, 512)          0           weather_snn_9_dense5[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense6 (Dense)    (None, 512)          262656      weather_snn_9_dropout6[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout7 (AlphaDr (None, 512)          0           weather_snn_9_dense6[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense7 (Dense)    (None, 512)          262656      weather_snn_9_dropout7[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout8 (AlphaDr (None, 512)          0           weather_snn_9_dense7[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense8 (Dense)    (None, 512)          262656      weather_snn_9_dropout8[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout9 (AlphaDr (None, 512)          0           weather_snn_9_dense8[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense9 (Dense)    (None, 512)          262656      weather_snn_9_dropout9[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout10 (AlphaD (None, 512)          0           weather_snn_9_dense9[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense10 (Dense)   (None, 512)          262656      weather_snn_9_dropout10[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout11 (AlphaD (None, 512)          0           weather_snn_9_dense10[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense11 (Dense)   (None, 512)          262656      weather_snn_9_dropout11[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout12 (AlphaD (None, 512)          0           weather_snn_9_dense11[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense12 (Dense)   (None, 512)          262656      weather_snn_9_dropout12[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout13 (AlphaD (None, 512)          0           weather_snn_9_dense12[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense13 (Dense)   (None, 512)          262656      weather_snn_9_dropout13[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout14 (AlphaD (None, 512)          0           weather_snn_9_dense13[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense14 (Dense)   (None, 512)          262656      weather_snn_9_dropout14[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout15 (AlphaD (None, 512)          0           weather_snn_9_dense14[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense15 (Dense)   (None, 512)          262656      weather_snn_9_dropout15[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout16 (AlphaD (None, 512)          0           weather_snn_9_dense15[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense16 (Dense)   (None, 512)          262656      weather_snn_9_dropout16[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout17 (AlphaD (None, 512)          0           weather_snn_9_dense16[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense17 (Dense)   (None, 512)          262656      weather_snn_9_dropout17[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout18 (AlphaD (None, 512)          0           weather_snn_9_dense17[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dense18 (Dense)   (None, 512)          262656      weather_snn_9_dropout18[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_dropout19 (AlphaD (None, 512)          0           weather_snn_9_dense18[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_output (Dense)    (None, 2)            1024        weather_snn_9_dropout19[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_distribution (Dis ((None, 1), (None, 1 0           weather_snn_9_output[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_9_projection (Dense (None, 50)           25600       weather_snn_9_dropout13[0][0]    \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 10 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_10\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_10_feature_vector ( [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout1 (AlphaD (None, 123)          0           weather_snn_10_feature_vector[0][\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense1 (Dense)   (None, 512)          63488       weather_snn_10_dropout1[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout2 (AlphaD (None, 512)          0           weather_snn_10_dense1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense2 (Dense)   (None, 512)          262656      weather_snn_10_dropout2[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout3 (AlphaD (None, 512)          0           weather_snn_10_dense2[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense3 (Dense)   (None, 512)          262656      weather_snn_10_dropout3[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout4 (AlphaD (None, 512)          0           weather_snn_10_dense3[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense4 (Dense)   (None, 512)          262656      weather_snn_10_dropout4[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout5 (AlphaD (None, 512)          0           weather_snn_10_dense4[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense5 (Dense)   (None, 512)          262656      weather_snn_10_dropout5[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout6 (AlphaD (None, 512)          0           weather_snn_10_dense5[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense6 (Dense)   (None, 512)          262656      weather_snn_10_dropout6[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout7 (AlphaD (None, 512)          0           weather_snn_10_dense6[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense7 (Dense)   (None, 512)          262656      weather_snn_10_dropout7[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout8 (AlphaD (None, 512)          0           weather_snn_10_dense7[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense8 (Dense)   (None, 512)          262656      weather_snn_10_dropout8[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout9 (AlphaD (None, 512)          0           weather_snn_10_dense8[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense9 (Dense)   (None, 512)          262656      weather_snn_10_dropout9[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout10 (Alpha (None, 512)          0           weather_snn_10_dense9[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense10 (Dense)  (None, 512)          262656      weather_snn_10_dropout10[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout11 (Alpha (None, 512)          0           weather_snn_10_dense10[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense11 (Dense)  (None, 512)          262656      weather_snn_10_dropout11[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout12 (Alpha (None, 512)          0           weather_snn_10_dense11[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense12 (Dense)  (None, 512)          262656      weather_snn_10_dropout12[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout13 (Alpha (None, 512)          0           weather_snn_10_dense12[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense13 (Dense)  (None, 512)          262656      weather_snn_10_dropout13[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout14 (Alpha (None, 512)          0           weather_snn_10_dense13[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense14 (Dense)  (None, 512)          262656      weather_snn_10_dropout14[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout15 (Alpha (None, 512)          0           weather_snn_10_dense14[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense15 (Dense)  (None, 512)          262656      weather_snn_10_dropout15[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout16 (Alpha (None, 512)          0           weather_snn_10_dense15[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense16 (Dense)  (None, 512)          262656      weather_snn_10_dropout16[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout17 (Alpha (None, 512)          0           weather_snn_10_dense16[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense17 (Dense)  (None, 512)          262656      weather_snn_10_dropout17[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout18 (Alpha (None, 512)          0           weather_snn_10_dense17[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dense18 (Dense)  (None, 512)          262656      weather_snn_10_dropout18[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_dropout19 (Alpha (None, 512)          0           weather_snn_10_dense18[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_output (Dense)   (None, 2)            1024        weather_snn_10_dropout19[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_distribution (Di ((None, 1), (None, 1 0           weather_snn_10_output[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_10_projection (Dens (None, 50)           25600       weather_snn_10_dropout13[0][0]   \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 11 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_11\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_11_feature_vector ( [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout1 (AlphaD (None, 123)          0           weather_snn_11_feature_vector[0][\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense1 (Dense)   (None, 512)          63488       weather_snn_11_dropout1[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout2 (AlphaD (None, 512)          0           weather_snn_11_dense1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense2 (Dense)   (None, 512)          262656      weather_snn_11_dropout2[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout3 (AlphaD (None, 512)          0           weather_snn_11_dense2[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense3 (Dense)   (None, 512)          262656      weather_snn_11_dropout3[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout4 (AlphaD (None, 512)          0           weather_snn_11_dense3[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense4 (Dense)   (None, 512)          262656      weather_snn_11_dropout4[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout5 (AlphaD (None, 512)          0           weather_snn_11_dense4[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense5 (Dense)   (None, 512)          262656      weather_snn_11_dropout5[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout6 (AlphaD (None, 512)          0           weather_snn_11_dense5[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense6 (Dense)   (None, 512)          262656      weather_snn_11_dropout6[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout7 (AlphaD (None, 512)          0           weather_snn_11_dense6[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense7 (Dense)   (None, 512)          262656      weather_snn_11_dropout7[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout8 (AlphaD (None, 512)          0           weather_snn_11_dense7[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense8 (Dense)   (None, 512)          262656      weather_snn_11_dropout8[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout9 (AlphaD (None, 512)          0           weather_snn_11_dense8[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense9 (Dense)   (None, 512)          262656      weather_snn_11_dropout9[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout10 (Alpha (None, 512)          0           weather_snn_11_dense9[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense10 (Dense)  (None, 512)          262656      weather_snn_11_dropout10[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout11 (Alpha (None, 512)          0           weather_snn_11_dense10[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense11 (Dense)  (None, 512)          262656      weather_snn_11_dropout11[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout12 (Alpha (None, 512)          0           weather_snn_11_dense11[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense12 (Dense)  (None, 512)          262656      weather_snn_11_dropout12[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout13 (Alpha (None, 512)          0           weather_snn_11_dense12[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense13 (Dense)  (None, 512)          262656      weather_snn_11_dropout13[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout14 (Alpha (None, 512)          0           