AE_models.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h4> Notebook that contains the training and evaluation of models that leverage the autoencoder representation and classical machine learning models without the AE representation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 105/105 [00:00<00:00, 1226.98it/s]\n"
     ]
    },
    {
     "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>sigId</th>\n",
       "      <th>ecg_lead_1</th>\n",
       "      <th>ecg_lead_2</th>\n",
       "      <th>peaks</th>\n",
       "      <th>frequencies</th>\n",
       "      <th>labels</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>S001</td>\n",
       "      <td>[0.0, 0.04, 0.03, 0.0, 0.03, 0.09, 0.18, 0.14,...</td>\n",
       "      <td>[0.08, 0.07, 0.1, 0.06, 0.06, 0.03, 0.1, 0.21,...</td>\n",
       "      <td>[29, 110, 191, 272, 353, 433, 514, 595, 676, 7...</td>\n",
       "      <td>128</td>\n",
       "      <td>[N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>S002</td>\n",
       "      <td>[-0.035, -0.045, -0.025, -0.035, -0.045, -0.05...</td>\n",
       "      <td>[-0.095, -0.105, -0.095, -0.095, -0.115, -0.09...</td>\n",
       "      <td>[48, 153, 243, 352, 440, 547, 636, 742, 831, 9...</td>\n",
       "      <td>128</td>\n",
       "      <td>[N, N, N, N, S, N, S, N, S, N, S, N, S, N, S, ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>S003</td>\n",
       "      <td>[-0.56, -0.56, -0.55, -0.47, -0.53, -0.47, -0....</td>\n",
       "      <td>[0.43, 0.56, 0.6, 0.41, 0.54, 0.48, 0.56, 0.46...</td>\n",
       "      <td>[91, 209, 326, 394, 537, 653, 745, 872, 984, 1...</td>\n",
       "      <td>128</td>\n",
       "      <td>[N, N, N, S, N, N, S, N, N, S, N, N, N, N, N, ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>S004</td>\n",
       "      <td>[-0.46, -0.49, -0.52, -0.58, -0.62, -0.69, -0....</td>\n",
       "      <td>[0.56, 0.61, 0.66, 0.66, 0.63, 0.66, 0.59, 0.5...</td>\n",
       "      <td>[98, 223, 349, 474, 599, 726, 853, 980, 1116, ...</td>\n",
       "      <td>128</td>\n",
       "      <td>[N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>S005</td>\n",
       "      <td>[-0.27, -0.17, -0.13, -0.23, -0.18, -0.23, -0....</td>\n",
       "      <td>[-0.02, -0.04, -0.01, -0.01, -0.02, -0.06, 0.0...</td>\n",
       "      <td>[27, 127, 225, 324, 423, 523, 623, 722, 822, 9...</td>\n",
       "      <td>128</td>\n",
       "      <td>[N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, ...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  sigId                                         ecg_lead_1  \\\n",
       "0  S001  [0.0, 0.04, 0.03, 0.0, 0.03, 0.09, 0.18, 0.14,...   \n",
       "1  S002  [-0.035, -0.045, -0.025, -0.035, -0.045, -0.05...   \n",
       "2  S003  [-0.56, -0.56, -0.55, -0.47, -0.53, -0.47, -0....   \n",
       "3  S004  [-0.46, -0.49, -0.52, -0.58, -0.62, -0.69, -0....   \n",
       "4  S005  [-0.27, -0.17, -0.13, -0.23, -0.18, -0.23, -0....   \n",
       "\n",
       "                                          ecg_lead_2  \\\n",
       "0  [0.08, 0.07, 0.1, 0.06, 0.06, 0.03, 0.1, 0.21,...   \n",
       "1  [-0.095, -0.105, -0.095, -0.095, -0.115, -0.09...   \n",
       "2  [0.43, 0.56, 0.6, 0.41, 0.54, 0.48, 0.56, 0.46...   \n",
       "3  [0.56, 0.61, 0.66, 0.66, 0.63, 0.66, 0.59, 0.5...   \n",
       "4  [-0.02, -0.04, -0.01, -0.01, -0.02, -0.06, 0.0...   \n",
       "\n",
       "                                               peaks frequencies  \\\n",
       "0  [29, 110, 191, 272, 353, 433, 514, 595, 676, 7...         128   \n",
       "1  [48, 153, 243, 352, 440, 547, 636, 742, 831, 9...         128   \n",
       "2  [91, 209, 326, 394, 537, 653, 745, 872, 984, 1...         128   \n",
       "3  [98, 223, 349, 474, 599, 726, 853, 980, 1116, ...         128   \n",
       "4  [27, 127, 225, 324, 423, 523, 623, 722, 822, 9...         128   \n",
       "\n",
       "                                              labels  \n",
       "0  [N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, ...  \n",
       "1  [N, N, N, N, S, N, S, N, S, N, S, N, S, N, S, ...  \n",
       "2  [N, N, N, S, N, N, S, N, N, S, N, N, N, N, N, ...  \n",
       "3  [N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, ...  \n",
       "4  [N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, ...  "
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import os \n",
    "import zipfile \n",
    "from scipy.io import loadmat \n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt \n",
    "import numpy as np\n",
    "import io\n",
    "from PIL import Image\n",
    "import tensorflow as tf \n",
    "from tensorflow import keras\n",
    "import sys\n",
    "from tqdm import tqdm\n",
    "import random\n",
    "from sklearn.model_selection import train_test_split\n",
    "import pywt \n",
    "\n",
    "\n",
    "SEED = 123\n",
    "os.environ['PYTHONHASHSEED'] = str(SEED)\n",
    "random.seed(SEED)\n",
    "np.random.seed(SEED)\n",
    "tf.random.set_seed(SEED)\n",
    "\n",
    "#\n",
    "# CHANGE ACCORDING YOUR SETUP \n",
    "dataset_current_folder = \"training_set.zip\" # where the zip is\n",
    "dataset_folder = \"C://Users//simon//Desktop//AppliedAI-project\" # where I want the dataset - avoid the current folder as Git doesn't allow huge uploads\n",
    "#\n",
    "#\n",
    "\n",
    "\n",
    "with zipfile.ZipFile(dataset_current_folder, 'r') as zip: # extract the zip file into the desired folder \n",
    "    zip.extractall(dataset_folder)\n",
    "\n",
    "def load_data(sample_prefix, input_dir):    # everything is returned as a numpy array which is easier to manipulate\n",
    "    label_filepath = os.path.join(input_dir, sample_prefix + '_ann.mat')\n",
    "    peak_filepath = os.path.join(input_dir, sample_prefix + '_rpk.mat')\n",
    "    signal_filepath = os.path.join(input_dir, sample_prefix + '.mat')\n",
    "    if os.path.isfile(label_filepath):\n",
    "        mat_file = loadmat(label_filepath)\n",
    "        label = np.asarray(mat_file['labels'])\n",
    "    if os.path.isfile(peak_filepath):\n",
    "        mat_file = loadmat(peak_filepath)\n",
    "        peaks = np.array(mat_file['rpeaks'],dtype=np.int64)\n",
    "    if os.path.isfile(signal_filepath):\n",
    "        mat_file = loadmat(signal_filepath)\n",
    "        signal = np.asarray(mat_file['ecg'] )\n",
    "\n",
    "    return label, peaks, signal\n",
    "\n",
    "\n",
    "labels = list()             # labels \n",
    "ids = list()                # Id of samples \n",
    "rpeaks = list()             # detected peaks of the signal \n",
    "ecg_signals = list()        # .mat ecg signal \n",
    "frequencies = list()        # sample frequency of the ecg signal \n",
    "\n",
    "\n",
    "for f in os.listdir(dataset_folder):\n",
    "  if f.lower().endswith('.mat'):\n",
    "    id = f[:4]\n",
    "    if id not in ids:\n",
    "      ids.append(id)\n",
    "      sample_prefix = f[:8]\n",
    "      label, peak, signal = load_data(sample_prefix, dataset_folder)\n",
    "      labels.append(label)\n",
    "      rpeaks.append(peak)\n",
    "      ecg_signals.append(signal)\n",
    "      frequencies.append(int(sample_prefix[5:]))\n",
    "cols = [\"sigId\",\"ecg_lead_1\",\"ecg_lead_2\",\"peaks\",\"frequencies\",\"labels\"]\n",
    "\n",
    "# ecg signals is 105 rows [,,,,,]\n",
    "\n",
    "first_lead_signals = []\n",
    "second_lead_signals = []\n",
    "\n",
    "for signal in ecg_signals:\n",
    "    first_lead_signals.append(signal[:,0].tolist())    # converting the array to list as list of array is deprecated \n",
    "    second_lead_signals.append(signal[:,1].tolist())\n",
    "\n",
    "df = pd.DataFrame(data =[ids,first_lead_signals,second_lead_signals,rpeaks,frequencies,labels]).T\n",
    "df.columns = cols\n",
    "\n",
    "# transform peaks \n",
    "for id in tqdm(df.index.tolist()):\n",
    "    peaks = df.iloc[id]['peaks']\n",
    "    p_list = list()\n",
    "    for p in peaks:\n",
    "        p_list.append(p[0])\n",
    "    df.iloc[id]['peaks'] = np.asarray(p_list).astype(np.int64)\n",
    "\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2>Downsampling "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 40/40 [00:04<00:00,  8.77it/s]\n"
     ]
    }
   ],
   "source": [
    "# because of te curse of dimensionality --> undersampling \n",
    "\n",
    "from scipy.signal import resample\n",
    "\n",
    "ids_128 = df[df['frequencies'] == 128].index.tolist() \n",
    "ids_250 = df[df['frequencies'] == 250].index.tolist() \n",
    "sampled_len =  len(df.iloc[ids_250[0]]['ecg_lead_1'])\n",
    "resampled_len = len(df.iloc[ids_128[0]]['ecg_lead_1'])\n",
    "\n",
    "for id in tqdm(ids_250):\n",
    "    row = df.iloc[id]\n",
    "    # first lead\n",
    "    signal = np.asarray(row['ecg_lead_1']).astype(np.float32)\n",
    "    resampled_1 = resample(signal,resampled_len)\n",
    "    \n",
    "    # second lead\n",
    "    \n",
    "    signal = np.asarray(row['ecg_lead_2']).astype(np.float32)\n",
    "    resampled_2 = resample(signal,resampled_len)\n",
    "    \n",
    "    df.iloc[id]['ecg_lead_1'] = resampled_1.tolist()\n",
    "    df.iloc[id]['ecg_lead_2'] = resampled_2.tolist()\n",
    "    \n",
    "    for i,p in enumerate(list(df.iloc[id]['peaks'])):\n",
    "        df.iloc[id]['peaks'][i] = int(resampled_len * p/sampled_len)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2>Bandpass filtering"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 105/105 [00:01<00:00, 54.39it/s]\n"
     ]
    }
   ],
   "source": [
    "#Preprocessing used in practical lessons\n",
    "from scipy.signal import resample, butter, lfilter, iirnotch\n",
    "\n",
    "def notch_filter(cutoff, fs, q=30):\n",
    "    nyq = 0.5*fs\n",
    "    freq = cutoff/nyq\n",
    "    b, a = iirnotch(freq, q)\n",
    "    return b, a\n",
    "\n",
    "\n",
    "def bandpass_filter(data, filter_order=6, lowcut = .8, highcut = 45, signal_freq=128):\n",
    "    \"\"\"\n",
    "    Method responsible for creating and applying Butterworth filter.\n",
    "    :param deque data: raw data\n",
    "    :param float lowcut: filter lowcut frequency value\n",
    "    :param float highcut: filter highcut frequency value\n",
    "    :param int signal_freq: signal frequency in samples per second (Hz)\n",
    "    :param int filter_order: filter order\n",
    "    :return array: filtered data\n",
    "    \"\"\"\n",
    "    powerline = 60\n",
    "    nyquist_freq = 0.5 * signal_freq\n",
    "    low = lowcut / nyquist_freq\n",
    "    high = highcut / nyquist_freq\n",
    "    b, a = butter(filter_order, [low, high], btype=\"band\")\n",
    "    y = lfilter(b, a, data,zi = None)\n",
    "    #b,a = notch_filter(powerline,signal_freq) not applied \n",
    "    #y = lfilter(b,a,y)\n",
    "    return y\n",
    "\n",
    "\n",
    "\n",
    "filtered_df = df.copy(deep= True)\n",
    "\n",
    "ids = filtered_df.index.tolist()\n",
    "\n",
    "for id_ in tqdm(ids):\n",
    "    \n",
    "    row = filtered_df.iloc[id_]\n",
    "    first = row['ecg_lead_1']\n",
    "    second = row['ecg_lead_2']\n",
    "    filtered_first = np.asarray(bandpass_filter(first))\n",
    "    filtered_second =  np.asarray(bandpass_filter(second))\n",
    "\n",
    "    filtered_df.iloc[id_]['ecg_lead_1'] = pd.Series(filtered_first)\n",
    "    filtered_df.iloc[id_]['ecg_lead_2'] = pd.Series(filtered_second)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2>Scaling to [0:1] in 10 seconds time window "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 105/105 [00:14<00:00,  7.48it/s]\n"
     ]
    }
   ],
   "source": [
    "import heartpy as hp\n",
    "\n",
    "fs = 128\n",
    "time_window = 10\n",
    "samples = fs * time_window # samples in the 10 sec time window \n",
    "\n",
    "for id in tqdm(filtered_df.index.tolist()):\n",
    "    row = filtered_df.loc[id]\n",
    "    first_lead = row['ecg_lead_1']\n",
    "    second_lead = row['ecg_lead_2']\n",
    "    l_ = len(first_lead)\n",
    "    scaled_1 = list()\n",
    "    scaled_2 = list()\n",
    "    \n",
    "    for i in range(int(l_/samples)):\n",
    "        to_scale_1 = first_lead[i*samples:i*samples+samples] \n",
    "        to_scale_2 = second_lead[i*samples:i*samples+samples]\n",
    "        \n",
    "        scaled_1 += list(hp.scale_data(to_scale_1,lower=0,upper=1))\n",
    "        scaled_2 += list(hp.scale_data(to_scale_2,lower=0,upper=1))\n",
    "   \n",
    "    filtered_df.loc[id]['ecg_lead_1'] = scaled_1\n",
    "    filtered_df.loc[id]['ecg_lead_2'] = scaled_2"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2>Outliers detection and removal using zscore</h2>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "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>mean_lead1</th>\n",
       "      <th>median_lead1</th>\n",
       "      <th>std_lead1</th>\n",
       "      <th>var_lead1</th>\n",
       "      <th>skew_lead1</th>\n",
       "      <th>kurt_lead1</th>\n",
       "      <th>mean_lead2</th>\n",
       "      <th>median_lead2</th>\n",
       "      <th>std_lead2</th>\n",
       "      <th>var_lead2</th>\n",
       "      <th>skew_lead2</th>\n",
       "      <th>kurt_lead2</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.476086</td>\n",
       "      <td>0.474618</td>\n",
       "      <td>0.128357</td>\n",
       "      <td>0.016476</td>\n",
       "      <td>0.065871</td>\n",
       "      <td>3.853435</td>\n",
       "      <td>0.600828</td>\n",
       "      <td>0.613177</td>\n",
       "      <td>0.119664</td>\n",
       "      <td>0.014319</td>\n",
       "      <td>-1.328497</td>\n",
       "      <td>6.504245</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.309628</td>\n",
       "      <td>0.315643</td>\n",
       "      <td>0.118704</td>\n",
       "      <td>0.014091</td>\n",
       "      <td>2.242173</td>\n",
       "      <td>11.961672</td>\n",
       "      <td>0.662428</td>\n",
       "      <td>0.672944</td>\n",
       "      <td>0.144363</td>\n",
       "      <td>0.020841</td>\n",
       "      <td>-1.038713</td>\n",
       "      <td>3.939288</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.462381</td>\n",
       "      <td>0.466913</td>\n",