weather_snn_11_dense13[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense14 (Dense)  (None, 512)          262656      weather_snn_11_dropout14[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout15 (Alpha (None, 512)          0           weather_snn_11_dense14[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense15 (Dense)  (None, 512)          262656      weather_snn_11_dropout15[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout16 (Alpha (None, 512)          0           weather_snn_11_dense15[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense16 (Dense)  (None, 512)          262656      weather_snn_11_dropout16[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout17 (Alpha (None, 512)          0           weather_snn_11_dense16[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense17 (Dense)  (None, 512)          262656      weather_snn_11_dropout17[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout18 (Alpha (None, 512)          0           weather_snn_11_dense17[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dense18 (Dense)  (None, 512)          262656      weather_snn_11_dropout18[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_dropout19 (Alpha (None, 512)          0           weather_snn_11_dense18[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_output (Dense)   (None, 2)            1024        weather_snn_11_dropout19[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_distribution (Di ((None, 1), (None, 1 0           weather_snn_11_output[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_11_projection (Dens (None, 50)           25600       weather_snn_11_dropout13[0][0]   \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 12 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_12\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_12_feature_vector ( [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout1 (AlphaD (None, 123)          0           weather_snn_12_feature_vector[0][\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense1 (Dense)   (None, 512)          63488       weather_snn_12_dropout1[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout2 (AlphaD (None, 512)          0           weather_snn_12_dense1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense2 (Dense)   (None, 512)          262656      weather_snn_12_dropout2[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout3 (AlphaD (None, 512)          0           weather_snn_12_dense2[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense3 (Dense)   (None, 512)          262656      weather_snn_12_dropout3[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout4 (AlphaD (None, 512)          0           weather_snn_12_dense3[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense4 (Dense)   (None, 512)          262656      weather_snn_12_dropout4[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout5 (AlphaD (None, 512)          0           weather_snn_12_dense4[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense5 (Dense)   (None, 512)          262656      weather_snn_12_dropout5[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout6 (AlphaD (None, 512)          0           weather_snn_12_dense5[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense6 (Dense)   (None, 512)          262656      weather_snn_12_dropout6[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout7 (AlphaD (None, 512)          0           weather_snn_12_dense6[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense7 (Dense)   (None, 512)          262656      weather_snn_12_dropout7[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout8 (AlphaD (None, 512)          0           weather_snn_12_dense7[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense8 (Dense)   (None, 512)          262656      weather_snn_12_dropout8[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout9 (AlphaD (None, 512)          0           weather_snn_12_dense8[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense9 (Dense)   (None, 512)          262656      weather_snn_12_dropout9[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout10 (Alpha (None, 512)          0           weather_snn_12_dense9[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense10 (Dense)  (None, 512)          262656      weather_snn_12_dropout10[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout11 (Alpha (None, 512)          0           weather_snn_12_dense10[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense11 (Dense)  (None, 512)          262656      weather_snn_12_dropout11[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout12 (Alpha (None, 512)          0           weather_snn_12_dense11[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense12 (Dense)  (None, 512)          262656      weather_snn_12_dropout12[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout13 (Alpha (None, 512)          0           weather_snn_12_dense12[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense13 (Dense)  (None, 512)          262656      weather_snn_12_dropout13[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout14 (Alpha (None, 512)          0           weather_snn_12_dense13[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense14 (Dense)  (None, 512)          262656      weather_snn_12_dropout14[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout15 (Alpha (None, 512)          0           weather_snn_12_dense14[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense15 (Dense)  (None, 512)          262656      weather_snn_12_dropout15[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout16 (Alpha (None, 512)          0           weather_snn_12_dense15[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense16 (Dense)  (None, 512)          262656      weather_snn_12_dropout16[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout17 (Alpha (None, 512)          0           weather_snn_12_dense16[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense17 (Dense)  (None, 512)          262656      weather_snn_12_dropout17[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout18 (Alpha (None, 512)          0           weather_snn_12_dense17[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dense18 (Dense)  (None, 512)          262656      weather_snn_12_dropout18[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_dropout19 (Alpha (None, 512)          0           weather_snn_12_dense18[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_output (Dense)   (None, 2)            1024        weather_snn_12_dropout19[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_distribution (Di ((None, 1), (None, 1 0           weather_snn_12_output[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_12_projection (Dens (None, 50)           25600       weather_snn_12_dropout13[0][0]   \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 13 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_13\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_13_feature_vector ( [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout1 (AlphaD (None, 123)          0           weather_snn_13_feature_vector[0][\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense1 (Dense)   (None, 512)          63488       weather_snn_13_dropout1[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout2 (AlphaD (None, 512)          0           weather_snn_13_dense1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense2 (Dense)   (None, 512)          262656      weather_snn_13_dropout2[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout3 (AlphaD (None, 512)          0           weather_snn_13_dense2[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense3 (Dense)   (None, 512)          262656      weather_snn_13_dropout3[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout4 (AlphaD (None, 512)          0           weather_snn_13_dense3[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense4 (Dense)   (None, 512)          262656      weather_snn_13_dropout4[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout5 (AlphaD (None, 512)          0           weather_snn_13_dense4[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense5 (Dense)   (None, 512)          262656      weather_snn_13_dropout5[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout6 (AlphaD (None, 512)          0           weather_snn_13_dense5[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense6 (Dense)   (None, 512)          262656      weather_snn_13_dropout6[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout7 (AlphaD (None, 512)          0           weather_snn_13_dense6[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense7 (Dense)   (None, 512)          262656      weather_snn_13_dropout7[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout8 (AlphaD (None, 512)          0           weather_snn_13_dense7[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense8 (Dense)   (None, 512)          262656      weather_snn_13_dropout8[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout9 (AlphaD (None, 512)          0           weather_snn_13_dense8[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense9 (Dense)   (None, 512)          262656      weather_snn_13_dropout9[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout10 (Alpha (None, 512)          0           weather_snn_13_dense9[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense10 (Dense)  (None, 512)          262656      weather_snn_13_dropout10[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout11 (Alpha (None, 512)          0           weather_snn_13_dense10[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense11 (Dense)  (None, 512)          262656      weather_snn_13_dropout11[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout12 (Alpha (None, 512)          0           weather_snn_13_dense11[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense12 (Dense)  (None, 512)          262656      weather_snn_13_dropout12[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout13 (Alpha (None, 512)          0           weather_snn_13_dense12[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense13 (Dense)  (None, 512)          262656      weather_snn_13_dropout13[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout14 (Alpha (None, 512)          0           weather_snn_13_dense13[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense14 (Dense)  (None, 512)          262656      weather_snn_13_dropout14[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout15 (Alpha (None, 512)          0           weather_snn_13_dense14[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense15 (Dense)  (None, 512)          262656      weather_snn_13_dropout15[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout16 (Alpha (None, 512)          0           weather_snn_13_dense15[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense16 (Dense)  (None, 512)          262656      weather_snn_13_dropout16[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout17 (Alpha (None, 512)          0           weather_snn_13_dense16[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense17 (Dense)  (None, 512)          262656      weather_snn_13_dropout17[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout18 (Alpha (None, 512)          0           weather_snn_13_dense17[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dense18 (Dense)  (None, 512)          262656      weather_snn_13_dropout18[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_dropout19 (Alpha (None, 512)          0           weather_snn_13_dense18[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_output (Dense)   (None, 2)            1024        weather_snn_13_dropout19[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_distribution (Di ((None, 1), (None, 1 0           weather_snn_13_output[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_13_projection (Dens (None, 50)           25600       weather_snn_13_dropout13[0][0]   \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 14 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_14\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_14_feature_vector ( [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout1 (AlphaD (None, 123)          0           weather_snn_14_feature_vector[0][\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense1 (Dense)   (None, 512)          63488       weather_snn_14_dropout1[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout2 (AlphaD (None, 512)          0           weather_snn_14_dense1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense2 (Dense)   (None, 512)          262656      weather_snn_14_dropout2[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout3 (AlphaD (None, 512)          0           weather_snn_14_dense2[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense3 (Dense)   (None, 512)          262656      weather_snn_14_dropout3[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout4 (AlphaD (None, 512)          0           weather_snn_14_dense3[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense4 (Dense)   (None, 512)          262656      weather_snn_14_dropout4[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout5 (AlphaD (None, 512)          0           weather_snn_14_dense4[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense5 (Dense)   (None, 512)          262656      weather_snn_14_dropout5[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout6 (AlphaD (None, 512)          0           weather_snn_14_dense5[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense6 (Dense)   (None, 512)          262656      weather_snn_14_dropout6[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout7 (AlphaD (None, 512)          0           weather_snn_14_dense6[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense7 (Dense)   (None, 512)          262656      weather_snn_14_dropout7[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout8 (AlphaD (None, 512)          0           weather_snn_14_dense7[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense8 (Dense)   (None, 512)          262656      weather_snn_14_dropout8[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout9 (AlphaD (None, 512)          0           weather_snn_14_dense8[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense9 (Dense)   (None, 512)          262656      weather_snn_14_dropout9[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout10 (Alpha (None, 512)          0           weather_snn_14_dense9[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense10 (Dense)  (None, 512)          262656      weather_snn_14_dropout10[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout11 (Alpha (None, 512)          0           weather_snn_14_dense10[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense11 (Dense)  (None, 512)          262656      weather_snn_14_dropout11[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout12 (Alpha (None, 512)          0           weather_snn_14_dense11[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense12 (Dense)  (None, 512)          262656      weather_snn_14_dropout12[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout13 (Alpha (None, 512)          0           weather_snn_14_dense12[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense13 (Dense)  (None, 512)          262656      weather_snn_14_dropout13[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout14 (Alpha (None, 512)          0           weather_snn_14_dense13[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense14 (Dense)  (None, 512)          262656      weather_snn_14_dropout14[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout15 (Alpha (None, 512)          0           weather_snn_14_dense14[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense15 (Dense)  (None, 512)          262656      weather_snn_14_dropout15[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout16 (Alpha (None, 512)          0           weather_snn_14_dense15[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense16 (Dense)  (None, 512)          262656      weather_snn_14_dropout16[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout17 (Alpha (None, 512)          0           weather_snn_14_dense16[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense17 (Dense)  (None, 512)          262656      weather_snn_14_dropout17[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout18 (Alpha (None, 512)          0           weather_snn_14_dense17[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dense18 (Dense)  (None, 512)          262656      weather_snn_14_dropout18[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_dropout19 (Alpha (None, 512)          0           weather_snn_14_dense18[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_output (Dense)   (None, 2)            1024        weather_snn_14_dropout19[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_distribution (Di ((None, 1), (None, 1 0           weather_snn_14_output[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_14_projection (Dens (None, 50)           25600       weather_snn_14_dropout13[0][0]   \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 15 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_15\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_15_feature_vector ( [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout1 (AlphaD (None, 123)          0           weather_snn_15_feature_vector[0][\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense1 (Dense)   (None, 512)          63488       weather_snn_15_dropout1[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout2 (AlphaD (None, 512)          0           weather_snn_15_dense1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense2 (Dense)   (None, 512)          262656      weather_snn_15_dropout2[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout3 (AlphaD (None, 512)          0           weather_snn_15_dense2[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense3 (Dense)   (None, 512)          262656      weather_snn_15_dropout3[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout4 (AlphaD (None, 512)          0           weather_snn_15_dense3[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense4 (Dense)   (None, 512)          262656      weather_snn_15_dropout4[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout5 (AlphaD (None, 512)          0           weather_snn_15_dense4[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense5 (Dense)   (None, 512)          262656      weather_snn_15_dropout5[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout6 (AlphaD (None, 512)          0           weather_snn_15_dense5[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense6 (Dense)   (None, 512)          262656      weather_snn_15_dropout6[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout7 (AlphaD (None, 512)          0           weather_snn_15_dense6[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense7 (Dense)   (None, 512)          262656      weather_snn_15_dropout7[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout8 (AlphaD (None, 512)          0           weather_snn_15_dense7[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense8 (Dense)   (None, 512)          262656      weather_snn_15_dropout8[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout9 (AlphaD (None, 512)          0           weather_snn_15_dense8[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense9 (Dense)   (None, 512)          262656      weather_snn_15_dropout9[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout10 (Alpha (None, 512)          0           weather_snn_15_dense9[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense10 (Dense)  (None, 512)          262656      weather_snn_15_dropout10[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout11 (Alpha (None, 512)          0           weather_snn_15_dense10[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense11 (Dense)  (None, 512)          262656      weather_snn_15_dropout11[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout12 (Alpha (None, 512)          0           weather_snn_15_dense11[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense12 (Dense)  (None, 512)          262656      weather_snn_15_dropout12[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout13 (Alpha (None, 512)          0           weather_snn_15_dense12[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense13 (Dense)  (None, 512)          262656      weather_snn_15_dropout13[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout14 (Alpha (None, 512)          0           weather_snn_15_dense13[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense14 (Dense)  (None, 512)          262656      weather_snn_15_dropout14[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout15 (Alpha (None, 512)          0           weather_snn_15_dense14[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense15 (Dense)  (None, 512)          262656      weather_snn_15_dropout15[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout16 (Alpha (None, 512)          0           weather_snn_15_dense15[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense16 (Dense)  (None, 512)          262656      weather_snn_15_dropout16[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout17 (Alpha (None, 512)          0           weather_snn_15_dense16[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense17 (Dense)  (None, 512)          262656      weather_snn_15_dropout17[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout18 (Alpha (None, 512)          0           weather_snn_15_dense17[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dense18 (Dense)  (None, 512)          262656      weather_snn_15_dropout18[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_dropout19 (Alpha (None, 512)          0           weather_snn_15_dense18[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_output (Dense)   (None, 2)            1024        weather_snn_15_dropout19[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_distribution (Di ((None, 1), (None, 1 0           weather_snn_15_output[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_15_projection (Dens (None, 50)           25600       weather_snn_15_dropout13[0][0]   \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 16 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_16\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_16_feature_vector ( [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout1 (AlphaD (None, 123)          0           weather_snn_16_feature_vector[0][\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense1 (Dense)   (None, 512)          63488       weather_snn_16_dropout1[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout2 (AlphaD (None, 512)          0           weather_snn_16_dense1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense2 (Dense)   (None, 512)          262656      weather_snn_16_dropout2[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout3 (AlphaD (None, 512)          0           weather_snn_16_dense2[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense3 (Dense)   (None, 512)          262656      weather_snn_16_dropout3[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout4 (AlphaD (None, 512)          0           weather_snn_16_dense3[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense4 (Dense)   (None, 512)          262656      weather_snn_16_dropout4[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout5 (AlphaD (None, 512)          0           weather_snn_16_dense4[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense5 (Dense)   (None, 512)          262656      weather_snn_16_dropout5[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout6 (AlphaD (None, 512)          0           weather_snn_16_dense5[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense6 (Dense)   (None, 512)          262656      weather_snn_16_dropout6[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout7 (AlphaD (None, 512)          0           weather_snn_16_dense6[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense7 (Dense)   (None, 512)          262656      weather_snn_16_dropout7[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout8 (AlphaD (None, 512)          0           weather_snn_16_dense7[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense8 (Dense)   (None, 512)          262656      weather_snn_16_dropout8[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout9 (AlphaD (None, 512)          0           weather_snn_16_dense8[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense9 (Dense)   (None, 512)          262656      weather_snn_16_dropout9[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout10 (Alpha (None, 512)          0           weather_snn_16_dense9[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense10 (Dense)  (None, 512)          262656      weather_snn_16_dropout10[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout11 (Alpha (None, 512)          0           weather_snn_16_dense10[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense11 (Dense)  (None, 512)          262656      weather_snn_16_dropout11[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout12 (Alpha (None, 512)          0           weather_snn_16_dense11[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense12 (Dense)  (None, 512)          262656      weather_snn_16_dropout12[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout13 (Alpha (None, 512)          0           weather_snn_16_dense12[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense13 (Dense)  (None, 512)          262656      weather_snn_16_dropout13[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout14 (Alpha (None, 512)          0           weather_snn_16_dense13[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense14 (Dense)  (None, 512)          262656      weather_snn_16_dropout14[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout15 (Alpha (None, 512)          0           weather_snn_16_dense14[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense15 (Dense)  (None, 512)          262656      weather_snn_16_dropout15[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout16 (Alpha (None, 512)          0           weather_snn_16_dense15[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense16 (Dense)  (None, 512)          262656      weather_snn_16_dropout16[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout17 (Alpha (None, 512)          0           weather_snn_16_dense16[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense17 (Dense)  (None, 512)          262656      weather_snn_16_dropout17[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout18 (Alpha (None, 512)          0           weather_snn_16_dense17[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dense18 (Dense)  (None, 512)          262656      weather_snn_16_dropout18[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_dropout19 (Alpha (None, 512)          0           weather_snn_16_dense18[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_output (Dense)   (None, 2)            1024        weather_snn_16_dropout19[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_distribution (Di ((None, 1), (None, 1 0           weather_snn_16_output[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_16_projection (Dens (None, 50)           25600       weather_snn_16_dropout13[0][0]   \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 17 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_17\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_17_feature_vector ( [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout1 (AlphaD (None, 123)          0           weather_snn_17_feature_vector[0][\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense1 (Dense)   (None, 512)          63488       weather_snn_17_dropout1[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout2 (AlphaD (None, 512)          0           weather_snn_17_dense1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense2 (Dense)   (None, 512)          262656      weather_snn_17_dropout2[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout3 (AlphaD (None, 512)          0           weather_snn_17_dense2[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense3 (Dense)   (None, 512)          262656      weather_snn_17_dropout3[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout4 (AlphaD (None, 512)          0           weather_snn_17_dense3[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense4 (Dense)   (None, 512)          262656      weather_snn_17_dropout4[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout5 (AlphaD (None, 512)          0           weather_snn_17_dense4[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense5 (Dense)   (None, 512)          262656      weather_snn_17_dropout5[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout6 (AlphaD (None, 512)          0           weather_snn_17_dense5[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense6 (Dense)   (None, 512)          262656      weather_snn_17_dropout6[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout7 (AlphaD (None, 512)          0           weather_snn_17_dense6[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense7 (Dense)   (None, 512)          262656      weather_snn_17_dropout7[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout8 (AlphaD (None, 512)          0           weather_snn_17_dense7[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense8 (Dense)   (None, 512)          262656      weather_snn_17_dropout8[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout9 (AlphaD (None, 512)          0           weather_snn_17_dense8[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense9 (Dense)   (None, 512)          262656      weather_snn_17_dropout9[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout10 (Alpha (None, 512)          0           weather_snn_17_dense9[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense10 (Dense)  (None, 512)          262656      weather_snn_17_dropout10[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout11 (Alpha (None, 512)          0           weather_snn_17_dense10[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense11 (Dense)  (None, 512)          262656      weather_snn_17_dropout11[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout12 (Alpha (None, 512)          0           weather_snn_17_dense11[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense12 (Dense)  (None, 512)          262656      weather_snn_17_dropout12[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout13 (Alpha (None, 512)          0           weather_snn_17_dense12[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense13 (Dense)  (None, 512)          262656      weather_snn_17_dropout13[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout14 (Alpha (None, 512)          0           weather_snn_17_dense13[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense14 (Dense)  (None, 512)          262656      weather_snn_17_dropout14[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout15 (Alpha (None, 512)          0           weather_snn_17_dense14[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense15 (Dense)  (None, 512)          262656      weather_snn_17_dropout15[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout16 (Alpha (None, 512)          0           weather_snn_17_dense15[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense16 (Dense)  (None, 512)          262656      weather_snn_17_dropout16[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout17 (Alpha (None, 512)          0           weather_snn_17_dense16[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense17 (Dense)  (None, 512)          262656      weather_snn_17_dropout17[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout18 (Alpha (None, 512)          0           weather_snn_17_dense17[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dense18 (Dense)  (None, 512)          262656      weather_snn_17_dropout18[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_dropout19 (Alpha (None, 512)          0           weather_snn_17_dense18[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_output (Dense)   (None, 2)            1024        weather_snn_17_dropout19[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_distribution (Di ((None, 1), (None, 1 0           weather_snn_17_output[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_17_projection (Dens (None, 50)           25600       weather_snn_17_dropout13[0][0]   \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 18 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_18\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_18_feature_vector ( [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout1 (AlphaD (None, 123)          0           weather_snn_18_feature_vector[0][\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense1 (Dense)   (None, 512)          63488       weather_snn_18_dropout1[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout2 (AlphaD (None, 512)          0           weather_snn_18_dense1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense2 (Dense)   (None, 512)          262656      weather_snn_18_dropout2[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout3 (AlphaD (None, 512)          0           weather_snn_18_dense2[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense3 (Dense)   (None, 512)          262656      weather_snn_18_dropout3[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout4 (AlphaD (None, 512)          0           weather_snn_18_dense3[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense4 (Dense)   (None, 512)          262656      weather_snn_18_dropout4[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout5 (AlphaD (None, 512)          0           weather_snn_18_dense4[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense5 (Dense)   (None, 512)          262656      weather_snn_18_dropout5[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout6 (AlphaD (None, 512)          0           weather_snn_18_dense5[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense6 (Dense)   (None, 512)          262656      weather_snn_18_dropout6[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout7 (AlphaD (None, 512)          0           weather_snn_18_dense6[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense7 (Dense)   (None, 512)          262656      weather_snn_18_dropout7[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout8 (AlphaD (None, 512)          0           weather_snn_18_dense7[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense8 (Dense)   (None, 512)          262656      weather_snn_18_dropout8[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout9 (AlphaD (None, 512)          0           weather_snn_18_dense8[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense9 (Dense)   (None, 512)          262656      weather_snn_18_dropout9[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout10 (Alpha (None, 512)          0           weather_snn_18_dense9[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense10 (Dense)  (None, 512)          262656      weather_snn_18_dropout10[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout11 (Alpha (None, 512)          0           weather_snn_18_dense10[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense11 (Dense)  (None, 512)          262656      weather_snn_18_dropout11[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout12 (Alpha (None, 512)          0           weather_snn_18_dense11[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense12 (Dense)  (None, 512)          262656      weather_snn_18_dropout12[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout13 (Alpha (None, 512)          0           weather_snn_18_dense12[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense13 (Dense)  (None, 512)          262656      weather_snn_18_dropout13[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout14 (Alpha (None, 512)          0           weather_snn_18_dense13[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense14 (Dense)  (None, 512)          262656      weather_snn_18_dropout14[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout15 (Alpha (None, 512)          0           weather_snn_18_dense14[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense15 (Dense)  (None, 512)          262656      weather_snn_18_dropout15[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout16 (Alpha (None, 512)          0           weather_snn_18_dense15[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense16 (Dense)  (None, 512)          262656      weather_snn_18_dropout16[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout17 (Alpha (None, 512)          0           weather_snn_18_dense16[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense17 (Dense)  (None, 