       "      <td>0.122431</td>\n",
       "      <td>0.014989</td>\n",
       "      <td>-0.007802</td>\n",
       "      <td>2.618824</td>\n",
       "      <td>0.570732</td>\n",
       "      <td>0.586303</td>\n",
       "      <td>0.134329</td>\n",
       "      <td>0.018044</td>\n",
       "      <td>-0.788964</td>\n",
       "      <td>3.169306</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.399816</td>\n",
       "      <td>0.366260</td>\n",
       "      <td>0.149092</td>\n",
       "      <td>0.022228</td>\n",
       "      <td>0.955791</td>\n",
       "      <td>1.976279</td>\n",
       "      <td>0.658729</td>\n",
       "      <td>0.681179</td>\n",
       "      <td>0.162172</td>\n",
       "      <td>0.026300</td>\n",
       "      <td>-1.171663</td>\n",
       "      <td>1.827310</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.478230</td>\n",
       "      <td>0.488412</td>\n",
       "      <td>0.131123</td>\n",
       "      <td>0.017193</td>\n",
       "      <td>-0.046548</td>\n",
       "      <td>4.308083</td>\n",
       "      <td>0.555436</td>\n",
       "      <td>0.561574</td>\n",
       "      <td>0.148968</td>\n",
       "      <td>0.022191</td>\n",
       "      <td>-0.156231</td>\n",
       "      <td>1.414820</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>100</th>\n",
       "      <td>0.394100</td>\n",
       "      <td>0.385903</td>\n",
       "      <td>0.132414</td>\n",
       "      <td>0.017534</td>\n",
       "      <td>0.582383</td>\n",
       "      <td>1.942821</td>\n",
       "      <td>0.426588</td>\n",
       "      <td>0.419286</td>\n",
       "      <td>0.116018</td>\n",
       "      <td>0.013460</td>\n",
       "      <td>0.446691</td>\n",
       "      <td>2.379165</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>101</th>\n",
       "      <td>0.437926</td>\n",
       "      <td>0.430825</td>\n",
       "      <td>0.127339</td>\n",
       "      <td>0.016215</td>\n",
       "      <td>0.479884</td>\n",
       "      <td>3.274700</td>\n",
       "      <td>0.444411</td>\n",
       "      <td>0.452348</td>\n",
       "      <td>0.126760</td>\n",
       "      <td>0.016068</td>\n",
       "      <td>0.327382</td>\n",
       "      <td>3.613294</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>102</th>\n",
       "      <td>0.345843</td>\n",
       "      <td>0.332378</td>\n",
       "      <td>0.115116</td>\n",
       "      <td>0.013252</td>\n",
       "      <td>1.653328</td>\n",
       "      <td>6.466284</td>\n",
       "      <td>0.403117</td>\n",
       "      <td>0.404505</td>\n",
       "      <td>0.114159</td>\n",
       "      <td>0.013032</td>\n",
       "      <td>0.996624</td>\n",
       "      <td>6.575469</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>103</th>\n",
       "      <td>0.514950</td>\n",
       "      <td>0.522097</td>\n",
       "      <td>0.098893</td>\n",
       "      <td>0.009780</td>\n",
       "      <td>-0.152302</td>\n",
       "      <td>9.197804</td>\n",
       "      <td>0.461116</td>\n",
       "      <td>0.445006</td>\n",
       "      <td>0.129433</td>\n",
       "      <td>0.016753</td>\n",
       "      <td>0.508997</td>\n",
       "      <td>1.575390</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>104</th>\n",
       "      <td>0.537678</td>\n",
       "      <td>0.539089</td>\n",
       "      <td>0.106207</td>\n",
       "      <td>0.011280</td>\n",
       "      <td>-0.579932</td>\n",
       "      <td>6.520010</td>\n",
       "      <td>0.659409</td>\n",
       "      <td>0.655174</td>\n",
       "      <td>0.129539</td>\n",
       "      <td>0.016781</td>\n",
       "      <td>-1.179600</td>\n",
       "      <td>6.316306</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>105 rows × 12 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     mean_lead1  median_lead1  std_lead1  var_lead1  skew_lead1  kurt_lead1  \\\n",
       "0      0.476086      0.474618   0.128357   0.016476    0.065871    3.853435   \n",
       "1      0.309628      0.315643   0.118704   0.014091    2.242173   11.961672   \n",
       "2      0.462381      0.466913   0.122431   0.014989   -0.007802    2.618824   \n",
       "3      0.399816      0.366260   0.149092   0.022228    0.955791    1.976279   \n",
       "4      0.478230      0.488412   0.131123   0.017193   -0.046548    4.308083   \n",
       "..          ...           ...        ...        ...         ...         ...   \n",
       "100    0.394100      0.385903   0.132414   0.017534    0.582383    1.942821   \n",
       "101    0.437926      0.430825   0.127339   0.016215    0.479884    3.274700   \n",
       "102    0.345843      0.332378   0.115116   0.013252    1.653328    6.466284   \n",
       "103    0.514950      0.522097   0.098893   0.009780   -0.152302    9.197804   \n",
       "104    0.537678      0.539089   0.106207   0.011280   -0.579932    6.520010   \n",
       "\n",
       "     mean_lead2  median_lead2  std_lead2  var_lead2  skew_lead2  kurt_lead2  \n",
       "0      0.600828      0.613177   0.119664   0.014319   -1.328497    6.504245  \n",
       "1      0.662428      0.672944   0.144363   0.020841   -1.038713    3.939288  \n",
       "2      0.570732      0.586303   0.134329   0.018044   -0.788964    3.169306  \n",
       "3      0.658729      0.681179   0.162172   0.026300   -1.171663    1.827310  \n",
       "4      0.555436      0.561574   0.148968   0.022191   -0.156231    1.414820  \n",
       "..          ...           ...        ...        ...         ...         ...  \n",
       "100    0.426588      0.419286   0.116018   0.013460    0.446691    2.379165  \n",
       "101    0.444411      0.452348   0.126760   0.016068    0.327382    3.613294  \n",
       "102    0.403117      0.404505   0.114159   0.013032    0.996624    6.575469  \n",
       "103    0.461116      0.445006   0.129433   0.016753    0.508997    1.575390  \n",
       "104    0.659409      0.655174   0.129539   0.016781   -1.179600    6.316306  \n",
       "\n",
       "[105 rows x 12 columns]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from scipy import stats\n",
    "\n",
    "# Function for the computation of rhythm data \n",
    "def get_2ECG_rhythmfeatures(data):\n",
    "  \n",
    "  #   mean\n",
    "  mean_lead = np.mean(data)\n",
    "\n",
    "  #   median\n",
    "  median_lead = np.median(data)\n",
    "\n",
    "  #   standard deviation\n",
    "  std_lead = np.std(data)\n",
    "\n",
    "  #   variance\n",
    "  var_lead = stats.tvar(data)\n",
    "\n",
    "  #   Skewness\n",
    "  skew_lead = stats.skew(data)\n",
    "\n",
    "  #   Kurtosis\n",
    "  kurt_lead = stats.kurtosis(data)\n",
    "\n",
    "  features = np.hstack([mean_lead,median_lead,std_lead,var_lead,skew_lead,kurt_lead])\n",
    "\n",
    "  return features\n",
    "#-------------------------------------------------------------------------------\n",
    "\n",
    "#Training\n",
    "\n",
    "features_rhythm_lead1 = list()\n",
    "features_rhythm_lead2 = list()\n",
    "\n",
    "\n",
    "for i in range(filtered_df.shape[0]):\n",
    "  tmp_lead1 = get_2ECG_rhythmfeatures(filtered_df['ecg_lead_1'][i])\n",
    "  features_rhythm_lead1.append(tmp_lead1)\n",
    "  tmp_lead2 = get_2ECG_rhythmfeatures(filtered_df['ecg_lead_2'][i])\n",
    "  features_rhythm_lead2.append(tmp_lead2)\n",
    "\n",
    "features_rhythm_lead1 = pd.DataFrame(features_rhythm_lead1, columns = ['mean_lead1', 'median_lead1', 'std_lead1', 'var_lead1', 'skew_lead1','kurt_lead1'])\n",
    "features_rhythm_lead2 = pd.DataFrame(features_rhythm_lead2, columns = ['mean_lead2', 'median_lead2', 'std_lead2', 'var_lead2', 'skew_lead2','kurt_lead2'])\n",
    "\n",
    "features_rhythm = pd.concat([features_rhythm_lead1, features_rhythm_lead2], axis=1)\n",
    "\n",
    "features = pd.DataFrame(features_rhythm, columns = ['mean_lead1', 'median_lead1', 'std_lead1', 'var_lead1', 'skew_lead1','kurt_lead1',\n",
    "                                                    'mean_lead2', 'median_lead2', 'std_lead2', 'var_lead2', 'skew_lead2','kurt_lead2'])\n",
    "display(features) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_outliersAnalysis(abs_zscores):\n",
    "  # number of outliers per subject \n",
    "  bool_zscores = (abs_zscores > 3).sum(axis=1)\n",
    "  bool_zscores.hist()\n",
    "  plt.title(\"Number of outliers per subject\")\n",
    "  plt.show()\n",
    "\n",
    "  # number of outliers per feature \n",
    "  bool_zscores = (abs_zscores > 3).sum(axis=0)\n",
    "  #bool_zscores.hist(bins = len(feature_list))\n",
    "  fig = plt.figure()\n",
    "  ax = fig.add_axes([0,0,1,1])\n",
    "\n",
    "  ax.bar(feature_list,bool_zscores)\n",
    "  plt.title(\"Number of outliers per feature\")\n",
    "  plt.xticks(np.arange(len(feature_list)), feature_list,rotation=90)\n",
    "  plt.show()\n",
    "  return"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train subjects before outlier removal: (105, 12)\n",
      "Train subjects after outlier removal: (99, 12)\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "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>mean_lead1</th>\n",
       "      <th>median_lead1</th>\n",
       "      <th>std_lead1</th>\n",
       "      <th>var_lead1</th>\n",
       "      <th>skew_lead1</th>\n",
       "      <th>kurt_lead1</th>\n",
       "      <th>mean_lead2</th>\n",
       "      <th>median_lead2</th>\n",
       "      <th>std_lead2</th>\n",
       "      <th>var_lead2</th>\n",
       "      <th>skew_lead2</th>\n",
       "      <th>kurt_lead2</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.476086</td>\n",
       "      <td>0.474618</td>\n",
       "      <td>0.128357</td>\n",
       "      <td>0.016476</td>\n",
       "      <td>0.065871</td>\n",
       "      <td>3.853435</td>\n",
       "      <td>0.600828</td>\n",
       "      <td>0.613177</td>\n",
       "      <td>0.119664</td>\n",
       "      <td>0.014319</td>\n",
       "      <td>-1.328497</td>\n",
       "      <td>6.504245</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.309628</td>\n",
       "      <td>0.315643</td>\n",
       "      <td>0.118704</td>\n",
       "      <td>0.014091</td>\n",
       "      <td>2.242173</td>\n",
       "      <td>11.961672</td>\n",
       "      <td>0.662428</td>\n",
       "      <td>0.672944</td>\n",
       "      <td>0.144363</td>\n",
       "      <td>0.020841</td>\n",
       "      <td>-1.038713</td>\n",
       "      <td>3.939288</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.462381</td>\n",
       "      <td>0.466913</td>\n",
       "      <td>0.122431</td>\n",
       "      <td>0.014989</td>\n",
       "      <td>-0.007802</td>\n",
       "      <td>2.618824</td>\n",
       "      <td>0.570732</td>\n",
       "      <td>0.586303</td>\n",
       "      <td>0.134329</td>\n",
       "      <td>0.018044</td>\n",
       "      <td>-0.788964</td>\n",
       "      <td>3.169306</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.399816</td>\n",
       "      <td>0.366260</td>\n",
       "      <td>0.149092</td>\n",
       "      <td>0.022228</td>\n",
       "      <td>0.955791</td>\n",
       "      <td>1.976279</td>\n",
       "      <td>0.658729</td>\n",
       "      <td>0.681179</td>\n",
       "      <td>0.162172</td>\n",
       "      <td>0.026300</td>\n",
       "      <td>-1.171663</td>\n",
       "      <td>1.827310</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.478230</td>\n",
       "      <td>0.488412</td>\n",
       "      <td>0.131123</td>\n",
       "      <td>0.017193</td>\n",
       "      <td>-0.046548</td>\n",
       "      <td>4.308083</td>\n",
       "      <td>0.555436</td>\n",
       "      <td>0.561574</td>\n",
       "      <td>0.148968</td>\n",
       "      <td>0.022191</td>\n",
       "      <td>-0.156231</td>\n",
       "      <td>1.414820</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>100</th>\n",
       "      <td>0.394100</td>\n",
       "      <td>0.385903</td>\n",
       "      <td>0.132414</td>\n",
       "      <td>0.017534</td>\n",
       "      <td>0.582383</td>\n",
       "      <td>1.942821</td>\n",
       "      <td>0.426588</td>\n",
       "      <td>0.419286</td>\n",
       "      <td>0.116018</td>\n",
       "      <td>0.013460</td>\n",
       "      <td>0.446691</td>\n",
       "      <td>2.379165</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>101</th>\n",
       "      <td>0.437926</td>\n",
       "      <td>0.430825</td>\n",
       "      <td>0.127339</td>\n",
       "      <td>0.016215</td>\n",
       "      <td>0.479884</td>\n",
       "      <td>3.274700</td>\n",
       "      <td>0.444411</td>\n",
       "      <td>0.452348</td>\n",
       "      <td>0.126760</td>\n",
       "      <td>0.016068</td>\n",
       "      <td>0.327382</td>\n",
       "      <td>3.613294</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>102</th>\n",
       "      <td>0.345843</td>\n",
       "      <td>0.332378</td>\n",
       "      <td>0.115116</td>\n",
       "      <td>0.013252</td>\n",
       "      <td>1.653328</td>\n",
       "      <td>6.466284</td>\n",
       "      <td>0.403117</td>\n",
       "      <td>0.404505</td>\n",
       "      <td>0.114159</td>\n",
       "      <td>0.013032</td>\n",
       "      <td>0.996624</td>\n",
       "      <td>6.575469</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>103</th>\n",
       "      <td>0.514950</td>\n",
       "      <td>0.522097</td>\n",
       "      <td>0.098893</td>\n",
       "      <td>0.009780</td>\n",
       "      <td>-0.152302</td>\n",
       "      <td>9.197804</td>\n",
       "      <td>0.461116</td>\n",
       "      <td>0.445006</td>\n",
       "      <td>0.129433</td>\n",
       "      <td>0.016753</td>\n",
       "      <td>0.508997</td>\n",
       "      <td>1.575390</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>104</th>\n",
       "      <td>0.537678</td>\n",
       "      <td>0.539089</td>\n",
       "      <td>0.106207</td>\n",
       "      <td>0.011280</td>\n",
       "      <td>-0.579932</td>\n",
       "      <td>6.520010</td>\n",
       "      <td>0.659409</td>\n",
       "      <td>0.655174</td>\n",
       "      <td>0.129539</td>\n",
       "      <td>0.016781</td>\n",
       "      <td>-1.179600</td>\n",
       "      <td>6.316306</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>99 rows × 12 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     mean_lead1  median_lead1  std_lead1  var_lead1  skew_lead1  kurt_lead1  \\\n",
       "0      0.476086      0.474618   0.128357   0.016476    0.065871    3.853435   \n",
       "1      0.309628      0.315643   0.118704   0.014091    2.242173   11.961672   \n",
       "2      0.462381      0.466913   0.122431   0.014989   -0.007802    2.618824   \n",
       "3      0.399816      0.366260   0.149092   0.022228    0.955791    1.976279   \n",
       "4      0.478230      0.488412   0.131123   0.017193   -0.046548    4.308083   \n",
       "..          ...           ...        ...        ...         ...         ...   \n",
       "100    0.394100      0.385903   0.132414   0.017534    0.582383    1.942821   \n",
       "101    0.437926      0.430825   0.127339   0.016215    0.479884    3.274700   \n",
       "102    0.345843      0.332378   0.115116   0.013252    1.653328    6.466284   \n",
       "103    0.514950      0.522097   0.098893   0.009780   -0.152302    9.197804   \n",
       "104    0.537678      0.539089   0.106207   0.011280   -0.579932    6.520010   \n",
       "\n",
       "     mean_lead2  median_lead2  std_lead2  var_lead2  skew_lead2  kurt_lead2  \n",
       "0      0.600828      0.613177   0.119664   0.014319   -1.328497    6.504245  \n",
       "1      0.662428      0.672944   0.144363   0.020841   -1.038713    3.939288  \n",
       "2      0.570732      0.586303   0.134329   0.018044   -0.788964    3.169306  \n",
       "3      0.658729      0.681179   0.162172   0.026300   -1.171663    1.827310  \n",
       "4      0.555436      0.561574   0.148968   0.022191   -0.156231    1.414820  \n",
       "..          ...           ...        ...        ...         ...         ...  \n",
       "100    0.426588      0.419286   0.116018   0.013460    0.446691    2.379165  \n",
       "101    0.444411      0.452348   0.126760   0.016068    0.327382    3.613294  \n",
       "102    0.403117      0.404505   0.114159   0.013032    0.996624    6.575469  \n",
       "103    0.461116      0.445006   0.129433   0.016753    0.508997    1.575390  \n",
       "104    0.659409      0.655174   0.129539   0.016781   -1.179600    6.316306  \n",
       "\n",
       "[99 rows x 12 columns]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from scipy.stats import zscore\n",
    "import numpy as np\n",
    "\n",
    "feature_list = features.columns.tolist()\n",
    "print('Train subjects before outlier removal: ' + str(features.shape))\n",
    "\n",
    "# apply zscore\n",
    "zscore_features = features.apply(zscore, axis=0)\n",
    "\n",
    "# select all those values with zscore < -+ 3 and discard outliers \n",
    "abs_zscores = np.abs(zscore_features)\n",
    "filtered_entries = (abs_zscores < 3).all(axis=1)\n",
    "\n",
    "features_or = features[filtered_entries]\n",
    "\n",
    "print('Train subjects after outlier removal: ' + str(features_or.shape))\n",
    "\n",
    "\n",
    "plot_outliersAnalysis(abs_zscores)\n",
    "display(features_or)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2> Remove outliers "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "features_or.index\n",
    "filtered_df = filtered_df.loc[features_or.index]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2> RR distances analysis"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 0/99 [00:00<?, ?it/s]C:\\Users\\simon\\AppData\\Local\\Temp/ipykernel_11076/1816920695.py:24: RuntimeWarning: Mean of empty slice.\n",
      "  RR_dict[sig_Id]['AVG_S_distance'] = RR_dict[sig_Id]['S_distances'].mean()\n",
      "100%|██████████| 99/99 [00:00<00:00, 374.71it/s]\n"
     ]
    },
    {
     "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>RR_distances</th>\n",
       "      <th>S_distances</th>\n",
       "      <th>Avg_RR_distance</th>\n",
       "      <th>AVG_S_distance</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>S114</th>\n",
       "      <td>[117, 116, 115, 112, 113, 114, 114, 112, 116, ...</td>\n",
       "      <td>[]</td>\n",
       "      <td>114.914745</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>S115</th>\n",
       "      <td>[113, 118, 80, 151, 118, 123, 76, 152, 119, 11...</td>\n",
       "      <td>[73]</td>\n",
       "      <td>89.041006</td>\n",
       "      <td>73.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>S116</th>\n",
       "      <td>[143, 145, 141, 146, 143, 142, 144, 142, 143, ...</td>\n",
       "      <td>[75]</td>\n",
       "      <td>122.611082</td>\n",
       "      <td>75.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>S117</th>\n",
       "      <td>[142, 143, 147, 148, 145, 139, 148, 139, 140, ...</td>\n",
       "      <td>[119, 99, 92, 100, 99, 92, 91, 92, 89, 87, 82,...</td>\n",
       "      <td>106.363048</td>\n",
       "      <td>83.558719</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>S118</th>\n",
       "      <td>[85, 84, 86, 81, 82, 85, 81, 83, 83, 82, 84, 8...</td>\n",
       "      <td>[]</td>\n",
       "      <td>93.967755</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                           RR_distances  \\\n",
       "S114  [117, 116, 115, 112, 113, 114, 114, 112, 116, ...   \n",
       "S115  [113, 118, 80, 151, 118, 123, 76, 152, 119, 11...   \n",
       "S116  [143, 145, 141, 146, 143, 142, 144, 142, 143, ...   \n",
       "S117  [142, 143, 147, 148, 145, 139, 148, 139, 140, ...   \n",
       "S118  [85, 84, 86, 81, 82, 85, 81, 83, 83, 82, 84, 8...   \n",
       "\n",
       "                                            S_distances  Avg_RR_distance  \\\n",
       "S114                                                 []       114.914745   \n",
       "S115                                               [73]        89.041006   \n",
       "S116                                               [75]       122.611082   \n",
       "S117  [119, 99, 92, 100, 99, 92, 91, 92, 89, 87, 82,...       106.363048   \n",
       "S118                                                 []        93.967755   \n",
       "\n",
       "      AVG_S_distance  \n",
       "S114             NaN  \n",
       "S115       73.000000  \n",
       "S116       75.000000  \n",
       "S117       83.558719  \n",
       "S118             NaN  "
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "signal_ids = filtered_df.index.tolist()\n",
    "RR_dict = {}\n",
    "\n",
    "for id_ in tqdm(signal_ids):  \n",
    "  sig_Id = df.iloc[id_]['sigId']\n",
    "  RR_dict[sig_Id] = {}\n",
    "  RR_dict[sig_Id]['RR_distances'] = list()\n",
    "  RR_dict[sig_Id]['S_distances'] = list()\n",
    "  \n",
    "  peaks = df.iloc[id_]['peaks']\n",
    "  labels = df.iloc[id_]['labels']\n",
    "\n",
    "  for i,p in enumerate(peaks[1:-1]):\n",
    "    RR_distance = p-peaks[i]\n",
    "    if(RR_distance < 250): # discard outliers\n",
    "      RR_dict[sig_Id]['RR_distances'].append(RR_distance)\n",
    "      if(labels[i+1] == 'S'):\n",
    "        RR_dict[sig_Id]['S_distances'].append(RR_distance)\n",
    "\n",
    "  RR_dict[sig_Id]['RR_distances'] = np.asarray(RR_dict[sig_Id]['RR_distances'])\n",
    "  RR_dict[sig_Id]['S_distances'] = np.asarray(RR_dict[sig_Id]['S_distances'])\n",
    "\n",
    "  RR_dict[sig_Id]['Avg_RR_distance'] = RR_dict[sig_Id]['RR_distances'].mean()\n",
    "  RR_dict[sig_Id]['AVG_S_distance'] = RR_dict[sig_Id]['S_distances'].mean()\n",
    "\n",
    "\n",
    "RR_df = pd.DataFrame.from_dict(RR_dict,orient=\"index\")\n",
    "RR_df.tail()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2> Patch creation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "import heartpy as hp \n",
    "\n",
    "classes = np.array([\"N\",\"S\",\"V\"])\n",
    "\n",
    "patch_length = 350 \n",
    "\n",
    "def convert_to_one_hot(label):\n",
    "    return np.array(classes == label,dtype=np.float32)\n",
    "\n",
    "def create_patch_dataset(df):\n",
    "    \n",
    "    dataset_dict = {}\n",
    "    ids = df['sigId']\n",
    "   \n",
    "    for id in tqdm(ids):\n",
    "        row = df[df['sigId'] == id]\n",
    "        sigId = row['sigId'].values[0]\n",
    "        peaks = row['peaks'].values[0]\n",
    "        first_lead_signal = row['ecg_lead_1'].values[0]\n",
    "        second_lead_signal = row['ecg_lead_2'].values[0]\n",
    "        labels = row['labels'].values[0]\n",
    "        avg_RR = RR_df.loc[sigId]['Avg_RR_distance'] # T for the signal \n",
    "\n",
    "        for i,peak in enumerate(peaks):\n",
    "            \n",
    "            stringIdx = str(sigId)+'_'+str(i)    \n",
    "            dataset_dict[stringIdx] = {}\n",
    "            dataset_dict[stringIdx][\"first_lead\"] = list()\n",
    "            dataset_dict[stringIdx][\"second_lead\"] = list()\n",
    "            size = list(range(int(peak-(avg_RR)),int(peak+avg_RR)))\n",
    "\n",
    "            for s in size:\n",
    "                if(s < 0 or s >= len(first_lead_signal)):\n",
    "                    dataset_dict[stringIdx][\"first_lead\"].append(0.5)\n",
    "                    dataset_dict[stringIdx][\"second_lead\"].append(0.5)\n",
    "                else:\n",
    "                    dataset_dict[stringIdx][\"first_lead\"].append(first_lead_signal[s])\n",
    "                    dataset_dict[stringIdx][\"second_lead\"].append(second_lead_signal[s])\n",
    "            \n",
    "            first_lead = dataset_dict[stringIdx][\"first_lead\"][:]\n",
    "            second_lead = dataset_dict[stringIdx][\"second_lead\"][:]\n",
    "            \n",
    "            first_lead = resample(first_lead,patch_length)\n",
    "            second_lead = resample(second_lead,patch_length)\n",
    "            \n",
    "            dataset_dict[stringIdx][\"first_lead\"] = first_lead[:]\n",
    "            dataset_dict[stringIdx][\"second_lead\"] = second_lead[:]\n",
    "            \n",
    "            # uncomment to create also discrete wavelet transform \n",
    "            #dataset_dict[stringIdx][\"dwt_1\"] = pywt.dwt(first_lead,'db4','constant')\n",
    "            #dataset_dict[stringIdx][\"dwt_2\"] = pywt.dwt(second_lead,'db4','constant')\n",
    "             # uncomment to create also discrete wavelet transform multilevel\n",
    "            #dataset_dict[stringIdx][\"dwt_1_m\"] = pywt.wavedec(first_lead, 'db3', level=2)\n",
    "            #dataset_dict[stringIdx][\"dwt_2_m\"] = pywt.wavedec(second_lead, 'db3', level=2)\n",
    "            \n",
    "            dataset_dict[stringIdx][\"stringLabel\"] = labels[i]\n",
    "            dataset_dict[stringIdx][\"label\"] = convert_to_one_hot(labels[i])\n",
    "\n",
    "    dataset_df = pd.DataFrame.from_dict(dataset_dict,orient=\"index\")\n",
    "    return dataset_df"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2> Split dataset by patient & creation of patches"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 79/79 [00:24<00:00,  3.17it/s]\n",
      "100%|██████████| 10/10 [00:03<00:00,  3.24it/s]\n",
      "100%|██████████| 10/10 [00:03<00:00,  3.09it/s]\n"
     ]
    }
   ],
   "source": [
    "\n",
    "training_df, test_df_ = train_test_split(filtered_df,test_size = 0.2,random_state=SEED)\n",
    "validation_df, test_df = train_test_split(test_df_,test_size = 0.5,random_state=SEED)\n",
    "\n",
    "training_patches_df_ = create_patch_dataset(training_df)\n",
    "validation_patches_df_ = create_patch_dataset(validation_df)\n",
    "test_patches_df = create_patch_dataset(test_df)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2> Drop and resampling "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "How many training indx: 168940\n",
      "Balancing training...\n",
      "pre drop: 181313\n",
      "post drop:27358\n",
      "20924\n",
      "1032\n",
      "1703\n",
      "Balancing validation\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "training_patches_df = training_patches_df_.copy(deep=True)\n",
    "validation_patches_df = validation_patches_df_.copy(deep=True)\n",
    "\n",
    "training_df.head()\n",
    "\n",
    "training_patients = training_df['sigId']\n",
    "validation_patients = validation_df['sigId']\n",
    "\n",
    "# calculate percentage of heartbeats to drop \n",
    "\n",
    "num_n = len(training_patches_df[(training_patches_df['stringLabel']=='N')])\n",
    "num_s = len(training_patches_df[(training_patches_df['stringLabel']=='S')])\n",
    "num_v = len(training_patches_df[(training_patches_df['stringLabel']=='V')])\n",
    "to_drop = num_n - max(num_s,num_v) * 2\n",
    "perc_to_drop_train = to_drop/(num_n)\n",
    "\n",
    "num_n = len(validation_patches_df[(validation_patches_df['stringLabel']=='N')])\n",
    "num_s = len(validation_patches_df[(validation_patches_df['stringLabel']=='S')])\n",
    "num_v = len(validation_patches_df[(validation_patches_df['stringLabel']=='V')])\n",
    "to_drop = 0\n",
    "perc_to_drop_valid = 0 \n",
    "# to_drop = num_n - max(num_s,num_v) ---- uncomment to balance valid\n",
    "# perc_to_drop_train = to_drop/(num_n)\n",
    "\n",
    "train_indexes = training_patches_df.index[training_patches_df['stringLabel'] == 'N'].tolist()\n",
    "valid_indexes = validation_patches_df.index[validation_patches_df['stringLabel'] == 'N'].tolist()\n",
    "\n",
    "print(f\"How many training indx: {len(train_indexes)}\")\n",
    "\n",
    "train_to_drop = list()\n",
    "valid_to_drop = list()\n",
    "lengths = list()\n",
    "dict_train = {}\n",
    "valid_train = {}\n",
    "\n",
    "print(\"Balancing training...\")\n",
    "\n",
    "for p in training_patients:\n",
    "    dict_train[p] = list()\n",
    "\n",
    "for hb in train_indexes:\n",
    "    dict_train[hb.split('_')[0]].append(hb)\n",
    "\n",
    "for p in dict_train:\n",
    "   random.shuffle(dict_train[p])\n",
    "\n",
    "for k in dict_train:\n",
    "    train_to_drop += dict_train[k][:int(len(dict_train[k]) * perc_to_drop_train)]\n",
    "\n",
    "print(f\"pre drop: {len(training_patches_df.index)}\")\n",
    "training_patches_df = training_patches_df.drop(train_to_drop)\n",
    "print(f\"post drop:{len(training_patches_df.index)}\")\n",