512)          262656      weather_snn_18_dropout17[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout18 (Alpha (None, 512)          0           weather_snn_18_dense17[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dense18 (Dense)  (None, 512)          262656      weather_snn_18_dropout18[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_dropout19 (Alpha (None, 512)          0           weather_snn_18_dense18[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_output (Dense)   (None, 2)            1024        weather_snn_18_dropout19[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_distribution (Di ((None, 1), (None, 1 0           weather_snn_18_output[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_18_projection (Dens (None, 50)           25600       weather_snn_18_dropout13[0][0]   \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 19 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_19\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_19_feature_vector ( [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout1 (AlphaD (None, 123)          0           weather_snn_19_feature_vector[0][\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense1 (Dense)   (None, 512)          63488       weather_snn_19_dropout1[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout2 (AlphaD (None, 512)          0           weather_snn_19_dense1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense2 (Dense)   (None, 512)          262656      weather_snn_19_dropout2[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout3 (AlphaD (None, 512)          0           weather_snn_19_dense2[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense3 (Dense)   (None, 512)          262656      weather_snn_19_dropout3[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout4 (AlphaD (None, 512)          0           weather_snn_19_dense3[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense4 (Dense)   (None, 512)          262656      weather_snn_19_dropout4[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout5 (AlphaD (None, 512)          0           weather_snn_19_dense4[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense5 (Dense)   (None, 512)          262656      weather_snn_19_dropout5[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout6 (AlphaD (None, 512)          0           weather_snn_19_dense5[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense6 (Dense)   (None, 512)          262656      weather_snn_19_dropout6[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout7 (AlphaD (None, 512)          0           weather_snn_19_dense6[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense7 (Dense)   (None, 512)          262656      weather_snn_19_dropout7[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout8 (AlphaD (None, 512)          0           weather_snn_19_dense7[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense8 (Dense)   (None, 512)          262656      weather_snn_19_dropout8[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout9 (AlphaD (None, 512)          0           weather_snn_19_dense8[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense9 (Dense)   (None, 512)          262656      weather_snn_19_dropout9[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout10 (Alpha (None, 512)          0           weather_snn_19_dense9[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense10 (Dense)  (None, 512)          262656      weather_snn_19_dropout10[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout11 (Alpha (None, 512)          0           weather_snn_19_dense10[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense11 (Dense)  (None, 512)          262656      weather_snn_19_dropout11[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout12 (Alpha (None, 512)          0           weather_snn_19_dense11[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense12 (Dense)  (None, 512)          262656      weather_snn_19_dropout12[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout13 (Alpha (None, 512)          0           weather_snn_19_dense12[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense13 (Dense)  (None, 512)          262656      weather_snn_19_dropout13[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout14 (Alpha (None, 512)          0           weather_snn_19_dense13[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense14 (Dense)  (None, 512)          262656      weather_snn_19_dropout14[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout15 (Alpha (None, 512)          0           weather_snn_19_dense14[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense15 (Dense)  (None, 512)          262656      weather_snn_19_dropout15[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout16 (Alpha (None, 512)          0           weather_snn_19_dense15[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense16 (Dense)  (None, 512)          262656      weather_snn_19_dropout16[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout17 (Alpha (None, 512)          0           weather_snn_19_dense16[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense17 (Dense)  (None, 512)          262656      weather_snn_19_dropout17[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout18 (Alpha (None, 512)          0           weather_snn_19_dense17[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dense18 (Dense)  (None, 512)          262656      weather_snn_19_dropout18[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_dropout19 (Alpha (None, 512)          0           weather_snn_19_dense18[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_output (Dense)   (None, 2)            1024        weather_snn_19_dropout19[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_distribution (Di ((None, 1), (None, 1 0           weather_snn_19_output[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_19_projection (Dens (None, 50)           25600       weather_snn_19_dropout13[0][0]   \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n",
      "========================================\n",
      " Self-Normalizing Network 20 for weather prediction\n",
      "========================================\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"weather_snn_20\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "weather_snn_20_feature_vector ( [(None, 123)]        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout1 (AlphaD (None, 123)          0           weather_snn_20_feature_vector[0][\n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense1 (Dense)   (None, 512)          63488       weather_snn_20_dropout1[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout2 (AlphaD (None, 512)          0           weather_snn_20_dense1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense2 (Dense)   (None, 512)          262656      weather_snn_20_dropout2[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout3 (AlphaD (None, 512)          0           weather_snn_20_dense2[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense3 (Dense)   (None, 512)          262656      weather_snn_20_dropout3[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout4 (AlphaD (None, 512)          0           weather_snn_20_dense3[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense4 (Dense)   (None, 512)          262656      weather_snn_20_dropout4[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout5 (AlphaD (None, 512)          0           weather_snn_20_dense4[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense5 (Dense)   (None, 512)          262656      weather_snn_20_dropout5[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout6 (AlphaD (None, 512)          0           weather_snn_20_dense5[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense6 (Dense)   (None, 512)          262656      weather_snn_20_dropout6[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout7 (AlphaD (None, 512)          0           weather_snn_20_dense6[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense7 (Dense)   (None, 512)          262656      weather_snn_20_dropout7[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout8 (AlphaD (None, 512)          0           weather_snn_20_dense7[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense8 (Dense)   (None, 512)          262656      weather_snn_20_dropout8[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout9 (AlphaD (None, 512)          0           weather_snn_20_dense8[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense9 (Dense)   (None, 512)          262656      weather_snn_20_dropout9[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout10 (Alpha (None, 512)          0           weather_snn_20_dense9[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense10 (Dense)  (None, 512)          262656      weather_snn_20_dropout10[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout11 (Alpha (None, 512)          0           weather_snn_20_dense10[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense11 (Dense)  (None, 512)          262656      weather_snn_20_dropout11[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout12 (Alpha (None, 512)          0           weather_snn_20_dense11[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense12 (Dense)  (None, 512)          262656      weather_snn_20_dropout12[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout13 (Alpha (None, 512)          0           weather_snn_20_dense12[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense13 (Dense)  (None, 512)          262656      weather_snn_20_dropout13[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout14 (Alpha (None, 512)          0           weather_snn_20_dense13[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense14 (Dense)  (None, 512)          262656      weather_snn_20_dropout14[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout15 (Alpha (None, 512)          0           weather_snn_20_dense14[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense15 (Dense)  (None, 512)          262656      weather_snn_20_dropout15[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout16 (Alpha (None, 512)          0           weather_snn_20_dense15[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense16 (Dense)  (None, 512)          262656      weather_snn_20_dropout16[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout17 (Alpha (None, 512)          0           weather_snn_20_dense16[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense17 (Dense)  (None, 512)          262656      weather_snn_20_dropout17[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout18 (Alpha (None, 512)          0           weather_snn_20_dense17[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dense18 (Dense)  (None, 512)          262656      weather_snn_20_dropout18[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_dropout19 (Alpha (None, 512)          0           weather_snn_20_dense18[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_output (Dense)   (None, 2)            1024        weather_snn_20_dropout19[0][0]   \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_distribution (Di ((None, 1), (None, 1 0           weather_snn_20_output[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "weather_snn_20_projection (Dens (None, 50)           25600       weather_snn_20_dropout13[0][0]   \n",
      "==================================================================================================\n",
      "Total params: 4,555,264\n",
      "Trainable params: 4,555,264\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "\n"
     ]
    }
   ],
   "source": [
    "for nn_idx in range(ensemble_size):\n",
    "    model_name = f'weather_snn_{nn_idx + 1}'\n",
    "    serialization_name = os.path.join(model_dir, model_name + '.h5')\n",
    "    regression_output_name = f'{model_name}_distribution'\n",
    "    projection_output_name = f'{model_name}_projection'\n",
    "    printable_name = f'Self-Normalizing Network ' \\\n",
    "                     f'{nn_idx + 1} for weather prediction'\n",
    "    print('========================================')\n",
    "    print(' ' + printable_name)\n",
    "    print('========================================')\n",
    "    print('')\n",
    "    new_model = build_neural_network(\n",