    "print(num_n)\n",
    "print(num_s)\n",
    "print(num_v)\n",
    "print(\"Balancing validation\")\n",
    "\n",
    "for p in validation_patients:\n",
    "    valid_train[p] = list()\n",
    "    \n",
    "for hb in valid_indexes:\n",
    "    valid_train[hb.split('_')[0]].append(hb)\n",
    "\n",
    "for p in valid_train:\n",
    "    random.shuffle(valid_train[p])\n",
    "\n",
    "for k in valid_train:\n",
    "    valid_to_drop += valid_train[k][:int(perc_to_drop_valid * len(valid_train[k]))]\n",
    "\n",
    "validation_balanced_patches_df = validation_patches_df.drop(valid_to_drop)\n",
    "\n",
    "\n",
    "training_labels = training_patches_df['stringLabel'].values\n",
    "validation_labels = validation_balanced_patches_df['stringLabel'].values\n",
    "\n",
    "classes = [\"N\",\"S\",\"V\"]\n",
    "\n",
    "training_counts = [0]*3\n",
    "validation_counts = [0]*3\n",
    "\n",
    "for l in training_labels:\n",
    "    training_counts[classes.index(l)] += 1\n",
    "for l in validation_labels:\n",
    "    validation_counts[classes.index(l)] += 1\n",
    "\n",
    "fig1, (ax1,ax2) = plt.subplots(1,2)\n",
    "\n",
    "ax1.pie(training_counts, explode=None, labels=classes, autopct='%1.1f%%',\n",
    "        shadow=False, startangle=90)\n",
    "ax1.axis('equal')  \n",
    "ax2.pie(validation_counts, explode=None, labels=classes, autopct='%1.1f%%',\n",
    "        shadow=False, startangle=90)\n",
    "ax2.axis('equal')  \n",
    "\n",
    "fig1.set_facecolor(\"white\")\n",
    "\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "S: 7473\n",
      "V: 4900\n",
      "N: 14985\n",
      "S to res\n",
      "7512\n",
      "V to res\n",
      "10085\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 7512/7512 [00:00<00:00, 18610.10it/s]\n",
      "100%|██████████| 10085/10085 [00:00<00:00, 19101.87it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "S: 14985\n",
      "V: 14985\n",
      "N: 14985\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "# resample v and s\n",
    "from sklearn.utils import shuffle\n",
    "\n",
    "\n",
    "ids_S = training_patches_df.index[training_patches_df['stringLabel']=='S']\n",
    "ids_V = training_patches_df.index[training_patches_df['stringLabel']=='V']\n",
    "ids_N = training_patches_df.index[training_patches_df['stringLabel']=='N']\n",
    "\n",
    "print(f\"S: {len(ids_S)}\")\n",
    "print(f\"V: {len(ids_V)}\")\n",
    "print(f\"N: {len(ids_N)}\")\n",
    "\n",
    "to_add = {}\n",
    "\n",
    "to_resample_s = len(ids_N) - len(ids_S)\n",
    "to_resample_v = len(ids_N) - len(ids_V)\n",
    "\n",
    "print(\"S to res\")\n",
    "print(to_resample_s)\n",
    "print(\"V to res\")\n",
    "print(to_resample_v)\n",
    "\n",
    "for i in tqdm(range(to_resample_s)):\n",
    "    id = random.choice(ids_S)\n",
    "    row = training_patches_df.loc[id]\n",
    "    idx = id + '_' + str(i)\n",
    "    to_add[idx] = {}\n",
    "    to_add[idx]['first_lead'] = row['first_lead']\n",
    "    to_add[idx]['second_lead'] = row['second_lead']\n",
    "    to_add[idx]['stringLabel'] = row['stringLabel']\n",
    "    to_add[idx]['label'] = row['label']\n",
    "\n",
    "\n",
    "\n",
    "for i in tqdm(range(to_resample_v)):\n",
    "    id = random.choice(ids_V)\n",
    "    row = training_patches_df.loc[id]\n",
    "    idx = id + '_' + str(i)\n",
    "    to_add[idx] = {}\n",
    "    to_add[idx]['first_lead'] = row['first_lead']\n",
    "    to_add[idx]['second_lead'] = row['second_lead']\n",
    "    to_add[idx]['stringLabel'] = row['stringLabel']\n",
    "    to_add[idx]['label'] = row['label']\n",
    "\n",
    "\n",
    "to_add_df = pd.DataFrame.from_dict(to_add,orient=\"index\")\n",
    "training_patches_df = pd.concat([training_patches_df,to_add_df],axis=0) \n",
    "training_patches_df = shuffle(training_patches_df)\n",
    "\n",
    "ids_S = training_patches_df.index[training_patches_df['stringLabel']=='S']\n",
    "ids_V = training_patches_df.index[training_patches_df['stringLabel']=='V']\n",
    "ids_N = training_patches_df.index[training_patches_df['stringLabel']=='N']\n",
    "\n",
    "print(f\"S: {len(ids_S)}\")\n",
    "print(f\"V: {len(ids_V)}\")\n",
    "print(f\"N: {len(ids_N)}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "\n",
    "training_labels = training_patches_df['stringLabel'].values\n",
    "validation_labels = validation_balanced_patches_df['stringLabel'].values\n",
    "\n",
    "classes = [\"N\",\"S\",\"V\"]\n",
    "\n",
    "training_counts = [0]*3\n",
    "validation_counts = [0]*3\n",
    "\n",
    "for l in training_labels:\n",
    "    training_counts[classes.index(l)] += 1\n",
    "for l in validation_labels:\n",
    "    validation_counts[classes.index(l)] += 1\n",
    "\n",
    "fig1, (ax1,ax2) = plt.subplots(1,2)\n",
    "\n",
    "ax1.pie(training_counts, explode=None, labels=classes, autopct='%1.1f%%',\n",
    "        shadow=False, startangle=90)\n",
    "ax1.axis('equal')  \n",
    "ax2.pie(validation_counts, explode=None, labels=classes, autopct='%1.1f%%',\n",
    "        shadow=False, startangle=90)\n",
    "ax2.axis('equal')  \n",
    "\n",
    "fig1.set_facecolor(\"white\")\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "# function to create input for models except the ones that leverage the autoencoder \n",
    "def create_input(df):\n",
    "    \n",
    "    x = list()\n",
    "    y = list()\n",
    "    \n",
    "    for id in tqdm(df.index.tolist()):\n",
    "        row = df.loc[id]\n",
    "        x.append(np.transpose(np.asarray([row['first_lead'],row['second_lead']]).astype(np.float32)))\n",
    "        y.append(row['label'])\n",
    "\n",
    "    x = np.asarray(x).astype(np.float32)\n",
    "    y = np.asarray(y).astype(np.float32)\n",
    "    \n",
    "    return x,y\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "# function to create input of the autoencoder \n",
    "def create_autoencoder_input(df):\n",
    "    x = list()\n",
    "    \n",
    "    for id in tqdm(df.index.tolist()):\n",
    "        row = df.loc[id]\n",
    "        x.append(np.transpose(np.asarray([row['first_lead']]).astype(np.float32)))\n",
    "\n",
    "    x = np.asarray(x).astype(np.float32)\n",
    "    \n",
    "    return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 44955/44955 [00:01<00:00, 22697.03it/s]\n",
      "100%|██████████| 23659/23659 [00:00<00:00, 24037.88it/s]\n"
     ]
    }
   ],
   "source": [
    "x_autoencoder = create_autoencoder_input(training_patches_df)\n",
    "x_valid_auto = create_autoencoder_input(validation_patches_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "176/176 [==============================] - 2s 7ms/step - loss: 0.0247 - val_loss: 0.0170\n",
      "Epoch 2/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0150 - val_loss: 0.0136\n",
      "Epoch 3/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0127 - val_loss: 0.0120\n",
      "Epoch 4/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0114 - val_loss: 0.0113\n",
      "Epoch 5/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0103 - val_loss: 0.0102\n",
      "Epoch 6/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0091 - val_loss: 0.0089\n",
      "Epoch 7/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0081 - val_loss: 0.0079\n",
      "Epoch 8/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0074 - val_loss: 0.0074\n",
      "Epoch 9/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0071 - val_loss: 0.0070\n",
      "Epoch 10/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0067 - val_loss: 0.0066\n",
      "Epoch 11/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0064 - val_loss: 0.0063\n",
      "Epoch 12/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0061 - val_loss: 0.0060\n",
      "Epoch 13/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0059 - val_loss: 0.0057\n",
      "Epoch 14/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0056 - val_loss: 0.0055\n",
      "Epoch 15/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0054 - val_loss: 0.0052\n",
      "Epoch 16/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0052 - val_loss: 0.0050\n",
      "Epoch 17/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0051 - val_loss: 0.0049\n",
      "Epoch 18/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0049 - val_loss: 0.0048\n",
      "Epoch 19/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0048 - val_loss: 0.0046\n",
      "Epoch 20/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0046 - val_loss: 0.0046\n",
      "Epoch 21/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0045 - val_loss: 0.0044\n",
      "Epoch 22/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0044 - val_loss: 0.0043\n",
      "Epoch 23/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0043 - val_loss: 0.0042\n",
      "Epoch 24/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0042 - val_loss: 0.0041\n",
      "Epoch 25/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0041 - val_loss: 0.0040\n",
      "Epoch 26/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0040 - val_loss: 0.0039\n",
      "Epoch 27/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0040 - val_loss: 0.0038\n",
      "Epoch 28/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0039 - val_loss: 0.0037\n",
      "Epoch 29/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0038 - val_loss: 0.0036\n",
      "Epoch 30/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0037 - val_loss: 0.0035\n",
      "Epoch 31/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0037 - val_loss: 0.0035\n",
      "Epoch 32/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0036 - val_loss: 0.0034\n",
      "Epoch 33/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0035 - val_loss: 0.0033\n",
      "Epoch 34/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0035 - val_loss: 0.0033\n",
      "Epoch 35/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0034 - val_loss: 0.0032\n",
      "Epoch 36/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0034 - val_loss: 0.0032\n",
      "Epoch 37/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0033 - val_loss: 0.0031\n",
      "Epoch 38/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0033 - val_loss: 0.0030\n",
      "Epoch 39/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0032 - val_loss: 0.0030\n",
      "Epoch 40/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0032 - val_loss: 0.0029\n",
      "Epoch 41/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0031 - val_loss: 0.0029\n",
      "Epoch 42/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0031 - val_loss: 0.0029\n",
      "Epoch 43/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0031 - val_loss: 0.0028\n",
      "Epoch 44/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0030 - val_loss: 0.0027\n",
      "Epoch 45/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0030 - val_loss: 0.0027\n",
      "Epoch 46/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0030 - val_loss: 0.0026\n",
      "Epoch 47/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0029 - val_loss: 0.0026\n",
      "Epoch 48/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0029 - val_loss: 0.0026\n",
      "Epoch 49/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0029 - val_loss: 0.0026\n",
      "Epoch 50/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.0028 - val_loss: 0.0026\n",
      "Epoch 51/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0028 - val_loss: 0.0025\n",
      "Epoch 52/100\n",
      "176/176 [==============================] - 2s 10ms/step - loss: 0.0028 - val_loss: 0.0025\n",
      "Epoch 53/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0028 - val_loss: 0.0024\n",
      "Epoch 54/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0027 - val_loss: 0.0024\n",
      "Epoch 55/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.0027 - val_loss: 0.0024\n",
      "Epoch 56/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.0027 - val_loss: 0.0023\n",
      "Epoch 57/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0027 - val_loss: 0.0023\n",
      "Epoch 58/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0026 - val_loss: 0.0023\n",
      "Epoch 59/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0026 - val_loss: 0.0022\n",
      "Epoch 60/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0026 - val_loss: 0.0022\n",
      "Epoch 61/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0026 - val_loss: 0.0022\n",
      "Epoch 62/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0026 - val_loss: 0.0022\n",
      "Epoch 63/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0026 - val_loss: 0.0022\n",
      "Epoch 64/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0025 - val_loss: 0.0021\n",
      "Epoch 65/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0025 - val_loss: 0.0022\n",
      "Epoch 66/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0025 - val_loss: 0.0021\n",
      "Epoch 67/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0025 - val_loss: 0.0021\n",
      "Epoch 68/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0025 - val_loss: 0.0021\n",
      "Epoch 69/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0024 - val_loss: 0.0021\n",
      "Epoch 70/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0024 - val_loss: 0.0021\n",
      "Epoch 71/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0024 - val_loss: 0.0020\n",
      "Epoch 72/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0024 - val_loss: 0.0020\n",
      "Epoch 73/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0024 - val_loss: 0.0020\n",
      "Epoch 74/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0024 - val_loss: 0.0020\n",
      "Epoch 75/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0024 - val_loss: 0.0020\n",
      "Epoch 76/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0024 - val_loss: 0.0020\n",
      "Epoch 77/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0023 - val_loss: 0.0019\n",
      "Epoch 78/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0023 - val_loss: 0.0019\n",