    "        input_size=num_features,\n",
    "        layer_size=nn_layer_size,\n",
    "        n_layers=nn_depth,\n",
    "        dropout_rate=nn_dropout,\n",
    "        scale_coeff=postprocessing_scalers[nn_idx].scale_[0],\n",
    "        nn_name=model_name\n",
    "    )\n",
    "    new_model.summary()\n",
    "    print('')\n",
    "    new_model.load_weights(serialization_name)\n",
    "    deep_ensemble.append(new_model)\n",
    "    del new_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "ff56ae34",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Row number is 50000.\n",
      "Column number is 129.\n"
     ]
    }
   ],
   "source": [
    "df_in = pd.read_csv(dev_in_name)\n",
    "print(f'Row number is {df_in.shape[0]}.')\n",
    "print(f'Column number is {df_in.shape[1]}.')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "81da8770",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>fact_time</th>\n",
       "      <th>fact_latitude</th>\n",
       "      <th>fact_longitude</th>\n",
       "      <th>fact_temperature</th>\n",
       "      <th>fact_cwsm_class</th>\n",
       "      <th>climate</th>\n",
       "      <th>topography_bathymetry</th>\n",
       "      <th>sun_elevation</th>\n",
       "      <th>climate_temperature</th>\n",
       "      <th>climate_pressure</th>\n",
       "      <th>...</th>\n",
       "      <th>cmc_0_1_66_0_grad</th>\n",
       "      <th>cmc_0_1_66_0_next</th>\n",
       "      <th>cmc_0_1_67_0_grad</th>\n",
       "      <th>cmc_0_1_67_0_next</th>\n",
       "      <th>cmc_0_1_68_0_grad</th>\n",
       "      <th>cmc_0_1_68_0_next</th>\n",
       "      <th>gfs_2m_dewpoint_grad</th>\n",
       "      <th>gfs_2m_dewpoint_next</th>\n",
       "      <th>gfs_total_clouds_cover_low_grad</th>\n",
       "      <th>gfs_total_clouds_cover_low_next</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1539162000</td>\n",
       "      <td>-40.350000</td>\n",
       "      <td>-9.880000</td>\n",
       "      <td>11.0</td>\n",
       "      <td>10</td>\n",
       "      <td>tropical</td>\n",
       "      <td>-843.0</td>\n",
       "      <td>31.782490</td>\n",
       "      <td>10.070714</td>\n",
       "      <td>765.631228</td>\n",
       "      <td>...</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>0.0000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.505035</td>\n",
       "      <td>2.647577</td>\n",
       "      <td>2.0</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1545006600</td>\n",
       "      <td>53.421299</td>\n",
       "      <td>-6.270070</td>\n",
       "      <td>4.0</td>\n",
       "      <td>10</td>\n",
       "      <td>mild temperate</td>\n",
       "      <td>67.0</td>\n",
       "      <td>-59.691521</td>\n",
       "      <td>7.005000</td>\n",
       "      <td>752.897615</td>\n",
       "      <td>...</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>0.0000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>-0.400024</td>\n",
       "      <td>1.249994</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1540094400</td>\n",
       "      <td>-19.757700</td>\n",
       "      <td>63.361000</td>\n",
       "      <td>26.0</td>\n",
       "      <td>10</td>\n",
       "      <td>dry</td>\n",
       "      <td>6.0</td>\n",
       "      <td>35.250889</td>\n",
       "      <td>23.327143</td>\n",
       "      <td>763.115016</td>\n",
       "      <td>...</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>0.0000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.100006</td>\n",
       "      <td>21.050012</td>\n",
       "      <td>-1.0</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1552611600</td>\n",
       "      <td>35.245899</td>\n",
       "      <td>47.009201</td>\n",
       "      <td>5.0</td>\n",
       "      <td>10</td>\n",
       "      <td>mild temperate</td>\n",
       "      <td>1390.0</td>\n",
       "      <td>-23.755615</td>\n",
       "      <td>3.109286</td>\n",
       "      <td>609.419333</td>\n",
       "      <td>...</td>\n",
       "      <td>1.69672</td>\n",
       "      <td>5.1653</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000049</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>-1.500000</td>\n",
       "      <td>-0.349982</td>\n",
       "      <td>-12.0</td>\n",
       "      <td>81.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1545631200</td>\n",
       "      <td>26.633333</td>\n",
       "      <td>118.150000</td>\n",
       "      <td>14.0</td>\n",
       "      <td>20</td>\n",
       "      <td>mild temperate</td>\n",
       "      <td>210.0</td>\n",
       "      <td>33.040438</td>\n",
       "      <td>12.172143</td>\n",
       "      <td>734.678037</td>\n",
       "      <td>...</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>0.0000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.102081</td>\n",
       "      <td>11.513879</td>\n",
       "      <td>-15.0</td>\n",
       "      <td>83.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 129 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "    fact_time  fact_latitude  fact_longitude  fact_temperature  \\\n",
       "0  1539162000     -40.350000       -9.880000              11.0   \n",
       "1  1545006600      53.421299       -6.270070               4.0   \n",
       "2  1540094400     -19.757700       63.361000              26.0   \n",
       "3  1552611600      35.245899       47.009201               5.0   \n",
       "4  1545631200      26.633333      118.150000              14.0   \n",
       "\n",
       "   fact_cwsm_class         climate  topography_bathymetry  sun_elevation  \\\n",
       "0               10        tropical                 -843.0      31.782490   \n",
       "1               10  mild temperate                   67.0     -59.691521   \n",
       "2               10             dry                    6.0      35.250889   \n",
       "3               10  mild temperate                 1390.0     -23.755615   \n",
       "4               20  mild temperate                  210.0      33.040438   \n",
       "\n",
       "   climate_temperature  climate_pressure  ...  cmc_0_1_66_0_grad  \\\n",
       "0            10.070714        765.631228  ...            0.00000   \n",
       "1             7.005000        752.897615  ...            0.00000   \n",
       "2            23.327143        763.115016  ...            0.00000   \n",
       "3             3.109286        609.419333  ...            1.69672   \n",
       "4            12.172143        734.678037  ...            0.00000   \n",
       "\n",
       "   cmc_0_1_66_0_next  cmc_0_1_67_0_grad  cmc_0_1_67_0_next  cmc_0_1_68_0_grad  \\\n",
       "0             0.0000                0.0           0.000000                0.0   \n",
       "1             0.0000                0.0           0.000000                0.0   \n",
       "2             0.0000                0.0           0.000000                0.0   \n",
       "3             5.1653                0.0           0.000049                0.0   \n",
       "4             0.0000                0.0           0.000000                0.0   \n",
       "\n",
       "   cmc_0_1_68_0_next  gfs_2m_dewpoint_grad  gfs_2m_dewpoint_next  \\\n",
       "0                0.0              0.505035              2.647577   \n",
       "1                0.0             -0.400024              1.249994   \n",
       "2                0.0              0.100006             21.050012   \n",
       "3                0.0             -1.500000             -0.349982   \n",
       "4                0.0              0.102081             11.513879   \n",
       "\n",
       "   gfs_total_clouds_cover_low_grad  gfs_total_clouds_cover_low_next  \n",
       "0                              2.0                              2.0  \n",
       "1                              0.0                              0.0  \n",
       "2                             -1.0                              1.0  \n",
       "3                            -12.0                             81.0  \n",
       "4                            -15.0                             83.0  \n",
       "\n",
       "[5 rows x 129 columns]"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_in.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "cc8b797f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Row number is 50000.\n",
      "Column number is 129.\n"
     ]
    }
   ],
   "source": [
    "df_out = pd.read_csv(dev_out_name)\n",
    "print(f'Row number is {df_out.shape[0]}.')\n",
    "print(f'Column number is {df_out.shape[1]}.')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "636d330e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>fact_time</th>\n",
       "      <th>fact_latitude</th>\n",
       "      <th>fact_longitude</th>\n",
       "      <th>fact_temperature</th>\n",
       "      <th>fact_cwsm_class</th>\n",
       "      <th>climate</th>\n",
       "      <th>topography_bathymetry</th>\n",
       "      <th>sun_elevation</th>\n",
       "      <th>climate_temperature</th>\n",
       "      <th>climate_pressure</th>\n",
       "      <th>...</th>\n",
       "      <th>cmc_0_1_66_0_grad</th>\n",
       "      <th>cmc_0_1_66_0_next</th>\n",
       "      <th>cmc_0_1_67_0_grad</th>\n",
       "      <th>cmc_0_1_67_0_next</th>\n",
       "      <th>cmc_0_1_68_0_grad</th>\n",
       "      <th>cmc_0_1_68_0_next</th>\n",
       "      <th>gfs_2m_dewpoint_grad</th>\n",
       "      <th>gfs_2m_dewpoint_next</th>\n",
       "      <th>gfs_total_clouds_cover_low_grad</th>\n",
       "      <th>gfs_total_clouds_cover_low_next</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1563124500</td>\n",
       "      <td>42.614899</td>\n",
       "      <td>-89.590401</td>\n",
       "      <td>28.0</td>\n",
       "      <td>0</td>\n",
       "      <td>snow</td>\n",
       "      <td>313.0</td>\n",
       "      <td>65.438862</td>\n",
       "      <td>19.171429</td>\n",
       "      <td>737.750555</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>2.399994</td>\n",
       "      <td>23.450006</td>\n",
       "      <td>26.0</td>\n",
       "      <td>71.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1565951400</td>\n",
       "      <td>61.885201</td>\n",
       "      <td>34.154701</td>\n",
       "      <td>18.0</td>\n",
       "      <td>10</td>\n",
       "      <td>snow</td>\n",
       "      <td>34.0</td>\n",
       "      <td>40.277528</td>\n",
       "      <td>15.127857</td>\n",
       "      <td>751.846067</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>-0.300018</td>\n",
       "      <td>9.449976</td>\n",
       "      <td>21.0</td>\n",
       "      <td>40.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1563850200</td>\n",
       "      <td>38.468201</td>\n",
       "      <td>43.332298</td>\n",
       "      <td>17.0</td>\n",
       "      <td>10</td>\n",
       "      <td>snow</td>\n",
       "      <td>1652.0</td>\n",
       "      <td>-0.237116</td>\n",
       "      <td>20.117857</td>\n",
       "      <td>609.203551</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.849335</td>\n",
       "      <td>3.365320</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1562806380</td>\n",
       "      <td>38.791000</td>\n",
       "      <td>-97.652199</td>\n",
       "      <td>28.0</td>\n",
       "      <td>0</td>\n",
       "      <td>snow</td>\n",
       "      <td>383.0</td>\n",
       "      <td>19.207866</td>\n",
       "      <td>33.193571</td>\n",
       "      <td>724.725255</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>-2.486816</td>\n",
       "      <td>16.263177</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1564509600</td>\n",
       "      <td>58.767200</td>\n",
       "      <td>-111.116997</td>\n",
       "      <td>20.0</td>\n",
       "      <td>20</td>\n",
       "      <td>snow</td>\n",
       "      <td>235.0</td>\n",
       "      <td>46.174496</td>\n",
       "      <td>19.481429</td>\n",
       "      <td>740.954837</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>-1.200012</td>\n",
       "      <td>7.050012</td>\n",
       "      <td>-1.0</td>\n",
       "      <td>5.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 129 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "    fact_time  fact_latitude  fact_longitude  fact_temperature  \\\n",