      "Epoch 79/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0023 - val_loss: 0.0019\n",
      "Epoch 80/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0023 - val_loss: 0.0019\n",
      "Epoch 81/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0023 - val_loss: 0.0019\n",
      "Epoch 82/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0023 - val_loss: 0.0019\n",
      "Epoch 83/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0023 - val_loss: 0.0019\n",
      "Epoch 84/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0023 - val_loss: 0.0019\n",
      "Epoch 85/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0022 - val_loss: 0.0019\n",
      "Epoch 86/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0022 - val_loss: 0.0018\n",
      "Epoch 87/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0022 - val_loss: 0.0018\n",
      "Epoch 88/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0022 - val_loss: 0.0018\n",
      "Epoch 89/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0022 - val_loss: 0.0018\n",
      "Epoch 90/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.0022 - val_loss: 0.0018\n",
      "Epoch 91/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0022 - val_loss: 0.0018\n",
      "Epoch 92/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0022 - val_loss: 0.0018\n",
      "Epoch 93/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0022 - val_loss: 0.0018\n",
      "Epoch 94/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0022 - val_loss: 0.0018\n",
      "Epoch 95/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0021 - val_loss: 0.0018\n",
      "Epoch 96/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0021 - val_loss: 0.0017\n",
      "Epoch 97/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0021 - val_loss: 0.0017\n",
      "Epoch 98/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0021 - val_loss: 0.0017\n",
      "Epoch 99/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0021 - val_loss: 0.0017\n",
      "Epoch 100/100\n",
      "176/176 [==============================] - 1s 6ms/step - loss: 0.0021 - val_loss: 0.0017\n"
     ]
    }
   ],
   "source": [
    "import keras\n",
    "from keras import layers\n",
    "\n",
    "# This is the size of our encoded representations\n",
    "encoding_dim = 40 # 350/50 = 7 compression factor\n",
    "\n",
    "input_shape = x_autoencoder.shape[1]\n",
    "\n",
    "input = keras.Input(shape=input_shape,name='Input_layer')\n",
    "\n",
    "# encoding \n",
    "\n",
    "dense1 = layers.Dense(512,activation='relu',name='Hidden_1')(input)\n",
    "dropout1 = layers.Dropout(0.1,name='Dropout_1')(dense1)\n",
    "dense2 = layers.Dense(256,activation='relu',name='Hidden_2')(dropout1)\n",
    "\n",
    "encoded = layers.Dense(encoding_dim, activation='sigmoid',name='Encoded_input')(dense2)\n",
    "\n",
    "dense3 = layers.Dense(256,activation='relu',name='Hidden_3')(encoded)\n",
    "dropout2 = layers.Dropout(0.1,name='Dropout_2')(dense3)\n",
    "dense4 = layers.Dense(512,activation='relu',name='Hidden_4')(dropout2)\n",
    "\n",
    "# decoding\n",
    "\n",
    "decoded = layers.Dense(input_shape, activation='sigmoid',name='Decoded_input')(dense4)\n",
    "\n",
    "# This model maps an input to its reconstruction\n",
    "autoencoder = keras.Model(input, decoded)\n",
    "\n",
    "# This model maps an input to its encoded representation\n",
    "encoder = keras.Model(input, encoded)\n",
    "\n",
    "\n",
    "\n",
    "dot_img_file = 'autencoder.png'\n",
    "autoencoder.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.0001), loss=\"mse\")\n",
    "\n",
    "\n",
    "tf.keras.utils.plot_model(autoencoder, to_file=dot_img_file, show_shapes=False)\n",
    "\n",
    "history = autoencoder.fit(x_autoencoder, x_autoencoder,\n",
    "                epochs=100,\n",
    "                batch_size=256,\n",
    "                shuffle=True,\n",
    "                validation_data=(x_valid_auto, x_valid_auto))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history.history[\"loss\"], label=\"Training Loss\")\n",
    "plt.plot(history.history[\"val_loss\"], label=\"Validation Loss\")\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2> Save the autoencoder weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Save the weights\n",
    "# change directory in case of errors \n",
    "autoencoder.save_weights('./models/autoencoder/autoencoder_weights')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_sample(signal):\n",
    "    fig,ax = plt.subplots()\n",
    "    x = list(range(len(signal)))\n",
    "    ax.plot(x,signal)\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "# try the autoencoder \n",
    "\n",
    "samples = x_autoencoder[:10000]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "decoded = autoencoder.predict(samples)\n",
    "encoded = encoder.predict(samples)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAD4CAYAAAD8Zh1EAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/YYfK9AAAACXBIWXMAAAsTAAALEwEAmpwYAAA+gklEQVR4nO2dd3xkZ3X3v8+MNGqjPup1e7XXW+xd93WDtQk2GAcbMJAXg0MLISYJJhCTmECAJMCbvJSYElNdAjYYXHE3ttfb+2p3tdpd9a5RG42mPe8fc2dWq1XXnarz/Xz245l7r+49vpr56dzznKK01giCIAiJjyXWBgiCIAjmIIIuCIKQJIigC4IgJAki6IIgCEmCCLogCEKSkBKrCzscDl1bWxurywuCICQku3fv7tZaF020L2aCXltby65du2J1eUEQhIREKXVmsn0SchEEQUgSRNAFQRCSBBF0QRCEJEEEXRAEIUkQQRcEQUgSphV0pdRPlFKdSqlDk+xXSqn/VErVK6UOKKU2mG+mIAiCMB0z8dAfBLZNsf9GYJnx727g+/M3SxAEQZgt0wq61vpVoHeKQ24BfqaDbAfylFJlZhkoCEL8MDzq4+dvnqZz0B1rU4QJMCOGXgE0jXnfbGwThKTg6YNtbPzKH/n4z3fH2pSY89M3T/OPvzvM9f/xCsOjvlibI4wjqouiSqm7lVK7lFK7urq6onlpQZgzj+1toWfYwzOH2+kd9sTanJjyxL5WAAbcPt461RNja4TxmCHoLUDVmPeVxrbz0Fo/oLXepLXeVFQ0YSsCQYgrfP4A20/2sKzYDsDO01NFH5Ob+s5B6toH+eJNq0hPtfDq8e5YmySMwwxBfwL4kJHtsgXo11q3mXBeQYg5B1r6GRz18clrlmBLsbDz1MIV9Nfrgx75trWlbF5UyKsn5Ck73phJ2uJDwJvACqVUs1LqLqXUx5VSHzcOeQpoAOqBHwKfjJi1ghBl9jU6Abh8qYN1lbnsbXLG1J5Y8tapHiryMqgqyOTalcU0dA1T3zkYa7OEMUzbbVFr/b5p9mvgU6ZZJAhxxInOQfIzUymyp7G02M5zhztibVJM0FqzvaGXrSuCodJta0v58hOH+cOBNj6xNROb1YJSKsZWClIpKghTcLxjiGUl2SilqCnMomfYw4DbG2uzos6ZHhe9wx421RQAUJKTzpbFBXzn+ROs+NIzfOf5EzG2UAARdEGYFK01x9sHWVGSDUBtYSYAjT2uWJoVE460DQCwtiInvO3/vX8DX7xpFQ67jR+91oDTtbAzgOIBEXRBmIT2ATeDoz6WlwQzXGoKswA43TMcS7NiwtG2AawWxXLjjxuAw57Gx65azC8+uplhj59f726OoYUCiKALwqQcaw8u+C0zRKzG8NDPLEQPvXWApUV20lOt5+1bWZrD0mI7rxyXrJdYI4IuCJNwomMIIOyVZtpSKMpO43T3wvPQj7QNsKose9L9W5cX8VZDLy6PVI/GEhF0QZiE4x2DOOxpFGTZwtsq8zNo7R+JoVXRx+3109bvZqlRXDURV68owuMPsL1BqkdjiQi6IEzC8Y7BcPw8RHluBm3OhdWYqrkv+AesMj9z0mMuri0gI9XKy8ck7BJLRNAFYQICAc2JzqFzFgEByvPSaXGOECy/WBg09wXXDCrzMyY9Jj3VyqVLCiWOHmNE0AVhApr6XLg8flaUnivoZbkZjPoC9LkWTi56k+GhVxVM7qEDbF1RxJke14JM64wXRNAFYQIOtvQDcEFF7jnby/OCXmqrc+HE0Zv7XNisForsaVMeFyo62tPYFw2zhAkQQReECTjY3I/Napkw5ALQsqAEfYSK/AwslqlL+5eX2MlItbJvAfe7iTUi6IIwAQdb+llZlo0t5dyvSMhDb1tggj5V/DxEitXCBQu8gVmsEUEXhHEEApqDLf2sHRduASjMsmFLsdDav3AyXZp7XVNmuIxlfVUeR1r78fgCEbZKmAgRdEEYR33XEINuHxuq88/bp5SiPDd9wcTQXR4fPcOeGXnoACtKs/H6NY29sjAaC0TQBWEcu04HF/U21pwv6BAMuywUQW8J56DPTNAXFwXz9hu6hiJmkzA5IuiCMI7dZ/oozLKFuyuOpyw3g7YFEnJpMnLQp0tZDLHIEWxg1rAA2yPEAyLogjCOvY19rK/On3RgQ0VeOh0Dbnz+5I8TN8/SQ8/NSMVht4mHHiNE0AVhDINuLw3dw6yrPH9BNERZXgYBDR2Do1G0LDY0942QljJ9DvpYFjvsnBIPPSaIoJtIIKD57d4Wvv3H4wvCe0tGjrYFW+ZOlOESYiEVF53pGaYyP2NW4+UWF2XR0CWCHgumnSkqzJydp3v57CP7gGD3uYmyJIT45pBRIbpmzGSe8ZTnBouLFoKgH2oZYH113qx+ZnFRcFRfv8tLbmZqZAwTJkQ8dBMJlYvDwhxTlgwcbh2gKDuN4uz0SY8pC3voyb0w2j00SotzhHWVebP6ucWOYKbLyW6Jo0cbEXQTOdjST2GWDaUW5lSbZOBE5yArSycf5ABgT0shJz2FtiTvi36g2QnAuqq8Wf3c4iIj00XCLlFHBN1EDrX0s6Emn7KcdM4swLmTiY7WmlPdw9Qas0OnYiHkou9tdGJR5w6GnglVBZmkWJRkusQAEXSTcHl8NHQPs7Y8l5rCrAU5SDjR6R32MOj2hXOppyIo6MkbctFa89TBNjbVFJBpm91SW6rVQnVhpmS6xAARdJNo7htB6+DjZq0jU0IuCUjoj/DMBD09qUfRHW4d4GTXMO9aXzGnn19Rks3h1gGTrRKmQwTdJELtVMvzMqjMz6Rn2MOIxx9jq4TZEIr51s5A0MtyM3C6vEk7FDk0eejGtaVz+vmNNfk09rroHEjep5h4RATdJELx1Iq8DPIzg0OF+0cWzlSbZOB0zzBWi5pRVWRFkme6HG0boKogg/wxA7JnQ6gPzq4zMuwimoigm0Src4QUi6IoO43cjGDurQh6YtHUO0J5Xjqp1um/FmVGLnp7kvZ0qWsfZEXJ7BZDx7KmPJf0VAs7TvWaaJUwHSLoJtHqdFOam47VosjJCC4iiaAnFq3OkbDnPR2O7GApfM9w8pX/j/r8nOoenjZ9cypsKRYuWVTIS8c6F9RA7Vgjgm4SLc6RcEm4eOiJSeuY3+F0OLKCgt6VhP1c6juH8Ac0K8vmLugAb1tdwpkeFyc6JX0xWoigm8RY704EPfHw+QO0D7hn7KHnZKSQYlH0DHsibFn0CS0OLy22z+s8b1tdAsAfj3TM2yZhZsxI0JVS25RSx5RS9UqpeyfYX62UekkptVcpdUApdZP5psYvWms6B0YpyQnGVUXQE4+OwVECmhl76EopCu02eoaSz0MPrQvM9F5MRnFOOsuK7ew6LXH0aDGtoCulrMB3gRuB1cD7lFKrxx32JeBRrfV64A7ge2YbGs+4PH48/gD5RiOi7HQR9ESjdUza6Uxx2NPoGUo+D72t302WzUp22vx7922syWdPo5NAQOLo0WAmHvolQL3WukFr7QEeBm4Zd4wGQkviuUCreSbGP05DuPMMQbdaFNnpKQyIoCcMZ9NOJ2/KNZ5CexrdyeihD4xQmps+q5a5k7GhOp/+Ea9MMIoSMxH0CqBpzPtmY9tY/gm4UynVDDwF/NVEJ1JK3a2U2qWU2tXV1TUHc+MTpyvopeVlns3Zzc1IFQ89gQgVhpXlzsJDz7LRnYQeeqvTPav7MBUbjHz0PY2Sjx4NzFoUfR/woNa6ErgJ+LlS6rxza60f0Fpv0lpvKioqMunSscfpMjz0jLO9n0XQE4tW5wh5malkzSLM4MhOo2d4NOnS8tr7gym4ZrDYkUWmzcoRaQMQFWYi6C1A1Zj3lca2sdwFPAqgtX4TSAccZhiYCIQFXTz0hKXV6aZ8ll5pYZYNtzfAcBK1ePD5A3QOusOFU/PFYlGsKssRQY8SMxH0ncAypdQipZSN4KLnE+OOaQSuA1BKrSIo6MkTU5kG50go5CIeeqIymxz0EAVGWXxvEoVduoaC2T5meegAq8tyONI2IAujUWBaQdda+4BPA88CRwlmsxxWSt2vlLrZOOxzwMeUUvuBh4C/0Mn2HDoFIQ89V0IuCUuLc2RWC6Jw9oksmX7PreG1BPMEfU15DkOjPpr6pANppJlRwFBr/RTBxc6x2+4b8/oIcLm5piUOTpeHjFQr6anW8DZ7WgrDo8nZiS/ZGHB7GXT7Zu2h56QnX4uH5r6goFflZ5p2ztDA7X1NTmpmMDxEmDtSKWoCTpf3nHALQGZaCi6PXx4zE4A259wKaUIDkJNR0Ctm0HFypqwszSbLZmWnFBhFHBF0E3COeM8JtwDY04Le+nCS9stOJs4WFc0uzJCMFcHNfS4cdtuspxRNRYrVwoaafHadltTFSCOCbgL9E3joofQ3VxJlQCQrjb3B2G5VwezCDCFBH3Ank6CPUGFiuCXExbUFHOsYpN+VPPcqHhFBN4E+lyc81CKE3RD0IYmjxz1nelxkpFopsqfN6ucyUq2kWlVSeehNva4ZDfiYLVsWF6I1vNnQbfq5hbOIoJuAc2QCD914ZJWF0finsXeY6oLMWZe6K6XISU+ebKZAQNPqdEdE0NdX52FPS+GV4yLokUQEfZ5orel3ecnNONdDzxIPPWE40+OiunBuYYZkSk9t7HXh8QeojUAmSqrVwqVLCnn1+IIpT4kJIujzZMQb7LQ43kMPhVyGRyWGHs8EAprGXhc1s4yfh8jJSE2aJmxH2oLVnGvK5z56biouW1JIi3Mkacf2xQMi6PNkoj4uAFmhLBfx0OOazsFRRn2BeXnoySLoh1v7sVoUy0vmN6loMi6qygOC+ehCZBBBnyd9E3RaBFkUTRQauoLj0eYaZkimkMvh1gGWFdvPKZAzk1VlOaRaFfubnRE5vyCCPm/6w425Jk5bFA89vgmFGVaVzS3MkEyCfqR1gNVzvA8zIT3VyqqyHPaLhx4xRNDnyfjhFiEybVaUSm5B39PYR1NvYvfnONI2QFF2GkXZs0tZDJGTkcKA25fwLXQH3V46B0dZWjK/OaLTcUFFLgdb+hP+fsUrIujz5GwM/dyQi1KKLFsKQ0m6KHqiY5Bbv/cGV37zJbY39MTanDkzX6/UnpaKP6BxewMmWhV9zvQE/zAvdkS218rq8hwG3b7wQBHBXETQ58nZGHrqefuy0qxJ66H/69N1QLBB1ed/cwC3N/H+cI36/NR3DrF6HlkdoRYPib5WEhoRt8gRWQ89FNqS/uiRQQR9nvSPeElPtUy4kJSVlsJQEvZy6Xd5eeV4F5/cuoTvfWAjZ3pc3PHAdv7h8YMJlWe8+0wfvoBmvZF9MReSZa3kVFdQ0GvmmO0zU1aWZqMUHG0bjOh1Fioi6PPE6fKcF24JkSwtdLXWnOgYDHeOfPFYB/6A5obVJVyxzMF//Pk6ugZH+f3+Vv7if3YkzPzIV453kWpVXLZ07sO1Qk2sEr0J2+meYSryMiKW4RIi05bCosIsDrf2R/Q6CxUR9HkyUevcEFm2xBf0QEBz1093ccO3X+X+PxwB4MkD7RRnp7GuMg+A92ys5PV7r+WNe6+lLDeDLz5+KCEWvV451sXGmvxwiulcSJYCslPdw9Q6Iuudh1hXlceexr6E+IwkGgkp6G819PDDVxuAYMjjkq8+zwtHO2Jii9N1fuvcEFlpib8o+lp9Ny/WdXJBRS4PvnGan/zpFC8d6+Q9GyuxWM7tfZKdnspfX7eMo20DvBnnC6Unu4aoax/k+lUl8zpPshSQdQy4Kc0xv4fLRGxeVED3kIeTRg2AYB4JKei3P7Cdrz51FH9As6/JSefgKM/HStBHzu+0GMKeBIuiD73VSEGWjUf+cgsXVORy/x+OENCa911cPeHxN19UTn5mKg++fjq6hs6S3+1twaLgnevK53WeZCggCwQ0XYOjFOfMLXVztmxeXAjA9gYZeGE2CSnoIVqdIxw0qs52n4lN3HbKkEuCx9AH3F5erOvkXRdVkGlL4b/et54Pbqnhv+/cOGmpfHqqlfdvrub5ox3hHHWvP75S+rTWPL6vhcuXOijJmd/szGRYFO1zefAFNMVzzMWfLbWFmZTlpvNKAi2gJwoJLeine4bZ3xxcXDneMRT1ij2tdXBa0SSCbk9LSWjP7aW6Tjz+ADddUApArSOLr7xrLW9bUzrlz925pQalFF/5wxHu/c0Bln/paT7y4E48vvgQ9l1n+mjqHeHd6yvmfa5Qm+RE/j13Do4CzPuP20xRSrFtbSmvHOtKmirbeCEhBT30mHu6x8XB5n5KjQ9itEuKR7x+PL7ApFkuWWkpjPoC+OLMQ52K10508ZM/ncLjC/DkgTaKstPYUJ0/q3OU5WZw77aVPHekg0d2NXHtimJerOvkif2tEbJ6dvx2bwsZqVbePs0fppkQiqEn8mSqkKBHy0MHuOWiCjz+AM8ebo/aNRcCCSnotpSg2QeanLQPuLnlomAc9HhHdHNbnZP0cQmRlWAZEL/b18IHf7yD+/9whCu/+SLPHengvZvOX/ycCR+7ajEPfWwLr/39Nfzow5tYWZrNj15riHlmg9aaF+s6uXp5Ufj3Mx9SrBbSUiwJHXLpHAi2sy3Ojo6HDrCuMpey3HRePtYZtWsuBBJO0LXW4ce0l4wPw8aafBx2G8c7BjnVPczr9dGZihIS9PxJQy5GFWEc5ihrrfnxn07xqV/u4XjHIKM+P//+3DFWl+XwpXesonfYQ0lOGp/YunTO17h0SSGV+cFJQB/YUkNd+yDHovxHdzwnOodo63dz9Yoi086Z6KG1sIcepUVRCIZdrljq4PX6HvwBSV80C/NGe0eJwVFf+APQPRQsu19cZGd5STbHOob4/K8PsK/Jyat/fw2luZH1OJwjweuPn1YUIp4XzH725hm+8ocjpFoVr53oYkNNPk29I/zirgu5YpmD915chd+v55WjPZYb15by5d8d4g/721hZGrmOftMRqmS9arl5gp7oi9+dA26y01MiXlQ0niuXF/G/u5s52NIf7pUuzI+E89Cdw0GveN2YD0B1QSbLS7LZ3+Rkx+lePP4AP3ytIeK2TNY6N0Q8jaH7/f5Wvv/ySboGR+keGuXfnz3GlcscvHDPVmwpVl4+1sXN68q5YlmwajInPZX8rIn/UM0Fhz2Ny5Y4ePJgW0zDLtsbeqktzKQiz7yc60ybNaHrDToHR6MaPw9x+ZJg+uJbcV6zkEgknIceaoa1bU1peBHUlmJhZWlwykqKRbGxJp/njrTzj3+2OsK2TCPocTIounPAzV89tBeAR3Y2UpydzqgvwJffuYbqwkxe/Nurg02qItgLG+AdF5bxhccOcrh1gLUVuRG91kRordnb2GdquAUSv8VDa7+bchP/wM2UQnsatYWZCdMqIhFIPA/diJ9fXHtu5sUtF1Xwb7ddyGufv4brV5XQ1DtClxEbjJwtRqfFSUMu8VFF+OAbp7Eo+NZ719Hn8rK7sY+vvnstS4uDnfVy0lPZUJ0f8UfubWtKsVoUvz8Qm2yXxl4XPcOeWWftTEdWWgquOFwnmSltzhHKIhyenIwN1fnsPuOM+WJ5spBwHrpzzMi3xz95WbjsPsNm5c83VQGwvjoPgL2NfdPmTM+HfpeXtBQLGbaJhfBsFWFsH8efPtTOlcuKuHVDJX92YTmjPj/Z6RM/VUSS/Cwb164s5uEdTfzVtctMi8/PlJAnaLag29NSaO5LzEEfHl+ArqFRynKj76EDrK/J57G9LTT1jsx5rqtwloTz0PuGg4Ken5nK+up8Fhed3795bUUuKRbF7gg/yvUMT172D2dj6LH03lqdI5zqHuZKIzZuS7HERMxDfHLrEvpHvPxi+5moX3vPGSdZNisrSs0dghzse5+YMfSOATdaEzMP/dLFBQC8ckKqRs0g4QQ9P8vGppr8SRtiQbD8/NIlhTy+pyWigxem638Riz4fXn+Arz11NJzN8cbJ4ILT5fNoEWsm66vzuXKZgx+91sBIlItx9jT2sa4qD+sc8uqnIjPOu2p6/QG6hyYOP7b1B3PQy2IQQwdYUmRnsSOL56TAyBRmJOhKqW1KqWNKqXql1L2THPNepdQRpdRhpdSvzDXzLLdcVMGvP3EZKdapTf/41UvoHByNqCfYNThKkX1yQU9LsWC1qKh+2X+5/QwPvNrAh36yg5ePdfJiXQcOu40VJeZ6pfPh09cspXvIw8M7GyNy/hbnCL/f33pOha7L46OufdD0cAsYi6Ke+J0r+uM/nWLTvzzPr3c3n7evrT84Cq48Rh66Uoob1pTw5smeiK95LQSmFXSllBX4LnAjsBp4n1Jq9bhjlgFfAC7XWq8BPmu+qbPjsiWFXLuymG88UxexcVddQ6NTDhcOzhWN3uO41x/gv16sZ0N1HpX5Gdz3u8M8dbCdd64rn1O1Z6TYvLiQS2oL+O9XGhj1mXtvtNZ86pd7+KuH9vKXP98d3r6/qR9/QLOhJs/U60EwtBbQxO1c0b1G6PE7zx8/b1+rM7YeOsDtxtrXRPYJs2MmHvolQL3WukFr7QEeBm4Zd8zHgO9qrfsAtNYxr+dVSvEff74Of0BHpF+EP6DpGZo+fzeaVYRvnOyhZ9jDx69ewmevX06j0e3wto2VUbn+bPj0tUtpH3Dzm90tpp73haOd7GtyUpSdxgt1neF2EG+d6kEp2FhdYOr1IP7nivYYBXihyuaxNHQNUZBli/oC9VgWF9m545IqHtnZRM8koSFhZsxE0CuApjHvm41tY1kOLFdKva6U2q6U2jbRiZRSdyuldimldnV1RX4RJD/LxtJiOweMFrtm0jvsIaCZ0kOH6FYRPnWgjSyblauWF3Hbxkp+9dHNfPv2dawpj37O93RcuczBuspcvvdyPSMeP7/YfoY7f/QWD75+asbn8PkDdA64w6PxAB7b24zDnsaTn7mCVKvikZ3Bj+4b9T2sLc+dtDPmfIjnimCA5r5gWGVo1Hdeo7j9zU7WVcb+8/H+S2rwBTRPHZJY+nwwa1E0BVgGbAXeB/xQKZU3/iCt9QNa601a601FReYWd0zGhZV5HGjuNz2+GYr3zUTQo+G5ub1+nj7Uxg2rS8L55JctdfDu9fHnnUPwCeqet62gxTnCxn/5I1/67SFOdA7yT78/wtMH2845Vms94e/v8785yCVfe4GP/2I3WmtcHh8v1XWxbW0JxdnpvG11KY/taabf5WVvUx+XLS2MyP9LZhy30B31+ekYdIeL3wbcZ20cdHs50TnERVXmryvMllVl2SwvsfP7ffHRkTNRmYmgtwBVY95XGtvG0gw8obX2aq1PAccJCnzMWVeZS8+whxbniKnn7RqamaBHq4rwqYNtDLh9vPfiqukPjhOuXl7Ejz+8iW1rSvnWe9fx2t9fy7JiO997