       "0  1563124500      42.614899      -89.590401              28.0   \n",
       "1  1565951400      61.885201       34.154701              18.0   \n",
       "2  1563850200      38.468201       43.332298              17.0   \n",
       "3  1562806380      38.791000      -97.652199              28.0   \n",
       "4  1564509600      58.767200     -111.116997              20.0   \n",
       "\n",
       "   fact_cwsm_class climate  topography_bathymetry  sun_elevation  \\\n",
       "0                0    snow                  313.0      65.438862   \n",
       "1               10    snow                   34.0      40.277528   \n",
       "2               10    snow                 1652.0      -0.237116   \n",
       "3                0    snow                  383.0      19.207866   \n",
       "4               20    snow                  235.0      46.174496   \n",
       "\n",
       "   climate_temperature  climate_pressure  ...  cmc_0_1_66_0_grad  \\\n",
       "0            19.171429        737.750555  ...                0.0   \n",
       "1            15.127857        751.846067  ...                0.0   \n",
       "2            20.117857        609.203551  ...                0.0   \n",
       "3            33.193571        724.725255  ...                0.0   \n",
       "4            19.481429        740.954837  ...                0.0   \n",
       "\n",
       "   cmc_0_1_66_0_next  cmc_0_1_67_0_grad  cmc_0_1_67_0_next  cmc_0_1_68_0_grad  \\\n",
       "0                0.0                0.0                0.0                0.0   \n",
       "1                0.0                0.0                0.0                0.0   \n",
       "2                0.0                0.0                0.0                0.0   \n",
       "3                0.0                0.0                0.0                0.0   \n",
       "4                0.0                0.0                0.0                0.0   \n",
       "\n",
       "   cmc_0_1_68_0_next  gfs_2m_dewpoint_grad  gfs_2m_dewpoint_next  \\\n",
       "0                0.0              2.399994             23.450006   \n",
       "1                0.0             -0.300018              9.449976   \n",
       "2                0.0              1.849335              3.365320   \n",
       "3                0.0             -2.486816             16.263177   \n",
       "4                0.0             -1.200012              7.050012   \n",
       "\n",
       "   gfs_total_clouds_cover_low_grad  gfs_total_clouds_cover_low_next  \n",
       "0                             26.0                             71.0  \n",
       "1                             21.0                             40.0  \n",
       "2                              0.0                              0.0  \n",
       "3                              0.0                              0.0  \n",
       "4                             -1.0                              5.0  \n",
       "\n",
       "[5 rows x 129 columns]"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_out.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "1a49d92f",
   "metadata": {},
   "outputs": [],
   "source": [
    "inputs = np.vstack([\n",
    "    df_in.iloc[:, 6:].to_numpy().astype(np.float64),\n",
    "    df_out.iloc[:, 6:].to_numpy().astype(np.float64)\n",
    "])\n",
    "targets = np.concatenate([\n",
    "    df_in['fact_temperature'].to_numpy().astype(np.float64),\n",
    "    df_out['fact_temperature'].to_numpy().astype(np.float64)\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "6440aa2e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The system predicts each sample from devset in 0.066 msecs.\n"
     ]
    }
   ],
   "source": [
    "start_inference_time = time.time()\n",
    "all_preds = predict_by_ensemble(\n",
    "    input_data=inputs,\n",
    "    preprocessing=common_preprocessor,\n",
    "    postprocessing=postprocessing_scalers,\n",
    "    ensemble=deep_ensemble,\n",
    "    minibatch=BATCH_SIZE\n",
    ")\n",
    "end_inference_time = time.time()\n",
    "inference_duration = end_inference_time - start_inference_time\n",
    "inference_duration *= 1000.0\n",
    "print('The system predicts each sample from devset in ' \\\n",
    "      '{0:.3f} msecs.'.format(inference_duration / inputs.shape[0]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "6a8d0bc4",
   "metadata": {},
   "outputs": [],
   "source": [
    "all_uncertainty = ensemble_uncertainties_regression(all_preds)\n",
    "uncertainties = all_uncertainty['tvar']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "d053138e",
   "metadata": {},
   "outputs": [],
   "source": [
    "all_preds_mean = all_preds[:,:,0]\n",
    "avg_preds = np.squeeze(np.mean(all_preds_mean, axis=0))\n",
    "errors = (avg_preds - targets) ** 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "c2b565b6",
   "metadata": {},
   "outputs": [],
   "source": [
    "rejection_mse = calc_uncertainty_regection_curve(errors, uncertainties)\n",
    "retention_mse = rejection_mse[::-1]\n",
    "retention_fractions = np.linspace(0, 1, len(retention_mse))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "id": "c7a6ef26",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "R-AUC MSE = 1.0145584671532237.\n"
     ]
    }
   ],
   "source": [
    "print(f'R-AUC MSE = {np.mean(retention_mse)}.')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "id": "8668a794",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "models\\yandex-shifts\\weather-baseline-catboost\\seed1.cbm = True\n",
      "CatBoost 1 is loaded from models\\yandex-shifts\\weather-baseline-catboost\\seed1.cbm.\n",
      "CatBoost 1 predicts on devset.\n",
      "\n",
      "models\\yandex-shifts\\weather-baseline-catboost\\seed2.cbm = True\n",
      "CatBoost 2 is loaded from models\\yandex-shifts\\weather-baseline-catboost\\seed2.cbm.\n",
      "CatBoost 2 predicts on devset.\n",
      "\n",
      "models\\yandex-shifts\\weather-baseline-catboost\\seed3.cbm = True\n",
      "CatBoost 3 is loaded from models\\yandex-shifts\\weather-baseline-catboost\\seed3.cbm.\n",
      "CatBoost 3 predicts on devset.\n",
      "\n",
      "models\\yandex-shifts\\weather-baseline-catboost\\seed4.cbm = True\n",
      "CatBoost 4 is loaded from models\\yandex-shifts\\weather-baseline-catboost\\seed4.cbm.\n",
      "CatBoost 4 predicts on devset.\n",
      "\n",
      "models\\yandex-shifts\\weather-baseline-catboost\\seed5.cbm = True\n",
      "CatBoost 5 is loaded from models\\yandex-shifts\\weather-baseline-catboost\\seed5.cbm.\n",
      "CatBoost 5 predicts on devset.\n",
      "\n",
      "models\\yandex-shifts\\weather-baseline-catboost\\seed6.cbm = True\n",
      "CatBoost 6 is loaded from models\\yandex-shifts\\weather-baseline-catboost\\seed6.cbm.\n",
      "CatBoost 6 predicts on devset.\n",
      "\n",
      "models\\yandex-shifts\\weather-baseline-catboost\\seed7.cbm = True\n",
      "CatBoost 7 is loaded from models\\yandex-shifts\\weather-baseline-catboost\\seed7.cbm.\n",
      "CatBoost 7 predicts on devset.\n",
      "\n",
      "models\\yandex-shifts\\weather-baseline-catboost\\seed8.cbm = True\n",
      "CatBoost 8 is loaded from models\\yandex-shifts\\weather-baseline-catboost\\seed8.cbm.\n",
      "CatBoost 8 predicts on devset.\n",
      "\n",
      "models\\yandex-shifts\\weather-baseline-catboost\\seed9.cbm = True\n",
      "CatBoost 9 is loaded from models\\yandex-shifts\\weather-baseline-catboost\\seed9.cbm.\n",
      "CatBoost 9 predicts on devset.\n",
      "\n",
      "models\\yandex-shifts\\weather-baseline-catboost\\seed10.cbm = True\n",
      "CatBoost 10 is loaded from models\\yandex-shifts\\weather-baseline-catboost\\seed10.cbm.\n",
      "CatBoost 10 predicts on devset.\n",
      "\n",
      "R-AUC MSE of CatBoost Ens = 1.226643512318076.\n"
     ]
    }
   ],
   "source": [
    "if CATBOOST_EXISTS and os.path.isdir(baseline_model_dir):\n",
    "    catboost_ensemble_size = 10\n",
    "    all_catboost_preds = []\n",
    "    for catboost_ind in range(1, catboost_ensemble_size + 1):\n",
    "        new_catboost = CatBoostRegressor()\n",
    "        cur_catboost_name = os.path.join(\n",
    "            baseline_model_dir,\n",
    "            f'seed{catboost_ind}.cbm'\n",
    "        )\n",
    "        print(f'{cur_catboost_name} = {os.path.isfile(cur_catboost_name)}')\n",
    "        new_catboost.load_model(cur_catboost_name)\n",
    "        print(f'CatBoost {catboost_ind} is loaded from {cur_catboost_name}.')\n",
    "        new_catboost_preds = new_catboost.predict(inputs)\n",
    "        if not isinstance(new_catboost_preds, np.ndarray):\n",
    "            new_catboost_preds = np.asarray(new_catboost_preds)\n",
    "        all_catboost_preds.append(new_catboost_preds)\n",
    "        del new_catboost_preds, new_catboost\n",
    "        print(f'CatBoost {catboost_ind} predicts on devset.')\n",
    "        print('')\n",
    "    all_catboost_preds = np.stack(all_catboost_preds, axis=0)\n",
    "    all_uncertainty_catboost = ensemble_uncertainties_regression(all_catboost_preds)\n",
    "    catboost_uncertainties = all_uncertainty_catboost['tvar']\n",
    "    all_catboost_preds_mean = all_catboost_preds[:,:,0]\n",
    "    avg_catboost_preds = np.squeeze(np.mean(all_catboost_preds_mean, axis=0))\n",
    "    catboost_errors = (avg_catboost_preds - targets) ** 2\n",
    "    catboost_rejection_mse = calc_uncertainty_regection_curve(\n",
    "        catboost_errors,\n",
    "        catboost_uncertainties\n",
    "    )\n",
    "    catboost_rejection_mse = catboost_rejection_mse[::-1]\n",
    "    if len(retention_mse) != len(catboost_rejection_mse):\n",
    "        err_msg_ = f'Predictions of our models do not correspond ' \\\n",
    "                   f'to predictions of baseline models! ' \\\n",
    "                   f'{len(retention_mse)} != {len(catboost_rejection_mse)}'\n",
    "        raise ValueError(err_msg_)\n",
    "    print(f'R-AUC MSE of CatBoost Ens = {np.mean(catboost_rejection_mse)}.')\n",
    "else:\n",
    "    catboost_rejection_mse = None"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "c43ab567",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 504x504 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "<Figure size 432x288 with 0 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(new_figure_id, figsize=(7, 7))\n",
    "new_figure_id += 1\n",
    "if catboost_rejection_mse is None:\n",
    "    plt.plot([0, 1], [0, retention_mse[-1]], 'k--')\n",
    "    plt.plot(retention_fractions, retention_mse, linewidth=2)\n",
    "else:\n",
    "    last_mse = max(retention_mse[-1], catboost_rejection_mse[-1])\n",
    "    plt.plot(\n",
    "        [0, 1],\n",
    "        [0, last_mse],\n",
    "        'k--'\n",
    "    )\n",
    "    if last_mse > retention_mse[-1]:\n",
    "        plt.plot(\n",
    "            retention_fractions.tolist() + [retention_fractions[-1]],\n",
    "            retention_mse.tolist() + [last_mse],\n",
    "            color='red', linewidth=2, label='Our deep ensemble'\n",
    "        )\n",
    "    else:\n",
    "        plt.plot(\n",
    "            retention_fractions,\n",
    "            retention_mse,\n",
    "            color='red', linewidth=2, label='Our deep ensemble'\n",
    "        )\n",
    "    if last_mse > catboost_rejection_mse[-1]:\n",
    "        plt.plot(\n",
    "            retention_fractions,\n",
    "            catboost_rejection_mse,\n",
    "            color='blue', linewidth=2, label='CatBoost ensemble baseline'\n",
    "        )\n",
    "    else:\n",
    "        plt.plot(\n",
    "            retention_fractions.tolist() + [retention_fractions[-1]],\n",
    "            catboost_rejection_mse.tolist() + [last_mse],\n",
    "            color='blue', linewidth=2, label='CatBoost ensemble baseline'\n",
    "        )\n",
    "plt.ylabel('MSE')\n",
    "plt.xlabel('Retention Fraction')\n",
    "plt.grid()\n",
    "plt.legend(loc='best')\n",
    "plt.show()\n",
    "plt.clf()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "36b96fed",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score at 95% retention: 0.6937173838917867\n"
     ]
    }
   ],
   "source": [
    "thresh = 1.0\n",
    "f_auc, f95, retention_f1 = f_beta_metrics(errors, uncertainties, thresh, beta=1.0)\n",