+WT4mEG3l9v/ezvv/+Fb58Tbe4ZGeWJ/8GP43JEOXj7exR+PdDDi9XPTBWUA3H5xFX0uLx/6nx14/ZprVhRH5P/DHsceepszmJa41nhKG9t//GBLP1rDuqrYe+hKKa5ZUcy+Jqfp6yoLiZkI+k5gmVJqkVLKBtwBPDHumN8S9M5RSjkIhmAi30xlBqw2Psh1beZ2+Qt76PapswOikaPs8QX4wSsnWeTI4tLFkfFCI8W1K0v41u0XceuGSmwpFu7cUsPBln4OGoNL/vn3R9jd2MebDT385wsnwj/3+N4WvH7Nk5+5gtrCTL725FF+8vppagsz2bIoeA+uWOpghdHj55LaAjYvMj9+DmMqguOwWjTkyKwpD7Z1CBXmARxuCSYLXGgM+441F1Xl4fEHTP+uLiSmFXSttQ/4NPAscBR4VGt9WCl1v1LqZuOwZ4EepdQR4CXg77TWcdFxZ5EjCwhONzKTVuOLMl3LUTNDLn3DnvBYt7H8+3PHON4xxJfesQql4iebZS68a30F6akWfrWjkc5Bd7BH+5Yatq0p5ZGdTeFxdi/WdbKyNJs15bnce+NKTnQOsb/JyYcurQ1n9Fgsikc/fimfuXYpX3/PBRG7N/Y47nt/pif4eVltCPpYD72+cwiH3UaBiU3Y5kOo4d7+CKx5LRRmtLSttX4KeGrctvvGvNbAPca/uCI/M5Wc9JR5C/o9j+xDA994z4XYUiyc6ByiqiBj2v4noRzl+eL2+rntB2/QM+zhpc9tDXdCfGhHIw+82sCdW6q5bp4T7OOB3IxU/uzCcp7Y14LXH8Dr13z4slpOdg7xzOF2XjvRxeZFhew83ctHLl8EwNvXlPLFm1aRl5nKezZUnne+e962IqI2x/OiaF37APa0lHDjtXMEvWuIJRNUWseKstx0irLT2Nfk5EOXxtqaxCThKkVni1KKWkdW2FOZCyc6BnlsbwuP720J58qe6BhkWfH0xTpmZbn89I3TnOwaZmDEy/81Qg/H2gf58hOHuXKZg3++ee28rxEvfPjSWoY9fn69u5n3b65mkSOLq1cUUZBl4zd7WnjjZA9evw73NFdK8bGrFvPnm6pikm8fz3NFj7QOsKosO5zdM2AIutaa+s6hcIO2eEApxbrKvKiPkkwmkl7QAWoLs+blof9qRyOpVsWFlbn84UAbPn+Ahq5hlpVM/2Wwp6Xg9et5L/S8eqKL1WU53Lqhkkd2NtE/4uW+3x0i02blO7dfZHo5eyy5oDKXh+/ewp1bqvmHm1YBkGq1cPO6cv54pIPvv1yPw57GxbWRiYnPlrNdNeMr5BIIaOraB1lVlhNulRHy0LuHPPSPeOPKQwe4qCo36Li4ZXj0XFgggp5JS9/InKbOa6159lA7W1cUc/vFVTT2unj+aAcef2BmHrpt/l92nz/A3kYnm2rz+YvLahnx+vngj9/irVO9/P3bV1I4RfuBRGXL4kL+5V0XnFPwctvGSjy+AHsanXxwS014tmysCc8VjbNF0cZeF0OjPlaX5ZCWYiU91RIW9JNdQwBx5aHD2Th6aFFcmB3x8Y2IMEuK7QR0cBFothxtG6S1380Nq0q4YVUJFgVff7oOIDxUYyrMiK8ebRvE5fGzqbaAtRW5vH9zNQea+3nHhWW875LESVOcL2srcvmfv7iYD11aw19cVhtrc84h3oZc+AOaLzx2kBSL4mIjuyc3IzUs6KE5orHqsjgZF1bkAbBPwi5zYkEI+nqjcGIu7XSfORQsctm6sojinHRu3VDJ6R4XG2vyw6lgU2FGx8VdZ3oB2FQT/P/42rsv4OW/3cr/vf2ihM9qmS3XrCzm/lvWRqTicz7E21zRHad6ebOhhy/fvCYcVsnLsIUFPdSGOl4yXELkZqayoiSb16Sd7pxYEIJeVZBBUXYae87MTtC7h0b5yeunuX5VsPIQ4J4blnPV8iLuv2XNjMTUDA9915k+ynPTzxkTVuvImrbjpBA94m2u6MvHO0m1Kt69/myXjpyMlLCg9wx7UIop21DHirevLWXHqV7pvjgHFoQiKKXYWJ3PztO9M24BMOD2cteDO3F7/dx748rw9vK8DH72kUtm3B9lvoOitdbsOt3LpjhZABQmJt5CLq8c62JTTcE5axDZ6anhz2HfsIfcjNS4dApuXFtKQMMz0iN91sTfbzNCXLeqmOa+ER7d1TTh/lGfn4/+dBef+MVuhkd9/NPvDnOodYDv37lxXgtH8y06ae4boWNglE21se+3IUxOlkn1BmYQ6v1+6ZJzq4az01MYNHq59Lo8cRduCbGyNJuVpdk8unPi76owOQtG0N+zoZLNiwr416frJpyU869P1fH80Q6ePdzODd96hcf2tvCpa5Zyw+r5FevMd1D0b/cG+5VctiSxSvoXGtFskzwdLUZ3xZpxMzrHCnrfsIeCKcYnxhKlFHdcXMXBln4OtUi2y2xYMIJusSj+9u0rcLq8/GbPuZNb+oY9PLSjkds3VfHABzfR5/Jy64YKPnPt0nlfdz6Loo09Ln70p1Ncv6qEpTNIkRRiR1aaFVecxNCbjIHVVQXjBT2VQbcXrTW9w55wtXE88u71laRaFb/bZ26//GRnwQg6BLNE1lbknDf67Lsv1TPqC/CRKxZx/eoS9t53A99670WmxBdDrVVn66H3u7zc+v3XAfi7t0e2dF2YP/GU5RLqf16Zf+4Uouz0UJFbgN5hD4VxLOi5malcvbyIPxxoO6ffvTA1C0rQlVLcuLaMQy0DdA646Rx0c9eDO/nRn07x/s3V4Wnw0/VnmQ22lGDRyeAsv+zff+UkPcMefvnRzaZPqRfMJ8sWP3NFm/tGSEuxnDfvNtt4Whxwe+lzxbeHDvDOdeW09bt5syEu+vwlBAtK0AGuXRnsif1CXSefe3Q/r5/s5jPXLeP+m9dE7Jp5man0TzD+azK8/gC/3H6GP7uwnLUVse9VLUxPaK7oiDf2YZfmPhcV+RnnpdVmpwdTFNv73Xj9Om5j6CHevqaUgiwb//P66VibkjAsOEFfWZpNdUEmX33yKK+d6OYf/2w199ywPKLpW3kZNvrG9KGejj1n+hgc9fEOY1CDEP/E01zR5r4RKvMzz9uenR700EON6uLdQ09PtXLn5mpeqOtgx6neWJuTECw4QVdK8fVbL2DY4+O6lcW8/5LqiF8zLzMV58jMPfRXT3SRYlERG5kmmE+o3iDWC6Naa870uKgaFz+Hsx76qe5gozqHPb4FHeDuq5dQU5DJ3zyyL9wLX5icBSfoEJy1+dxnr+K7H9gQldL52YZcXjnexYbqfHLS46+KT5iY+RaQmUWfy0v/iDc82GUsIQ/9rKDHf1M3e1oK971zNS3OEZ6VQqNpWZCCDrCsJNvUxc+pmE3IpXtolEMtA1y9IjpDtAVziJe5oqe6gw3oJmqLG7KxwRD06ebhxgtXLy+muiCTX25vnP7gBc6CFfRokpcVDLnMJAMi1JToqmUi6IlEuGdPjKtFT3YFxXoiDz30xHfKaJ0br5Wi47FaFDeuLWX3mT7ccbDoHM+IoEeBvAwbHl8At/f8GKDb6+ezD+/lsw/vpXPQzbOHOijMss2ok6MQP5xdFI2t4JzqHibVqs7LQQewp4fSFn3kZ6aSGod9XCZjY00+Hn9AKkenYUYzRYX5kWe0eu1zeciwnftF23m6l9/uawVgT6OTxl4Xn9y6JCaj1IS5M9cCMrMIBDSDbh8nOoaoLsicMGvLalFk2awMe/wJET8fywajdfTuM33SqG4KEudPdAKTbwi6c4KF0Z2nerFaFJ/ftpLGXhfpqRb+jzH8WEgcYjko2h/QfPqhPVz81ed5oa4jPGt1IhxG3DxR4uchHPY0Fjmy2DXLFtgLDfHQo0BuRjBW6ZxgYfStU72sLc/hE1uX8M51ZWideF82wZxRg3PlN3uaeepge7gr6OfeNnmriMuWODjT0xh+akwkNtbk81JdJ1rrBTfYZaaIhx4FQl+e8bnoPn+AvU3O8CNkZX7meQ2VhMQgxWohPTU2c0WfOthGdUEmf/ybq3j2s1ed0wN9PFuN7KlQcVEisbEmn55hD6cT0PZoIYIeBULZBD3D53robf1uPL4Ay0via1CvMDdi0UJ3wO3l9fputq0tRSmFdZq1l8uXOgC4KQGrkEMjGHedlqrRyRBBjwIOexpWi6K9f+Sc7U29E7c5FRITe1oKQ+7oCvrLx7rw+jVvm2HffntaCkfv38Ynty6JsGXms6TITkGWjRfrOmNtypxxe/388q0znOgYjMj5RdCjgNWiKMlOo83pPmd7oyHo1SLoSUFORioD7plXBJvBC0c7KMiysb565hOtMmzWhIxBWyyK2zZW8tyRDlqdI9P/QBwyMOLli48fYkeEnjJE0KNEeV4GreM89MZeFykWRVnu+TnDQuKRk54anggUDXz+AC8f6+KaFcXThlqShQ9uqQHgO88fj7Elc8Nj9KOJVA2ACHqUKMvLoK3/fA+9Ij9jwXwZk53s9BQGZtGEbb7sPtNH/4iX61YVR+2asaaqIJOPXrmIR3c1szsBUxg9vqCg20TQE5vy3HTanO5zpq809bok3JJERNtDf6Guk1Sr4spljqhdMx746+uWkWWz8sjOxOvt4vUHv//ioSc4ZbnpePyBczJdGntdsiCaRGSnp0Q1hv7C0Q42LyoMt8VdKGTaUti2toynD7YnXG+XUAtgW0oMBV0ptU0pdUwpVa+UuneK496jlNJKqU3mmZgclOUF4+RtRhx90O2lz+UVDz2JyMlIxeXxR6Vv9+nuYU52DS+ocMtYbt1QweCoj+ePdsTalFlxNoYemTDrtIKulLIC3wVuBFYD71NKrZ7guGzgr4G3zDYyGagtDHa/O2l0umvqDQq7CHryEOo3Ho3UxZCQXbdyZumKycaWxYWU5qTz+J6WWJsyK7xxEEO/BKjXWjdorT3Aw8AtExz3FeAbgHuCfQueJUVZpKdaONg8AEjKYjISak8bjbDL80c7WFZsp7pwYX5+rBbFLevLefl4V8RyuiNB2EOPYcilAmga877Z2BZGKbUBqNJaPznViZRSdyuldimldnV1dc3a2EQmxWphTXkuB1ucgBQVJSMhDz1SC6OBgKZzwM2R1gG2N/TyznXlEblOovCxKxeTk57CPY/uxxXjPvQzxRvvaYtKKQvwLeBz0x2rtX5Aa71Ja72pqGjhDXC4oCKXw60D+AOaxl4XOekp5GYsrAWtZCbH+F1GKnXxh681cMnXXuCm/3yNTJs1nJO9UHHY0/jmbes43NrP5x7dH2tzZoTHF8pyiVEMHWgBqsa8rzS2hcgG1gIvK6VOA1uAJ2Rh9HzWVeXi8vg53NrP6Z5h8c6TjOwxAyTMJhDQ/Hz7GVaX5fCXVy3m325bR36CTByKJDesLuFT1yzl6UPtnDZG68UzIQ89LYYhl53AMqXUIqWUDbgDeCK0U2vdr7V2aK1rtda1wHbgZq31rohYnMBsXV5MikXx2J4Wdp7uZcMsyrWF+CeSMfTtp3po7hvh41uX8IWbVvGOCxOvuVakuHNLDVaL4lc74j8vPeYhF621D/g08CxwFHhUa31YKXW/UurmiFiVpORn2bh8qYMH3ziN2xvg+hk2VBISg0iGXPY2OoGz7W+Fs5TkpHPNiiL+sL91RnN7Y0moUjSmMXSt9VNa6+Va6yVa668a2+7TWj8xwbFbxTufnI9duTj8estiGaWVTGSnpaBUZAS9rn2QyvyM8FOAcC7b1pbR2u/mQHN8zxyNtIcuE4uizBXLHDz3N1cx6PaSlmKNtTmCiVgsiryMVHonmEw1X+raBlhZKoPDJ+OGVSWkWBS/3t3Muqq8WJszKR6j9F96uSQRy0uy2Vgj3nkyUpBlo2/YXA/d7fXT0D3MytJsU8+bTORmpvLei6t4aEdjXOelx0XpvyAIM6Mgy0bvsLkeen3nEP6AZmWZCPpU3HPDcrLTU/joz3ZNOL83HvD6Ylz6LwjCzMnPNF/Q69qDHqeEXKbGYU/jRx/exJkeFw/taJr+B2KA1x9AKSLWMlsEXRBMpNBuMz2GXtc2QFqKhdoFWuY/GzbWFHBxbT7/u7spLjNeRv0BUq2WiE2MEkEXBBPJz7TRN+wxVUyOdQyyvCSblAgtpCUbt22spKFrmMOtA7E25Ty8Ph2xBVEQQRcEUynIsuELaAZHzasWPdo2KAuis+CalcGWwn+q746xJefj9QciFj8HEXRBMJUCoxy/d8icsEvnoJvuoVFWiKDPmOLsdFaUZPOnE/Ep6JHKcAERdEEwlVB/FbPi6PuMCtGL4ji3Oh65fKmDHad7ozpBaiZ4jBh6pBBBFwQTKcgMCnqfSZkue5ucpFgUaytyTTnfQuHWDRV4fAEeibNsF48vIDF0QUgUQiGXHpNCLnsb+1hdnkN6qlQVz4a1FblsXlTAd54/zg9eORk3GS9e8dAFIXFw2NMA6B4enfe5fP4AB5r7WS/hljnxb7etY8viQr7+dB3ffv5ErM0BwOvXpKbIoqggJAQZNitZNivdg/P30I93DOHy+FkvbZbnRHVhJj/68CZuXlfOD145SatzJNYmBRdFxUMXhMTBkZ1G99D8PfS9TX0ArK/Om/e5FipKKf5+2wrQ8He/3k97f2xHHnt8EnIRhITCYTdJ0BudFGTZZJD4PKnMz+Rfb72A1+t7uOzrL/DkgbaY2eKRtEVBSCwcdptJgt7H+qq8iJWJLyTes7GSpz5zJRdU5vGFxw7ErHmXLIoKQoIR9NDnJxj9Li8nu4Yl3GIiq8tz+Motaxhw+3jmUHtMbPD6tFSKCkIi4bCn0efy4DN6X8+Ffc1OAFkQNZkLKnJZ5Mji9weC4+p+9VYjO071Ru364qELQoLhyE5Da+bVRndvYx9KwYWVUlBkJkopbl5Xzhsne/jEL/bwD48f5K6f7qSp1xWV60sMXRASjCJ7sLioax5x9L2NTpYXZ5MtM0RN564rF1GcncYzh9u54+IqAgHN15+ui8q1I522KDNFBcFkwsVFc4yjBwKafU1OblxbaqZZgkFOeio//vDF1HcOcctF5RTabXzv5ZMcaR1gdXlkh4hI2qIgJBiFIUEfnJuHfqpnmP4RryyIRpC1Fbm8a30FSik+esViCrPS+NjPdpmSnTQVXr8WQReERMJhhFzmKg57jQ6LsiAaHfKzbDz4fy6mY8DNN58xP/SyvaGHB14N9pMZ9fklhi4IiYQ9LYW0FMucBX1fUx/2tBSWFNlNtkyYjLUVuXzkikU8uquZxp7gAumPXmvgt3tb5n3un795hm8+c4wBtw+vX5ObEbl1ERF0QTAZpdS8ctH3N/VzQUVuxAYJCxPzwS01ADx7uJ0zPcP8y5NH+ewj+3hoR+O8zlvfOYTPWBcBRNAFIdGYaz+XUZ+fuvYBLqySdMVoU1WQyaqyHJ470s4jO4N91AuzbBMK+tCoj0DgbEvefpeXuvbgDNMn9rfywR+/xb4mJ/6A5lT3MAC7Tgfz3fMyIyfokuUiCBGgyG6jxTn7RlBH2wbx+jXrKvPMN0qYlndcUMq/P3ecI60D3LC6hM2LCviXJ49S3znI0uLgGEC31881//4y1QWZPHL3FixK8Y7/eo3mvhFe/btr+MxDewFYU55LXkYqHqPAbKch6OKhC0KCMdcGXQeMClEpKIoNH7qslsIsGy6vn3tuWM7N68rJsln5yIO7+NQv99De7+b3+1vpGhxl95k+HtrRyJMH22juC7bm/ewje8PnOtExyMmuIQAsCnaeDnbPjKSgi4cuCBHAYU+jd9hDIKCxzCIWvr+pn8IsGxV5GRG0TpiMnPRUvveBDTT1jbCqLJiT/r07N/KF3xzgxbpOjrQNYFGwrNiOL6B5oa6TnPRUirLTKM9NZ0+jE5vVwg2rSzjQ4gwL+qaaAnaIhy4IiYnDbsMf0PTNsqvfgWYn66TDYkzZvLiQ2zZWht9fvbyIN75wHT+76xKa+1yc7Brmi+9YxVXLHGxv6OGlY51csdTBd+5YzxVLHXz62qWsKM2mqXeEgy0DOOw21lScLViKZAx9RoKulNqmlDqmlKpXSt07wf57lFJHlFIHlFIvKKVqzDdVEBIHR/bsq0WHRn3Udw1JuCVOubi2gO9/YCNfe/cFbF1RzBXLinB7Awy6fVy+1MEiRxa/+OhmPnPdMpYVB1NOnz3czpIiO5X5Z3va29MiFxiZ9sxKKSvwXeAGoBnYqZR6Qmt9ZMxhe4FNWmuXUuoTwDeB2yNhsCAkAoVZIUEfZQXZM/qZQy39aI0siMYx168uCb++enkRH7l8Ecc7BrluZfE5x62tCP5R9vgCLCm2nxNCi+TT10z+VFwC1GutGwxjHgZuAcKCrrV+aczx24E7zTRSEBKNouzZV4vKgmhiYUuxcN87V0+4r6ogk7zMVJwuL0uL7FTmR2dNZCYhlwqgacz7ZmPbZNwFPD3RDqXU3UqpXUqpXV1dXTO3UhASjFCDrq5Z9HPZ39xPRV5GuBeMkNiEKn0r8jOoyo/OGEFTF0WVUncCm4B/m2i/1voBrfUmrfWmoqIiMy8tCHFFbkYqqVY1qxh6cEFUvPNk4RvvuZArljq4dEkhORnRSSicyVVagKox7yuNbeeglLoe+CJwtdY6si3LBCHOUUpRmDXzXPS+YQ9NvSN8YLPkEyQLS4vt/OKjm8PvFzuyIt5wbSaCvhNYppRaRFDI7wDeP/YApdR64L+BbVrrTtOtFIQExJE982HRh1r7AbiwQjz0ZOXFv90a8WtMG3LRWvuATwPPAkeBR7XWh5VS9yulbjYO+zfADvyvUmqfUuqJiFksCAnCbKpFD7cG+4BEesCCkNzMKLCjtX4KeGrctvvGvL7eZLsEIeFx2NOoaxuc0bFHWgeoyMsgL9MWYauEZEYqRQUhQjjsafQMj6K1nvbYw6394p0L80YEXRAihMNuw+vXDLh9Ux7n8vho6B5mjQi6ME9E0AUhQpwdFj11HL2ufRCtYXWZCLowP0TQBSFCFBqzRXumyUUPLYiukQwXYZ6IoAtChAj1c+mZxkM/0jpAXmYq5bnp0TBLSGJE0AUhQjhC/VyGp/bQj7T2s7osR1rmCvNGBF0QIkSBkYLYPUU/F601xzuGWFkq8XNh/oigC0KESLFayM9MpWd4ckF3uryMeP1R68YnJDci6IIQQRz2tCkXRVucwVmU5TJyTjABEXRBiCCFdtuUgt5qCLrMEBXMQARdECJIoT2NrimyXFrDHrpkuAjzRwRdECJISXY6nQPuSfe39rtJS7FQkCU9XIT5I4IuCBGkOCeNYY+fodGJy/9bnSOU52VIyqJgCiLoghBBSnKCxUWTeelBQZdwi2AOIuiCEEFKsoNi3TEwcRy91emmPFcWRAVzEEEXhAhSHPLQB8/30L3+AB2DbklZFExDBF0QIkhxTtBD75zAQ2/vd6O1pCwK5iGCLggRJDsthfRUCx0TxNBDKYtlEkMXTEIEXRAiiFKKkpx0Oibo59LaL1WigrmIoAtChCnNSafN8MbH0uoMeu2yKCqYhQi6IESYyvxMmvpc521vdY5QkGUjw2aNgVVCMiKCLggRpqogg46BUUZ9/nO2Sw66YDYi6IIQYaryMwFo6Ts37CI56ILZiKALQoSpKggKetN5gj4iC6KCqYigC0KEqQ4Jeu/ZOPqA28vgqE9CLoKpiKALQoQpzk7DlmI5R9BbZbCFEAFE0AUhwlgsikWFWdR3DoW3iaALkUAEXRCiwMqybOraB8PvW4wcdCn7F8xEBF0QosDK0hxanCP0u7xAMOMlxaJw2NNibJmQTIigC0IUWFWWDUBd+wAAxzsGWVJkx2qRwRaCecxI0JVS25RSx5RS9UqpeyfYn6aUesTY/5ZSqtZ0SwUhgVlVlgPA4dagoB9tGwiLvCCYxbSCrpSyAt8FbgRWA+9TSq0ed9hdQJ/WeinwbeAbZhsqCIlMcXYa1QWZvF7fjdPloa3fHRZ5QTCLmXjolwD1WusGrbUHeBi4ZdwxtwA/NV7/GrhOyZBEQQijlOLalcW8frKbvU1OABF0wXRmIugVQNOY983GtgmP0Vr7gH6g0AwDBSFZ2LqiCLc3wDefOYZSsLpcBF0wl6guiiql7lZK7VJK7erq6ormpQUh5ly+1MG6ylyOtg3wgc3VkuEimM5MBL0FqBrzvtLYNuExSqkUIBfoGX8irfUDWutNWutNRUVFc7NYEBKUVKuFH3xwIx+/egl/9/aVsTZHSEJmIug7gWVKqUVKKRtwB/DEuGOeAD5svL4NeFFrrc0zUxCSg7LcDO69cSW5GamxNkVIQlKmO0Br7VNKfRp4FrACP9FaH1ZK3Q/s0lo/AfwY+LlSqh7oJSj6giAIQhSZVtABtNZPAU+N23bfmNdu4M/NNU0QBEGYDVIpKgiCkCSIoAuCICQJIuiCIAhJggi6IAhCkiCCLgiCkCSIoAuCICQJKlb1P0qpLuDMHH/cAXSbaE6kEXsjRyLZCollbyLZCgvH3hqt9YSl9jET9PmglNqltd4UaztmitgbORLJVkgsexPJVhB7QUIugiAISYMIuiAIQpKQqIL+QKwNmCVib+RIJFshsexNJFtB7E3MGLogCIJwPonqoQuCIAjjEEEXBEFIEhJO0JVS25RSx5RS9Uqpe2Ntz3iUUqeVUgeVUvuUUruMbQVKqT8qpU4Y/82PoX0/UUp1KqUOjdk2oX0qyH8a9/qAUmpDnNj7T0qpFuMe71NK3TRm3xcMe48ppd4eZVurlFIvKaWOKKUOK6X+2tgel/d3Cnvj7v4qpdKVUjuUUvsNW//Z2L5IKfWWYdMjxhAelFJpxvt6Y39ttGydxt4HlVKnxtzbi4zt5nwWtNYJ84/ggI2TwGLABuwHVsfarnE2ngYc47Z9E7jXeH0v8I0Y2ncVsAE4NJ19wE3A04ACtgBvxYm9/wT87QTHrjY+E2nAIuOzYo2irWXABuN1NnDcsCku7+8U9sbd/TXukd14nQq8ZdyzR4E7jO0/AD5hvP4k8APj9R3AI1G+t5PZ+yBw2wTHm/JZSDQP/RKgXmvdoLX2AA8Dt8TYpplwC/BT4/VPgXfFyhCt9asEp0qNZTL7bgF+poNsB/KUUmVRMdRgEnsn4xbgYa31qNb6FFBP8DMTFbTWbVrrPcbrQeAoUEGc3t8p7J2MmN1f4x4NGW9TjX8auBb4tbF9/L0N3fNfA9cppVQ0bIUp7Z0MUz4LiSboFUDTmPfNTP0BjAUaeE4ptVspdbexrURr3Wa8bgdKYmPapExmXzzf708bj6Y/GRPCiht7jUf89QQ9s7i/v+PshTi8v0opq1JqH9AJ/JHgE4JTa+2bwJ6wrcb+fqAwWrZOZK/WOnRvv2rc228rpdLG22swp3ubaIKeCFyhtd4A3Ah8Sil11didOvh8Fbe5ovFun8H3gSXARUAb8B8xtWYcSik78Bvgs1rrgbH74vH+TmBvXN5frbVfa30RUEnwyWBlbC2amvH2KqXWAl8gaPfFQAHweTOvmWiC3gJUjXlfaWyLG7TWLcZ/O4HHCX7wOkKPT8Z/O2Nn4YRMZl9c3m+tdYfxZQkAP+TsY3/M7VVKpRIUx19qrR8zNsft/Z3I3ni+v4Z9TuAl4FKCoYnQbOSx9oRtNfbnAj3RtTTIGHu3GWEurbUeBf4Hk+9togn6TmCZsbJtI7jY8USMbQqjlMpSSmWHXgNvAw4RtPHDxmEfBn4XGwsnZTL7ngA+ZKzAbwH6x4QOYsa42OK7Cd5jCNp7h5HhsAhYBuyIol0K+DFwVGv9rTG74vL+TmZvPN5fpVSRUirPeJ0B3EAw5v8ScJtx2Ph7G7rntwEvGk9HUWESe+vG/GFXBOP9Y+/t/D8L0Vz5NeMfwdXg4wTjZ1+MtT3jbFtMMAtgP3A4ZB/B2N0LwAngeaAghjY+RPAx2kswTnfXZPYRXHH/rnGvDwKb4sTenxv2HDC+CGVjjv+iYe8x4MYo23oFwXDKAWCf8e+meL2/U9gbd/cXuBDYa9h0CLjP2L6Y4B+VeuB/gTRje7rxvt7YvzjK93Yye1807u0h4BeczYQx5bMgpf+CIAhJQqKFXARBEIRJEEEXBEFIEkTQBUEQkgQRdEEQhCRBBF0QBCFJEEEXBEFIEkTQBUEQkoT/D7yoPCl/cBt1AAAAAElFTkSuQmCC",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "id = 1000\n",
    "plot_sample(samples[id])\n",
    "plot_sample(decoded[id])\n",
    "plot_sample(encoded[id])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "def encode_input(encoder,x_input):\n",
    "    transformed = encoder.predict(x_input)\n",
    "    return np.array(transformed)\n",
    "\n",
    "\n",
    "x_encoded = encode_input(encoder,x_autoencoder)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [],
   "source": [
    "import seaborn as sns \n",
    "def plot_confusion_matrix(y_true, y_pred):\n",
    "  \"\"\"\n",
    "    :param ndarray y_true: contains the ground truth labels in the shape (n_samples,)\n",
    "    :param ndarray y_pred: contains the predicted labels in the shape (n_samples,)\n",
    "  \"\"\"\n",
    "  #y_true = np.argmax(y_true,axis=1) already 1dimensional\n",
    "  #y_pred = np.argmax(y_pred,axis=1)\n",
    "  \n",
    "  con_mat = tf.math.confusion_matrix(labels=y_true, predictions=y_pred).numpy()\n",
    "  con_mat_norm = np.around(con_mat.astype('float') / con_mat.sum(axis=1)[:, np.newaxis], decimals=2)\n",
    "  con_mat_df = pd.DataFrame(con_mat_norm,\n",
    "                  index = classes, \n",
    "                  columns = classes)\n",
    "  figure = plt.figure(figsize=(8, 8))\n",
    "  sns.heatmap(con_mat_df, annot=True,cmap=plt.cm.Blues)\n",
    "  plt.tight_layout()\n",
    "  plt.ylabel('True label')\n",
    "  plt.xlabel('Predicted label')\n",
    "  plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "# redefine input to take just the first lead \n",
    "def create_input(df):\n",
    "    \n",
    "    x = list()\n",
    "    y = list()\n",
    "    \n",
    "    for id in tqdm(df.index.tolist()):\n",
    "        row = df.loc[id]\n",
    "        x.append(np.transpose(np.asarray([row['first_lead']]).astype(np.float32)))\n",
    "        y.append(row['label'])\n",
    "\n",
    "    x = np.asarray(x).astype(np.float32)\n",
    "    y = np.asarray(y).astype(np.float32)\n",
    "    \n",
    "    return x,y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 44955/44955 [00:02<00:00, 17699.04it/s]\n",
      "100%|██████████| 23659/23659 [00:01<00:00, 19221.19it/s]\n",
      "100%|██████████| 24866/24866 [00:01<00:00, 19475.72it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(44955, 40)\n",
      "[1. 0. 0.]\n",
      "Sv trained..\n",
      "Scoring:\n"
     ]
    }
   ],
   "source": [
    "from sklearn.svm import SVC\n",
    "\n",
    "\n",
    "x_train,y_train = create_input(training_patches_df)\n",
    "x_valid,y_valid = create_input(validation_patches_df)\n",
    "x_test,y_test = create_input(test_patches_df)\n",
    "\n",
    "\n",
    "x_train = encode_input(encoder,x_train)\n",
    "x_valid = encode_input(encoder,x_valid)\n",
    "x_test = encode_input(encoder,x_test)\n",
    "\n",
    "print(x_train.shape)\n",
    "print(y_train[0])\n",
    "\n",
    "y_train = np.argmax(y_train,axis=1)\n",
    "y_valid = np.argmax(y_valid,axis=1)\n",
    "y_test = np.argmax(y_test,axis=1)\n",
    "\n",
    "\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "\n",
    "sv = SVC(C=1, kernel='rbf', degree=3,   # C is a regularization parameter -> the higher and the more we weights points outside the margin \n",
    "         gamma='scale', coef0=0.0, shrinking=True, \n",
    "         probability=False, tol=0.001, cache_size=200, \n",
    "         class_weight=None, verbose=False, max_iter=- 1, \n",
    "         decision_function_shape='ovr', break_ties=False, \n",
    "         random_state=SEED)\n",
    "\n",
    "print(\"training ...\",end=\"\\r\")\n",
    "sv.fit(x_train,y_train)\n",
    "print(\"Sv trained\",end=\"\\n\")\n",
    "print(\"Scoring:\")\n",
    "#sv.score(x_valid[:,:,0],y_valid)\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Validation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       0.99      0.94      0.97     20924\n",
      "           1       0.56      0.86      0.68      1032\n",
      "           2       0.70      0.92      0.79      1703\n",
      "\n",
      "    accuracy                           0.94     23659\n",
      "   macro avg       0.75      0.91      0.81     23659\n",
      "weighted avg       0.95      0.94      0.94     23659\n",
      "\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import classification_report\n",
    "y_pred = sv.predict(x_valid)\n",
    "print(classification_report(y_valid, y_pred))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 576x576 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_confusion_matrix(y_valid,y_pred)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h3>Save the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pickle\n",
    "\n",
    "path = os.path.join(os.getcwd(),\"models\",\"svc\")\n",
    "\n",
    "pkl_filename = \"svc.pkl\"\n",
    "with open(os.path.join(path,pkl_filename), 'wb') as file:\n",
    "    pickle.dump(sv, file)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2>KNN</h2>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "import seaborn as sns \n",
    "def plot_confusion_matrix(y_true, y_pred):\n",
    "  \"\"\"\n",
    "    :param ndarray y_true: contains the ground truth labels in the shape (n_samples,)\n",
    "    :param ndarray y_pred: contains the predicted labels in the shape (n_samples,)\n",
    "  \"\"\"\n",
    "  y_true = np.argmax(y_true,axis=1)\n",
    "  y_pred = np.argmax(y_pred,axis=1)\n",
    "  \n",
    "  con_mat = tf.math.confusion_matrix(labels=y_true, predictions=y_pred).numpy()\n",
    "  con_mat_norm = np.around(con_mat.astype('float') / con_mat.sum(axis=1)[:, np.newaxis], decimals=2)\n",
    "  con_mat_df = pd.DataFrame(con_mat_norm,\n",
    "                  index = classes, \n",
    "                  columns = classes)\n",
    "  figure = plt.figure(figsize=(8, 8))\n",
    "  sns.heatmap(con_mat_df, annot=True,cmap=plt.cm.Blues)\n",
    "  plt.tight_layout()\n",
    "  plt.ylabel('True label')\n",
    "  plt.xlabel('Predicted label')\n",
    "  plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 44955/44955 [00:02<00:00, 21401.60it/s]\n",
      "100%|██████████| 23659/23659 [00:01<00:00, 22998.92it/s]\n",
      "100%|██████████| 24866/24866 [00:01<00:00, 23181.63it/s]\n"
     ]
    }
   ],
   "source": [
    "x_train,y_train = create_input(training_patches_df)\n",
    "x_valid,y_valid = create_input(validation_patches_df)\n",
    "x_test,y_test = create_input(test_patches_df)\n",
    "\n",
    "\n",
    "x_train = encode_input(encoder,x_train)\n",
    "x_valid = encode_input(encoder,x_valid)\n",
    "x_test = encode_input(encoder,x_test)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "KNN trained.\n"
     ]
    }
   ],
   "source": [
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "\n",
    "KNN = KNeighborsClassifier(n_neighbors=3,weights='distance',)\n",
    "print(\"training ...\",end=\"\\r\")\n",
    "KNN.fit(x_train,y_train)\n",
    "print(\"KNN trained\",end=\"\\n\")\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "validation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       0.99      0.92      0.96     20924\n",
      "           1       0.49      0.79      0.60      1032\n",
      "           2       0.57      0.84      0.68      1703\n",
      "\n",
      "   micro avg       0.91      0.91      0.91     23659\n",
      "   macro avg       0.68      0.85      0.75     23659\n",
      "weighted avg       0.94      0.91      0.92     23659\n",
      " samples avg       0.91      0.91      0.91     23659\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\simon\\anaconda3\\envs\\tf-gpu\\lib\\site-packages\\sklearn\\metrics\\_classification.py:1308: UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in samples with no predicted labels. Use `zero_division` parameter to control this behavior.\n",
      "  _warn_prf(average, modifier, msg_start, len(result))\n"
     ]
    }
   ],
   "source": [
    "y_pred = KNN.predict(x_valid)\n",
    "print(classification_report(y_valid, y_pred))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 576x576 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_confusion_matrix(y_valid, y_pred)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2>RFC"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 44955/44955 [00:02<00:00, 20846.21it/s]\n",
      "100%|██████████| 23659/23659 [00:01<00:00, 20893.70it/s]\n",
      "100%|██████████| 23659/23659 [00:01<00:00, 21769.06it/s]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "RandomForestClassifier(n_estimators=200, random_state=123)"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.ensemble import RandomForestClassifier\n",
    "\n",
    "x_train,y_train = create_input(training_patches_df)\n",
    "x_valid,y_valid = create_input(validation_balanced_patches_df)\n",
    "x_test,y_test = create_input(validation_patches_df)\n",
    "\n",
    "x_train = encode_input(encoder,x_train)\n",
    "x_valid = encode_input(encoder,x_valid)\n",
    "x_test = encode_input(encoder,x_test)\n",
    "\n",
    "\n",
    "rf = RandomForestClassifier(\n",
    "    n_estimators= 200,\n",
    "    random_state= SEED,\n",
    ")\n",
    "\n",
    "rf.fit(x_train,y_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Validation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       1.00      0.84      0.91     20924\n",
      "           1       0.84      0.66      0.74      1032\n",
      "           2       0.73      0.74      0.74      1703\n",
      "\n",
      "   micro avg       0.97      0.83      0.89     23659\n",
      "   macro avg       0.85      0.75      0.80     23659\n",
      "weighted avg       0.97      0.83      0.89     23659\n",
      " samples avg       0.83      0.83      0.83     23659\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\simon\\anaconda3\\envs\\tf-gpu\\lib\\site-packages\\sklearn\\metrics\\_classification.py:1308: UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in samples with no predicted labels. Use `zero_division` parameter to control this behavior.\n",
      "  _warn_prf(average, modifier, msg_start, len(result))\n"
     ]
    }
   ],
   "source": [
    "y_pred = rf.predict(x_valid)\n",
    "print(classification_report(y_valid, y_pred))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAicAAAJGCAYAAACA+CUiAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/YYfK9AAAACXBIWXMAAAsTAAALEwEAmpwYAAAuH0lEQVR4nO3dd5hcZdnH8e+9u0kIppPsplICoYTeiy+hSS8BbFix5sVXULGCKGgUFBBFpAZBQUWkCAQIJkoXgRDpCQKhppAGIRBKyu7z/rGTZTcku0tkz5wz+/14zXXtOfPMmeesc2Vvfuc+z0RKCUmSpLyoKvcEJEmSmrM4kSRJuWJxIkmScsXiRJIk5YrFiSRJypWack9gdbpve6y3EWmNLXzg3HJPQQW1vN5/erRmenSLyOq9svwb+dZD52Z2XiuYnEiSpFzJbXIiSZJWIyo7W6jss5MkSYVjcSJJknLFyzqSJBVNdr23ZWFyIkmScsXkRJKkorEhVpIkKTsmJ5IkFY09J5IkSdkxOZEkqWjsOZEkScqOyYkkSUVjz4kkSVJ2TE4kSSoae04kSZKyY3IiSVLR2HMiSZKUHZMTSZKKxp4TSZKk7JicSJJUNPacSJIkZcfkRJKkorHnRJIkKTsWJ5IkKVe8rCNJUtHYECtJkpQdkxNJkorGhlhJkqTsmJxIklQ0JieSJEnZMTmRJKloqrxbR5IkKTMmJ5IkFY09J5IkSdkxOZEkqWhcIVaSJCk7JieSJBWNPSeSJEnZMTmRJKlo7DmRJEnKjsmJJElFY8+JJElSdkxOJEkqGntOJEmSsmNxIkmScsXLOpIkFY0NsZIkSdkxOZEkqWhsiJUkScqOyYkkSUVjz4kkSVJ2TE4kSSoae04kSZKyY3IiSVLR2HMiSZKUHZMTSZKKxuREkiQpOyYnkiQVjXfrSJIkZcfkRJKkorHnRJIkKTsmJ5IkFY09J5IkSdkxOZEkqWjsOZEkScqOxYkkScoVL+tIklQ0NsRKkiRlx+REkqSCCZMTSZKk7JicSJJUMCYnkiRJGTI5kSSpaCo7ODE5kSRJ+WJyIklSwdhzIkmSlCGTE0mSCsbkRJIkKUMmJ5IkFYzJiSRJUoZMTiRJKhiTE0mSpAyZnEiSVDSVHZyYnEiSpHyxOMmZfXfbjEeu+yGP33AK3/78vu96ft1BfZlw4XFM/suJTLz46wyp7dP03KlfH82/rzmJh679AWd99yMZzlpFcc/dd3HYwftzyAH7csnF48o9HeXIv/55N0ceegCjD96P313y7s/G0qVLOeE7xzP64P347Cc/xuxZMwG47957+NTHj+RjRx7Kpz5+JJPvvy/rqXdKEZHZoxwsTnKkqio4+4SPMfrY89n2wz/lowdsz6bDB7YY87Pjj+BPN09mp4//jNPG3cLY4w4DYJetN2DXbYaz48dOY/uPnsr2m6/H7tuPKMdpKKfq6+s57dSxnH/hb7lu/M38bcJNPDN9ermnpRyor6/n56eN5ZwLLuaa629i4i038+wzLT8b1//1Gnr16sUNN0/iU585mnPOPguAPn36cvZvLuCqv97Ij3/6c04+6bvlOAVVGIuTHNlxi/V5ZsYCnp/1MsuW13P1xAc5ZM+tWozZdPgg7pz8JAB3PvAUh+y5JQApQbeuXejapYZuXWuoqalm3iuvZX4Oyq/HH3uUYcPWY+iwYXTp2pUDDjqYO26/tdzTUg5MffxRhq27LkOHDqNLl67sd8BB7/ps3HnHrRxy2OEA7LPv/ky+/15SSmy62UgG1NYBsOFGI1jy9hKWLl2a9Smowlic5Mjg2t7MnLuwaXvW3IUMGdC7xZjHnprF6L23AWD03lvTq0d3+vX+APc/+hx3TXma5/5+Ks9NOo1//OsJnnxubpbTV87NmzuXgYPeSeJq6+qYO9fPiBo/G3V1g5q26+oGMn9ey8/G/LnzmsbU1NTQo0dPXn311RZjbv37RDbdbCRdu3bt8Dl3dpV+WadD7taJiJNbeTqllH6ymteNAcYA1Azdk5r+m3fE9ArtxF9dx6++91E+fdjO3PPgdGbNXUh9fQPDh/Vnkw3q2Gj/HwBw84XH8cF/bcg9Dz1T5hlL6gyemf4055x9FudddEm5p6IK0FG3Er+xin1rA18C1gFWWZyklMYB4wC6b3ts6qC55dbseYsYWte3aXtIXV9mzV/UYsxL8xdx1Ld/C8AHunfl8H22YdHit/jCkbsx+bHneeOtxjh14j1T2XmrDSxO1KS2ro45L81p2m78r+W6Ms5IedGYor3UtD137pymSzUrDKirZe7cl6gbOJDly5ezePHr9OnTp3H8nDl8+/hjGXvq6Qwbtm6WU++0XIRtDaSUzlrxoLHY6A58AbgSGN4R71kJpkx9gY3WHcB6g9ehS001H91/O26+49EWY9bp84GmD+V3vrA/l93Q2Bk/Y85Cdt9+I6qrq6ipqWL37Ubwn+fmvOs91HltvsWWvPji88ycOYNlS5fytwk3s8dee5d7WsqBkZtvyYwXXmDWzJksW7aUSX+bwB57tvxs7LHn3tw0/nqg8fLNjjvtQkTw+muv8fVj/5fjvv4tttl2uzLMXpWowxZhi4h+wDeBTwGXAdullBa2/qrOrb6+geNPv4obz/8q1VXBZTfcxxPPzuGHXzmYB6e9yM13PsaoHUYw9rjDSAn++eB0vvGzqwD46z8eYo8dN2bKVd8nkfj7v55gwl2Pl/mMlCc1NTWceNLJfGXMl2hoqOfwIz7MRht5R5caPxvf/f4POfYrX6S+voHRh3+YDTcawQXnncPIkVuwx157M/qIj/DD73+X0QfvR+/evTntjF8C8Jcr/8SMF1/k4ovO5+KLzgfgvAsvod8665TzlCpepScnkdL7f/UkIs4EjqQxNTkvpbT4vR6jM17W0ftn4QPnlnsKKqjl9f7TozXTo1t2FcM6n/1zZh/Uly//ROaVUEfdrfMtYDDwA2B2RLxWerweEd7fKknSfyMyfJRBR/WcVKWUuqeUeqaUejV79Ewp9eqI95QkSdmLiAMi4smImB4RJ6zi+XUj4vaIeCgiHo2Ig9o6pl/8J0lSweSl5yQiqoHzgH2BmcADETE+pTSt2bAfAFellC6IiJHABGD91o7rImySJGlN7QRMTyk9m1JaSuNduaNXGpOAFVdNegOz2zqoyYkkSQWTZXLSfIHUknGldckAhgAzmj03E9h5pUP8CJgUEccBHwA+1NZ7WpxIkqTVar5A6hr6BPD7lNJZEbEr8IeI2CKl1LC6F1icSJJUMHnpOQFmAcOabQ8t7Wvui8ABACmleyNiLaA/MG91B7XnRJIkrakHgBERsUFEdAWOAsavNOZFYB+AiNgMWAuY39pBTU4kSSqanAQnKaXlEXEsMBGoBi5NKU2NiLHAlJTSeBrXPrs4Io6nsTn2c6mNFWAtTiRJ0hpLKU2g8fbg5vtObvbzNOCD7+WYFieSJBVMjnpOOoQ9J5IkKVdMTiRJKhiTE0mSpAxZnEiSpFzxso4kSQXjZR1JkqQMmZxIklQwJieSJEkZMjmRJKloKjs4MTmRJEn5YnIiSVLB2HMiSZKUIZMTSZIKxuREkiQpQyYnkiQVjMmJJElShkxOJEkqmsoOTkxOJElSvpicSJJUMPacSJIkZcjkRJKkgjE5kSRJypDJiSRJBWNyIkmSlCGLE0mSlCte1pEkqWC8rCNJkpQhkxNJkoqmsoMTkxNJkpQvJieSJBWMPSeSJEkZMjmRJKlgTE4kSZIyZHIiSVLBVHhwYnIiSZLyxeREkqSCsedEkiQpQyYnkiQVTIUHJyYnkiQpX0xOJEkqGHtOJEmSMmRyIklSwVR4cGJyIkmS8sXkRJKkgqmqquzoxOREkiTlisWJJEnKFS/rSJJUMDbESpIkZcjkRJKkgnERNkmSpAyZnEiSVDAVHpyYnEiSpHwxOZEkqWDsOZEkScqQyYkkSQVjciJJkpQhkxNJkgqmwoMTkxNJkpQvJieSJBWMPSeSJEkZMjmRJKlgKjw4MTmRJEn5YnIiSVLB2HMiSZKUIZMTSZIKpsKDE5MTSZKULxYnkiQpV7ysI0lSwdgQK0mSlCGTE0mSCqbCgxOTE0mSlC8mJ5IkFYw9J5IkSRkyOZEkqWAqPDjJb3Hyj6t+Uu4pqMBGnXFHuaeggrr+/3Yr9xRUUD26dS33FCpGbosTSZK0avacSJIkZcjkRJKkgqnw4MTkRJIk5YvJiSRJBWPPiSRJUoZMTiRJKpgKD05MTiRJUr6YnEiSVDD2nEiSJGXI5ESSpIIxOZEkScqQxYkkScoVL+tIklQwFX5Vx+REkiTli8mJJEkFY0OsJElShkxOJEkqmAoPTkxOJElSvpicSJJUMPacSJIkZcjkRJKkgqnw4MTkRJIk5YvJiSRJBVNV4dGJyYkkScoVixNJkgomIrtH23OJAyLiyYiYHhEnrGbMxyJiWkRMjYgr2jqml3UkSdIaiYhq4DxgX2Am8EBEjE8pTWs2ZgRwIvDBlNLCiKht67gWJ5IkFUyO1jnZCZieUnoWICKuBEYD05qN+TJwXkppIUBKaV5bB/WyjiRJWq2IGBMRU5o9xjR7eggwo9n2zNK+5jYGNo6IeyLivog4oK33NDmRJKlgqjIMTlJK44Bx/8UhaoARwJ7AUOCuiNgypfTq6l5gciJJktbULGBYs+2hpX3NzQTGp5SWpZSeA56isVhZLYsTSZIKJiIye7ThAWBERGwQEV2Bo4DxK425nsbUhIjoT+NlnmdbO6jFiSRJWiMppeXAscBE4AngqpTS1IgYGxGHlYZNBF6OiGnA7cB3Ukovt3Zce04kSdIaSylNACastO/kZj8n4JulR7tYnEiSVDD5uZO4Y3hZR5Ik5YrJiSRJBRNUdnRiciJJknLF5ESSpILJchG2cjA5kSRJuWJyIklSweToi/86hMmJJEnKFZMTSZIKpsKDE5MTSZKULyYnkiQVTFWFRycmJ5IkKVdMTiRJKpgKD05MTiRJUr6YnEiSVDCucyJJkpQhkxNJkgqmwoMTkxNJkpQvJieSJBWM65xIkiRlyOJEkiTlipd1JEkqmMq+qGNyIkmScsbkRJKkgnERNkmSpAyZnEiSVDBVlR2cmJxIkqR8MTmRJKlg7DmRJEnKkMmJJEkFU+HBicmJJEnKF5MTSZIKptJ7TlZbnETEb4C0uudTSl/rkBlJkqROrbXkZEpms5AkSe1W6eucrLY4SSld1nw7ItZOKb3Z8VOSJEmdWZsNsRGxa0RMA/5T2t46Is7v8JlJkqRViojMHuXQnrt1zgb2B14GSCk9AozqwDlJkqROrF1366SUZqxUPdV3zHQkSVJbKrzlpF3FyYyI2A1IEdEF+DrwRMdOS5IkdVbtKU6OAX4NDAFmAxOBr3bkpCRJ0upVddZ1TlZIKS0APpXBXCRJktp1t87wiLgxIuZHxLyIuCEihmcxOUmS1Pm0526dK4CrgEHAYOBq4M8dOSlJkrR6Edk9yqE9xcnaKaU/pJSWlx5/BNbq6IlJkqTOqbXv1ulX+vGWiDgBuJLG79r5ODAhg7lJkqRV6LRf/Af8m8ZiZMVv4H+bPZeAEztqUpIkqfNq7bt1NshyIpIkqX0qPDhp3wqxEbEFMJJmvSYppcs7alKSJKnzarM4iYhTgD1pLE4mAAcC/wQsTiRJKoNKX4StPXfrfATYB5iTUvo8sDXQu0NnJUmSOq32XNZ5K6XUEBHLI6IXMA8Y1sHz6rQe+/e9XDHuV6SGBnbf7zAO/uhnWzw/8boruGvSeKqrq+nZqy+f/8ZJ9K8dBMAXD9uNoettCMA6A+r42sm/yHz+Kq9dhvfjW/tuRFUENzzyEpff++K7xnxoswF8aff1IcHT8xbzwxsavyqrrlc3TjpoE+p6dSMBx//lMV5a9Ha2J6CymXzvPzn3l6dT31DPwYcdySeP/lKL55cuXcrPfvx9nvrPNHr17sMpPz2TgYOHsGzZMn75sx/z5H+mElHFcd88gW2237FMZ9F5VHhw0q7iZEpE9AEupvEOnsXAvR05qc6qob6eP17wC77103Pot04tY4//PNvsvDtD1n2nN3ndDTfh5F/9nm5rrcXtE67l6t+dy1e+dyoAXbt248e/+UO5pq8yqwr47v4jOPbPjzDvtSVc9vntufvpBTy34M2mMcP6