    "print(f'F1 score at 95% retention: {f95}')\n",
    "retention_fractions = np.linspace(0, 1, len(retention_f1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "477fa3d3",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 504x504 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "<Figure size 432x288 with 0 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(new_figure_id, figsize=(7, 7))\n",
    "new_figure_id += 1\n",
    "plt.plot(retention_fractions, retention_f1)\n",
    "plt.ylabel('F1')\n",
    "plt.xlabel('Retention Fraction')\n",
    "plt.show()\n",
    "plt.clf()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "79377b79",
   "metadata": {},
   "outputs": [],
   "source": [
    "ids = np.arange(1, inputs.shape[0] + 1)\n",
    "preds = np.mean(all_preds[:,:,0], axis=0)\n",
    "df_submission = pd.DataFrame(data={\n",
    "    'ID' : ids,\n",
    "    'PRED' : preds,\n",
    "    'UNCERTAINTY' : uncertainties\n",
    "})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "92b3b77a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>ID</th>\n",
       "      <th>PRED</th>\n",
       "      <th>UNCERTAINTY</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>10.438510</td>\n",
       "      <td>0.616079</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>4.866150</td>\n",
       "      <td>0.528755</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3</td>\n",
       "      <td>26.137645</td>\n",
       "      <td>0.645760</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>4.418600</td>\n",
       "      <td>2.852262</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5</td>\n",
       "      <td>13.519255</td>\n",
       "      <td>1.341938</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   ID       PRED  UNCERTAINTY\n",
       "0   1  10.438510     0.616079\n",
       "1   2   4.866150     0.528755\n",
       "2   3  26.137645     0.645760\n",
       "3   4   4.418600     2.852262\n",
       "4   5  13.519255     1.341938"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_submission.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "id": "16843244",
   "metadata": {},
   "outputs": [],
   "source": [
    "out_file = os.path.join(model_dir, 'df_submission_dev.csv')\n",
    "df_submission.to_csv(out_file, index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "id": "7632803b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>topography_bathymetry</th>\n",
       "      <th>sun_elevation</th>\n",
       "      <th>climate_temperature</th>\n",
       "      <th>climate_pressure</th>\n",
       "      <th>cmc_0_0_0_1000</th>\n",
       "      <th>cmc_0_0_0_2_grad</th>\n",
       "      <th>cmc_0_0_0_2_interpolated</th>\n",
       "      <th>cmc_0_0_0_2_next</th>\n",
       "      <th>cmc_0_0_0_2</th>\n",
       "      <th>cmc_0_0_0_500</th>\n",
       "      <th>...</th>\n",
       "      <th>cmc_0_1_66_0_grad</th>\n",
       "      <th>cmc_0_1_66_0_next</th>\n",
       "      <th>cmc_0_1_67_0_grad</th>\n",
       "      <th>cmc_0_1_67_0_next</th>\n",
       "      <th>cmc_0_1_68_0_grad</th>\n",
       "      <th>cmc_0_1_68_0_next</th>\n",
       "      <th>gfs_2m_dewpoint_grad</th>\n",
       "      <th>gfs_2m_dewpoint_next</th>\n",
       "      <th>gfs_total_clouds_cover_low_grad</th>\n",
       "      <th>gfs_total_clouds_cover_low_next</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>298.0</td>\n",
       "      <td>-16.337754</td>\n",
       "      <td>7.780000</td>\n",
       "      <td>742.524856</td>\n",
       "      <td>277.912061</td>\n",
       "      <td>-2.039062</td>\n",
       "      <td>275.247046</td>\n",
       "      <td>273.207983</td>\n",
       "      <td>275.247046</td>\n",
       "      <td>256.772266</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>-0.558990</td>\n",
       "      <td>-10.608984</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>376.0</td>\n",
       "      <td>30.120293</td>\n",
       "      <td>22.282143</td>\n",
       "      <td>723.331713</td>\n",
       "      <td>292.545093</td>\n",
       "      <td>-0.317285</td>\n",
       "      <td>291.059949</td>\n",
       "      <td>290.954187</td>\n",
       "      <td>291.271472</td>\n",
       "      <td>250.076855</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>2.099976</td>\n",
       "      <td>8.749994</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>83.0</td>\n",
       "      <td>1.232164</td>\n",
       "      <td>14.190000</td>\n",
       "      <td>761.034372</td>\n",
       "      <td>274.791968</td>\n",
       "      <td>-4.542041</td>\n",
       "      <td>273.794141</td>\n",
       "      <td>272.280127</td>\n",
       "      <td>276.822168</td>\n",
       "      <td>257.556323</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.100006</td>\n",
       "      <td>-10.349982</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>5.0</td>\n",
       "      <td>-13.621620</td>\n",
       "      <td>7.946429</td>\n",
       "      <td>764.253037</td>\n",
       "      <td>276.940283</td>\n",
       "      <td>-1.329907</td>\n",
       "      <td>277.582633</td>\n",
       "      <td>277.139331</td>\n",
       "      <td>278.469238</td>\n",
       "      <td>257.171875</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.125946</td>\n",
       "      <td>-2.341925</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>257.0</td>\n",
       "      <td>-20.644158</td>\n",
       "      <td>6.874286</td>\n",
       "      <td>653.700954</td>\n",
       "      <td>285.847852</td>\n",
       "      <td>-0.947852</td>\n",
       "      <td>277.626253</td>\n",
       "      <td>277.310303</td>\n",
       "      <td>278.258154</td>\n",
       "      <td>252.567334</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.002677</td>\n",
       "      <td>0.00358</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.199982</td>\n",
       "      <td>2.350000</td>\n",
       "      <td>26.0</td>\n",
       "      <td>91.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 123 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   topography_bathymetry  sun_elevation  climate_temperature  \\\n",
       "0                  298.0     -16.337754             7.780000   \n",
       "1                  376.0      30.120293            22.282143   \n",
       "2                   83.0       1.232164            14.190000   \n",
       "3                    5.0     -13.621620             7.946429   \n",
       "4                  257.0     -20.644158             6.874286   \n",
       "\n",
       "   climate_pressure  cmc_0_0_0_1000  cmc_0_0_0_2_grad  \\\n",
       "0        742.524856      277.912061         -2.039062   \n",
       "1        723.331713      292.545093         -0.317285   \n",
       "2        761.034372      274.791968         -4.542041   \n",
       "3        764.253037      276.940283         -1.329907   \n",
       "4        653.700954      285.847852         -0.947852   \n",
       "\n",
       "   cmc_0_0_0_2_interpolated  cmc_0_0_0_2_next  cmc_0_0_0_2  cmc_0_0_0_500  \\\n",
       "0                275.247046        273.207983   275.247046     256.772266   \n",
       "1                291.059949        290.954187   291.271472     250.076855   \n",
       "2                273.794141        272.280127   276.822168     257.556323   \n",
       "3                277.582633        277.139331   278.469238     257.171875   \n",
       "4                277.626253        277.310303   278.258154     252.567334   \n",
       "\n",
       "   ...  cmc_0_1_66_0_grad  cmc_0_1_66_0_next  cmc_0_1_67_0_grad  \\\n",
       "0  ...                0.0                0.0           0.000000   \n",
       "1  ...                0.0                0.0           0.000000   \n",
       "2  ...                0.0                0.0           0.000000   \n",
       "3  ...                0.0                0.0           0.000000   \n",
       "4  ...                0.0                0.0           0.002677   \n",
       "\n",
       "   cmc_0_1_67_0_next  cmc_0_1_68_0_grad  cmc_0_1_68_0_next  \\\n",
       "0            0.00000                0.0                0.0   \n",
       "1            0.00000                0.0                0.0   \n",
       "2            0.00000                0.0                0.0   \n",
       "3            0.00000                0.0                0.0   \n",
       "4            0.00358                0.0                0.0   \n",
       "\n",
       "   gfs_2m_dewpoint_grad  gfs_2m_dewpoint_next  \\\n",
       "0             -0.558990            -10.608984   \n",
       "1              2.099976              8.749994   \n",
       "2              0.100006            -10.349982   \n",
       "3              0.125946             -2.341925   \n",
       "4              0.199982              2.350000   \n",
       "\n",
       "   gfs_total_clouds_cover_low_grad  gfs_total_clouds_cover_low_next  \n",
       "0                              0.0                              0.0  \n",
       "1                              0.0                              0.0  \n",
       "2                              0.0                              0.0  \n",
       "3                              0.0                              0.0  \n",
       "4                             26.0                             91.0  \n",
       "\n",
       "[5 rows x 123 columns]"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_eval = pd.read_csv(eval_name)\n",
    "df_eval.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "id": "c719d2b3",
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_inputs = df_eval.to_numpy().astype(np.float64)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "id": "84e3c4ae",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The system predicts each sample from evalset in 0.063 msecs.\n"
     ]
    }
   ],
   "source": [
    "start_inference_time = time.time()\n",
    "all_preds = predict_by_ensemble(\n",
    "    input_data=eval_inputs,\n",
    "    preprocessing=common_preprocessor,\n",
    "    postprocessing=postprocessing_scalers,\n",
    "    ensemble=deep_ensemble,\n",
    "    minibatch=BATCH_SIZE\n",
    ")\n",
    "end_inference_time = time.time()\n",
    "inference_duration = end_inference_time - start_inference_time\n",
    "inference_duration *= 1000.0\n",
    "print('The system predicts each sample from evalset in ' \\\n",
    "      '{0:.3f} msecs.'.format(inference_duration / eval_inputs.shape[0]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "b6bd2690",
   "metadata": {},
   "outputs": [],
   "source": [
    "all_uncertainty = ensemble_uncertainties_regression(all_preds)\n",
    "uncertainties = all_uncertainty['tvar']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "f999a9a2",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>ID</th>\n",
       "      <th>PRED</th>\n",
       "      <th>UNCERTAINTY</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>2.899178</td>\n",
       "      <td>0.719325</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>17.571943</td>\n",
       "      <td>1.277610</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3</td>\n",
       "      <td>1.696646</td>\n",
       "      <td>0.905027</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>2.293385</td>\n",
       "      <td>2.332876</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5</td>\n",
       "      <td>9.411125</td>\n",
       "      <td>3.744693</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   ID       PRED  UNCERTAINTY\n",
       "0   1   2.899178     0.719325\n",
       "1   2  17.571943     1.277610\n",
       "2   3   1.696646     0.905027\n",
       "3   4   2.293385     2.332876\n",
       "4   5   9.411125     3.744693"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ids = np.arange(1, len(df_eval) + 1)\n",
    "preds = np.mean(all_preds[:,:,0], axis=0)\n",
    "df_submission = pd.DataFrame(data={\n",
    "    'ID' : ids,\n",
    "    'PRED' : preds,\n",
    "    'UNCERTAINTY' : uncertainties\n",
    "})\n",
    "df_submission.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "id": "20748bdc",
   "metadata": {},
   "outputs": [],
   "source": [
    "out_file = os.path.join(model_dir, 'df_submission.csv')\n",
    "df_submission.to_csv(out_file, index=False)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}