dufoXdfly5c/xOtvL6fv2l2anvvRoZvxu3teYPLzC+nepZqGlMpxGiqD+vp6fn3mqZz5m3EMqB3IMZ87it1234v1h2/YNGbC+L/Ss2cv/nTtBG6bdAsXnfcrTjn1F9x0/TUAXHrFdSx85WW+942vcOHvr6Sqqj3BvLRqbX56Ukr/l1J6NaV0IbAvcHTp8o7eZ88+NY3aQUOpHTiEmi5d2HnUvjx8310txmy21fZ0W6uxL3n4JluwcMG8ckxVObT54F7MXPgWs199m+UNiUnT5jFqRP8WYw7fZhDX/Hs2r7+9HICFby4DYIP+a1NdFUx+fiEAby2rZ8nyhmxPQGXzn2mPMXjougweMowuXbqw974Hcs9dt7cYc89dt7P/wYcBsMfe+/LgA/eTUuKF555h2x12BqBvv3Xo0bMXTz4xNfNz6GwiIrNHOay2OImI7VZ+AP2AmtLPqxURO0bEwGbbny19J885zRZ300pefXk+/QbUNm337V/Lwpfnr3b83ZNuZMvtd23aXrZ0KT/+xuf46be+yIP33tmhc1X+DOjZjbmvLWnanvf6Egb07NZizLr91mbdft25+DPbcsnR27HL8H5N+xe/vZzTP7w5f/jC9hy393CqKjw21jsWzJtHbV3TP9kMqK1jwfy5LcfMn0dtbeOY6poaevTowWuLXmXDEZvwr7tvp375cl6aPZOn/jONeXPnZDp/VZ7WLuuc1cpzCdi7lecvAj4EEBGjgJ8DxwHbAONobLJ9l4gYA4wB+M7YXzL6qM+18had272338Lz05/gez+/oGnfmZdeR9/+tcybM4szv/9Vhq6/IbWDhpZxlsqb6qpgWL/uHPOnh6nr2Y2LPrMNn7h4CtVVwTbDevPpS6cwd9ESTj1iJIdsNZDxj/hHRq076NAjePH5Z/nfzx1F3cBBbLHl1lRXe0mno1X6b7i1Rdj2+i+OW51SeqX088eBcSmla4FrI+LhVt5zHI3FC/c8vbDTXfDus84AXpn/zmWahQvm0XedAe8aN/Xhydz0l9/zvZ9fQJcuXZv29+3fmLrUDhzCpltux4vPPGVx0onMf30Jdb3eSUpqe3Zj/utLWoyZ9/oSHp/9GvUNidmL3ubFV95iWL/uzHttCU/NW8zsVxsbYO98agFbDOkFFiedQv/a2hZpx/x5c+k/oK7lmAG1zJs3hwF1A6lfvpzFixfTq3cfIoKvHv+9pnHHfunTDB22flZTV4XqqOKrOiJWFD77ALc1e65dC791RhtsvBlzZ89g/pzZLF+2jPvv+jvb7Lx7izEvPPMkl597Ol/74Zn06vPOFbI3Fr/GsmVLAXh90as8Pe1RBq3rIr+dybTZrzOsb3cG916Lmqpgv5G13P30ghZj7nhqAduv2weA3t27sG6/7sx+9S2mvfQaPbvV0KfUILvDen14bsEbWZ+CymTTzbZg1owXeGn2TJYtW8Ztf7+F3Ubt2WLMbrvvycSbxwNw521/Z9sddiIiePvtt3jrrcam6yn3/4vq6uoWjbTqGJXec9JRhcKfgTsjYgHwFnA3QERsBCzqoPcsvOrqGj59zLf55clfp6Ghgf/Z9xCGrDec6/44jvVHbMq2O4/iqkt/w5K33+T8n58EvHPL8Esznueyc08nIkgpcdBHP9viLh9VvvqUOHPS05xz1FZUVQU3PvISzy54kzGj1ueJl17n7qdf5r5nX2GXDfpy5ZgdaWhInHPbsyx6q7E59te3PsN5n9yaAP4zZzHXP/RSeU9ImamuqeFr3/4+3/3aMTQ01HPgoUewwfCNuPSic9lks8354Ki9OPiwIzntRyfyqQ8fRK9evfnhT88A4NVXXuG7Xz+GqAr6D6jlxB/9rMxno0oQqYNuF4yIXYBBwKSU0hulfRsDPVJKD7b1+s54WUfvn+OvfqTcU1BBXf9/u5V7CiqowX26ZhYzfOOG/2T2N/Ls0ZtmHp+0Z/n6AD4FDE8pjY2IdYGBKaXJrb0upXTfKvY9tcYzlSRJnUJ7LuucDzTQeHfOWOB14FrAJQAlSSqDSr/Vvz3Fyc4ppe0i4iGAlNLCiOja1oskSZLWRHuKk2URUU3j2iZExAAakxRJklQG5bqLJivtuZX4HOA6oDYiTgX+CZzWobOSJEmdVpvJSUrpTxHxbxrXKwng8JTSEx0+M0mS1Cm1526ddYE3gRub70spvfu72CVJUoezIRZuprHfJIC1gA2AJ4HNO3BekiSpk2rPZZ0tm2+XvpH4/zpsRpIkqVUV3g/73r9bp7S6684dMBdJkqR29Zx8s9lmFbAdMLvDZiRJklpVVeHRSXt6Tno2+3k5jT0o13bMdCRJUmfXanFSWnytZ0rp2xnNR5IkteE992QUzGrPLyJqUkr1wAcznI8kSerkWktOJtPYX/JwRIwHrgbeWPFkSumvHTw3SZK0ChXectKunpO1gJdp/FbiFeudJMDiRJIkve9aK05qS3fqPM47RckKqUNnJUmSVqsz361TDfSgZVGygsWJJEnqEK0VJy+llMZmNhNJktQuFR6ctHo3UoWfuiRJyqPWkpN9MpuFJElqt0r/VuLVJicppVeynIgkSRK071ZiSZKUI5V+t06lr4ArSZIKxuREkqSCqfDgxOREkiTli8WJJEnKFS/rSJJUMJ32VmJJkqRyMDmRJKlgosIXcTc5kSRJuWJyIklSwdhzIkmSlCGTE0mSCsbkRJIkKUMmJ5IkFUxU+Pr1JieSJClXTE4kSSoYe04kSZIyZHIiSVLBVHjLicmJJEnKF4sTSZIKpiois0dbIuKAiHgyIqZHxAmtjPtwRKSI2KHN83uPvw9JkiQAIqIaOA84EBgJfCIiRq5iXE/g68D97TmuxYkkSQVTFdk92rATMD2l9GxKaSlwJTB6FeN+ApwOvN2u83sPvwtJktTJRMSYiJjS7DGm2dNDgBnNtmeW9jV//XbAsJTSze19T+/WkSSpYLK8WyelNA4YtyavjYgq4JfA597L60xOJEnSmpoFDGu2PbS0b4WewBbAHRHxPLALML6tpliLE0mStKYeAEZExAYR0RU4Chi/4smU0qKUUv+U0voppfWB+4DDUkpTWjuol3UkSSqYKvKxCltKaXlEHAtMBKqBS1NKUyNiLDAlpTS+9SOsmsWJJElaYymlCcCElfadvJqxe7bnmBYnkiQVjMvXS5IkZcjkRJKkgmnH4miFZnIiSZJyxeREkqSCac8X8hWZyYkkScoVkxNJkgqmwoMTkxNJkpQvJieSJBWMPSeSJEkZMjmRJKlgKjw4MTmRJEn5YnIiSVLBVHqyUOnnJ0mSCsbkRJKkgokKbzoxOZEkSbliciJJUsFUdm5iciJJknLG4kSSJOWKl3UkSSoYl6+XJEnKkMmJJEkFU9m5icmJJEnKGZMTSZIKpsJbTkxOJElSvpicSJJUMC5fL0mSlCGTE0mSCqbSk4VKPz9JklQwJieSJBWMPSeSJEkZMjmRJKlgKjs3MTmRJEk5Y3IiSVLB2HMiSZKUodwmJ12qrZu05m795qhyT0EFtdUJt5R7Ciqo584+OLP3qvS/kJV+fpIkqWBym5xIkqRVs+dEkiQpQxYnkiQpV7ysI0lSwVT2RR2TE0mSlDMmJ5IkFUyF98OanEiSpHwxOZEkqWCqKrzrxOREkiTlismJJEkFY8+JJElShkxOJEkqmLDnRJIkKTsmJ5IkFYw9J5IkSRkyOZEkqWBc50SSJClDJieSJBWMPSeSJEkZMjmRJKlgTE4kSZIyZHIiSVLBuEKsJElShixOJElSrnhZR5Kkgqmq7Ks6JieSJClfTE4kSSoYG2IlSZIyZHIiSVLBuAibJElShkxOJEkqGHtOJEmSMmRyIklSwbjOiSRJUoZMTiRJKhh7TiRJkjJkciJJUsG4zokkSVKGTE4kSSqYCg9OTE4kSVK+mJxIklQwVRXedGJyIkmScsXkRJKkgqns3MTkRJIk5YzFiSRJyhUv60iSVDQVfl3H5ESSJOWKyYkkSQXjF/9JkiRlyOREkqSCqfA12ExOJElSvpicSJJUMBUenJicSJKkfDE5kSSpaCo8OjE5kSRJuWJyIklSwbjOiSRJUoZMTiRJKhjXOZEkScqQyYkkSQVT4cGJyYkkScoXkxNJkoqmwqMTkxNJkpQrJieSJBWM65xIkiStRkQcEBFPRsT0iDhhFc9/MyKmRcSjEXFrRKzX1jEtTiRJ0hqJiGrgPOBAYCTwiYgYudKwh4AdUkpbAdcAZ7R1XIsTSZIKJiK7Rxt2AqanlJ5NKS0FrgRGNx+QUro9pfRmafM+YGhbB7U4kSRJqxURYyJiSrPHmGZPDwFmNNueWdq3Ol8EbmnrPW2IlSSpYLJsh00pjQPG/bfHiYhPAzsAe7Q11uJEkiStqVnAsGbbQ0v7WoiIDwEnAXuklJa0dVCLE0mSiiY/dxI/AIyIiA1oLEqOAj7ZfEBEbAtcBByQUprXnoPacyJJktZISmk5cCwwEXgCuCqlNDUixkbEYaVhZwI9gKsj4uGIGN/WcU1OJEkqmDwtwpZSmgBMWGnfyc1+/tB7PabJiSRJyhWTE0mSCqYd648UmsmJJEnKFZMTSZIKpsKDE5MTSZKULyYnkiQVTYVHJyYnkiQpV0xOJEkqmDytc9IRTE4kSVKumJxIklQwlb7OicVJzjw65V7+cOFZNDQ0sOcBozn0Y0e3eP6Wv/6JO/42nurqanr27sOXj/8h/esGsWDuS5z9k++SUgP1y5ez72EfY5+DP1yms1C5/Oueu/nF6afR0NDA4Ud8hM998cstnl+6dCmnnPQ9nnhiGr179+FnZ/ySwUOG8Phjj3LaT04BIKXEmGO+yl777FuOU1CZjNp0AKccOZKqCP5y3wwuvPWZFs//4PDN2HXEOgB071LNOj27sfWJk5qe79GthkknjuLvj83llGunZjp3VR6LkxxpqK/nsvPO4HunnUu//rWc/PWj2W7n3Rmy3vCmMettuAljz7mMbmutxT9uuoYrL/0Nx554Gn369eeUX15Cl65defutNznxmE+w3S6j6LvOgDKekbJUX1/P6af9hPMuuoS6ujo++8mPMWrPvRi+4UZNY2647hp69urN9TdNZOItN/Obs3/Bz878FRttNILLr7iampoaFsyfxyc+egS777EXNTX+E9EZVAWM/cjmfOaC+5nz6tvc8M3/4R+Pz2X63MVNY356/RNNPx+9+/qMHNqrxTG+edDGTH7mlczm3NlVeHDy/vecRMR5EfHB9/u4ncEzT02lbvBQagcNoaZLF3bZYz/+fd9dLcaM3HoHuq21FgAbbbolryxo/Pbpmi5d6NK1KwDLli0lpYZsJ6+ym/r4owwbti5Dhw6jS5eu7HfAQdx5x20txtx5+20ccthoAPbZd38mT76PlBJrde/eVIgsWbKUqPTMWC1svV4fXljwJjNefotl9YkbH5rNvlvWrXb8odsN5sZ/z27a3mJoL/r37MbdTy7IYrrqBDqiIfYp4BcR8XxEnBER23bAe1SkhQvm02/AO/8g9Otfy8KX5692/J2TxrPVDrs2bb88fy7f/8on+cZnD+Xgj37W1KSTmTdvHnUDBzZt19bWMW/u3JXGzKVu4CAAampq6NGjJ4tefRWAxx99hI8dcQhHfWQ0J/7gFFOTTmRg77V4aeFbTdtzXn2bgb3XWuXYIX27M6xfd/71dGMhEgEnHT6S0254YpXjpTXxvhcnKaVfp5R2BfYAXgYujYj/RMQpEbFxa6+NiDERMSUiplz359+/31OrKPfcdgvPPfUEB3/4M0371hlQx2kXXMEvLvkr//zHzSxa+HIZZ6ii2WKrrbnqupu4/Iqr+N0lF7NkyZJyT0k5dMh2g7jlkTk0pMbtz3xwPe6YNo85i94u78Q6m8jwUQYd9p9GKaUXgNOB00vpyaXAyUB1K68ZB4wDmPzsotRRc8urvv0H8Mr8d/5L95UF81aZfjz+0GTGX/k7vn/GhU2XclocZ50BDF1vQ558/GF22n2fDp2z8qO2tpa5c+Y0bc+bN5faurqVxtQxd85L1NUNZPny5Sxe/Dq9+/RpMWaD4Ruy9tpr88z0pxm5+RZZTF1lNmfR2wzq271pe2CftVZbbBy67WBOvuadhtdt1+/Ljhv249P/sx5rd62hS03wxpLlnHHTkx0+b1WuDlvnJCJqIuLQiPgTcAvwJHBkR71fJRi+8UjmzJ7BvDmzWL5sGffdOYntdtm9xZjnpz/J7875Gcef8gt69+nXtP+V+XNZuqTxH5M3Xn+Np6Y9zKCh62U6f5XXyM23ZMaLLzBr5kyWLVvKpL9NYNQee7UYM2rPvbhp/A0A3Pr3iey40y5EBLNmzmT58uUAvDR7Fs8//yyDBw/J/BxUHo++uIj1+3+Aof2606U6OHTbwfzj8bnvGje89gP0XrsLDz6/sGnf8X98mP/58W3sPvZ2Thv/BNc9MMvCJAOR4f/K4X1PTiJiX+ATwEHAZOBKYExK6Y33+70qTXV1DZ/9ync48wdfo6G+gVH7HcrQ9Tbk2ssvYoONN2O7XUZx5SXn8Pbbb/Gb004EYJ0BA/nmj85i1ozn+fPFv26M4BIceOSnGbbBRq2/oSpKTU0N3znxBxz3lS9R39DAYYcfyYYbjeDC885hs823YI8992b0ER/h5JO+x+GH7E+vXr057YyzAHj4oX9z2aUXU9OlCxHBCd8/mT59+5b5jJSV+obEKdc+zuXH7ERVVXD1/TN5es5ijj9wYx578VX+MbWx8f7Q7QZz44Oz2zia9N+LlN7fqycRcRtwBXBtSmlhW+NXpzNe1tH7Z7PBPcs9BRXUVifcUu4pqKCeO/vgzGKGJ+e8mdnfyE0Grp15fPK+Jycppb3f72NKkqTOw3sFJUkqmEpficgv/pMkSbliciJJUtFUeHRiciJJknLF5ESSpIIp1/ojWTE5kSRJuWJyIklSwVT6F4ebnEiSpFwxOZEkqWAqPDgxOZEkSfliciJJUtFUeHRiciJJknLF5ESSpIJxnRNJkqQMWZxIkqRc8bKOJEkF4yJskiRJGTI5kSSpYCo8ODE5kSRJ+WJyIklS0VR4dGJyIkmScsXkRJKkgnERNkmSpAyZnEiSVDCucyJJkpQhkxNJkgqmwoMTkxNJkpQvJieSJBWMPSeSJEkZMjmRJKlwKjs6MTmRJEm5YnIiSVLB2HMiSZKUIZMTSZIKpsKDE5MTSZKULxYnkiQpV7ysI0lSwdgQK0mSlCGTE0mSCiYqvCXW5ESSJOWKyYkkSUVT2cGJyYkkScoXkxNJkgqmwoMTkxNJkpQvJieSJBWM65xIkiRlyOREkqSCcZ0TSZKkDJmcSJJUNJUdnJicSJKkfDE5kSSpYCo8ODE5kSRJ+WJyIklSwbjOiSRJUoZMTiRJKhjXOZEkScqQxYkkScoVL+tIklQwNsRKkiRlyOJEkiTlisWJJEnKFXtOJEkqGHtOJEmSMmRyIklSwbgImyRJUoZMTiRJKhh7TiRJkjJkciJJUsFUeHBiciJJkvLF5ESSpKKp8OjE5ESSJOWKyYkkSQXjOieSJEkZMjmRJKlgXOdEkiQpQyYnkiQVTIUHJyYnkiQpXyxOJElSrnhZR5Kkoqnw6zomJ5IkKVdMTiRJKhgXYZMkSVqNiDggIp6MiOkRccIqnu8WEX8pPX9/RKzf1jEtTiRJKpiI7B6tzyOqgfOAA4GRwCciYuRKw74ILEwpbQT8Cji9rfOzOJEkSWtqJ2B6SunZlNJS4Epg9EpjRgOXlX6+BtgnovWyJ7c9JzsN713ZF9T+SxExJqU0rtzzUPH42Wndc2cfXO4p5JafnfxYqya7ppOIGAOMabZrXLPPwRBgRrPnZgI7r3SIpjEppeURsQhYB1iwuvc0OSmuMW0PkVbJz47WlJ+dTiilNC6ltEOzR4cXqBYnkiRpTc0ChjXbHlrat8oxEVED9AZebu2gFieSJGlNPQCMiIgNIqIrcBQwfqUx44GjSz9/BLgtpZRaO2hue07UJq/7ak352dGa8rOjFko9JMcCE4Fq4NKU0tSIGAtMSSmNBy4B/hAR04FXaCxgWhVtFC+SJEmZ8rKOJEnKFYsTSZKUKxYnBRIRKSLOarb97Yj4URmnpIKJiJMiYmpEPBoRD0fEyusRSC1ExO0Rsf9K+74REReUa06qfBYnxbIEODIi+pd7IiqeiNgVOATYLqW0FfAhWi6eJK3Kn3l3A+NRpf1Sh7A4KZblNHbLH1/uiaiQBgELUkpLAFJKC1JKs8s8J+XfNcDBpdtEKX1p22Dg7nJOSpXN4qR4zgM+FRG9yz0RFc4kYFhEPBUR50fEHuWekPIvpfQKMJnGL3aDxtTkqrbWqZD+GxYnBZNSeg24HPhaueeiYkkpLQa2p3EJ8vnAXyLic2WdlIqi+aUdL+mow7nOSYFExOKUUo+I6Ac8CPyOxv8Pf1TemamIIuIjwNEppUPLPRflW0T0AJ4FDgCuTCltXOYpqcKZnBRQKWa9Cvhiueei4oiITSJiRLNd2wAvlGk6KpBS6nY7cCmmJsqAxUlxnQV4147eix7AZRExLSIeBUYCPyrvlFQgfwa2xuJEGfCyjiRJyhWTE0mSlCsWJ5IkKVcsTiRJUq5YnEiSpFyxOJEkSblicSJ1sIioL30D8OMRcXVErP1fHOv3pcXTiIjfRsTIVsbuGRG7rcF7PL+qL5dc3f6Vxix+j+/1o4j49nudo6TKZnEidby3UkrbpJS2AJYCxzR/MiJq1uSgKaUvpZSmtTJkT+A9FyeSVG4WJ1K27gY2KqUad0fEeGBaRFRHxJkR8UBEPBoR/wsQjc6NiCcj4h9A7YoDRcQdEbFD6ecDIuLBiHgkIm4tfXPsMcDxpdRm94gYEBHXlt7jgYj4YOm160TEpIiYGhG/BaKtk4iI6yPi36XXjFnpuV+V9t8aEQNK+zaMiL+VXnN3RGz6vvw2JVWkNfovNknvXSkhORD4W2nXdsAWKaXnSn/gF6WUdoyIbsA9ETEJ2BbYhMbVXOuAaTQuId78uAOAi4FRpWP1Sym9EhEXAotTSr8ojbsC+FVK6Z8RsS4wEdgMOAX4Z0ppbEQcTPu+FuELpffoDjwQEdemlF4GPgBMSSkdHxEnl459LDAOOCal9HRE7AycD+y9Br9GSZ2AxYnU8bpHxMOln+8GLqHxcsvklNJzpf37AVut6CcBegMjgFHAn1NK9cDsiLhtFcffBbhrxbFK3720Kh8CRkY0BSO9Sl/oNgo4svTamyNiYTvO6WsRcUTp52Glub4MNAB/Ke3/I/DX0nvsBlzd7L27teM9JHVSFidSx3srpbRN8x2lP9JvNN8FHJdSmrjSuIPex3lUAbuklN5exVzaLSL2pLHQ2TWl9GZE3AGstZrhqfS+r678O5Ck1bHnRMqHicBXIqILQERsHBEfAO4CPl7qSRkE7LWK194HjIqIDUqv7Vfa/zrQs9m4ScBxKzYiYpvSj3cBnyztOxDo28ZcewMLS4XJpjQmNytUASvSn0/SeLnoNeC5iPho6T0iIrZu4z0kdWIWJ1I+/JbGfpIHI+Jx4CIak83rgKdLz10O3LvyC1NK84ExNF5CeYR3LqvcCByxoiEW+BqwQ6nhdhrv3DX0YxqLm6k0Xt55sY25/g2oiYgngJ/TWByt8AawU+kc9gbGlvZ/CvhiaX5TgdHt+J1I6qT8VmJJkpQrJieSJClXLE4kSVKuWJxIkqRcsTiRJEm5YnEiSZJyxeJEkiTlisWJJEnKlf8H7jQ+zPkKu6oAAAAASUVORK5CYII=",
      "text/plain": [
       "<Figure size 576x576 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_confusion_matrix(y_valid, y_pred)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2>Neural Network "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 44955/44955 [00:02<00:00, 19761.35it/s]\n",
      "100%|██████████| 23659/23659 [00:01<00:00, 20278.39it/s]\n",
      "100%|██████████| 24866/24866 [00:01<00:00, 20852.28it/s]\n"
     ]
    }
   ],
   "source": [
    "\n",
    "x_train,y_train = create_input(training_patches_df)\n",
    "x_valid,y_valid = create_input(validation_patches_df)\n",
    "x_test,y_test = create_input(test_patches_df)\n",
    "\n",
    "\n",
    "x_train = encode_input(encoder,x_train)\n",
    "x_valid = encode_input(encoder,x_valid)\n",
    "x_test = encode_input(encoder,x_test)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "176/176 [==============================] - 3s 13ms/step - loss: 0.6098 - accuracy: 0.5079 - lr: 1.0000e-04 - val_loss: 0.6058 - val_accuracy: 0.4909 - val_lr: 1.0000e-04\n",
      "Epoch 2/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.5290 - accuracy: 0.6201 - lr: 1.0000e-04 - val_loss: 0.5363 - val_accuracy: 0.6065 - val_lr: 1.0000e-04\n",
      "Epoch 3/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.4674 - accuracy: 0.6884 - lr: 1.0000e-04 - val_loss: 0.5231 - val_accuracy: 0.6081 - val_lr: 1.0000e-04\n",
      "Epoch 4/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.4105 - accuracy: 0.7448 - lr: 1.0000e-04 - val_loss: 0.4506 - val_accuracy: 0.7181 - val_lr: 1.0000e-04\n",
      "Epoch 5/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.3608 - accuracy: 0.7853 - lr: 1.0000e-04 - val_loss: 0.4001 - val_accuracy: 0.7751 - val_lr: 1.0000e-04\n",
      "Epoch 6/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.3247 - accuracy: 0.8117 - lr: 1.0000e-04 - val_loss: 0.3763 - val_accuracy: 0.7945 - val_lr: 1.0000e-04\n",
      "Epoch 7/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.2958 - accuracy: 0.8343 - lr: 1.0000e-04 - val_loss: 0.3322 - val_accuracy: 0.8293 - val_lr: 1.0000e-04\n",
      "Epoch 8/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.2750 - accuracy: 0.8492 - lr: 1.0000e-04 - val_loss: 0.3411 - val_accuracy: 0.8261 - val_lr: 1.0000e-04\n",
      "Epoch 9/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.2599 - accuracy: 0.8596 - lr: 1.0000e-04 - val_loss: 0.3084 - val_accuracy: 0.8439 - val_lr: 1.0000e-04\n",
      "Epoch 10/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.2459 - accuracy: 0.8670 - lr: 1.0000e-04 - val_loss: 0.2850 - val_accuracy: 0.8595 - val_lr: 1.0000e-04\n",
      "Epoch 11/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.2367 - accuracy: 0.8721 - lr: 1.0000e-04 - val_loss: 0.2884 - val_accuracy: 0.8534 - val_lr: 1.0000e-04\n",
      "Epoch 12/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.2274 - accuracy: 0.8781 - lr: 1.0000e-04 - val_loss: 0.2605 - val_accuracy: 0.8742 - val_lr: 1.0000e-04\n",
      "Epoch 13/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.2193 - accuracy: 0.8830 - lr: 1.0000e-04 - val_loss: 0.2847 - val_accuracy: 0.8556 - val_lr: 1.0000e-04\n",
      "Epoch 14/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.2136 - accuracy: 0.8872 - lr: 1.0000e-04 - val_loss: 0.2555 - val_accuracy: 0.8729 - val_lr: 1.0000e-04\n",
      "Epoch 15/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.2079 - accuracy: 0.8917 - lr: 1.0000e-04 - val_loss: 0.3033 - val_accuracy: 0.8465 - val_lr: 1.0000e-04\n",
      "Epoch 16/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.2023 - accuracy: 0.8923 - lr: 1.0000e-04 - val_loss: 0.2491 - val_accuracy: 0.8796 - val_lr: 1.0000e-04\n",
      "Epoch 17/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1991 - accuracy: 0.8942 - lr: 1.0000e-04 - val_loss: 0.2461 - val_accuracy: 0.8795 - val_lr: 1.0000e-04\n",
      "Epoch 18/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1935 - accuracy: 0.8985 - lr: 1.0000e-04 - val_loss: 0.2327 - val_accuracy: 0.8937 - val_lr: 1.0000e-04\n",
      "Epoch 19/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1895 - accuracy: 0.9005 - lr: 1.0000e-04 - val_loss: 0.2366 - val_accuracy: 0.8825 - val_lr: 1.0000e-04\n",
      "Epoch 20/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.1856 - accuracy: 0.9024 - lr: 1.0000e-04 - val_loss: 0.2436 - val_accuracy: 0.8792 - val_lr: 1.0000e-04\n",
      "Epoch 21/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.1819 - accuracy: 0.9045 - lr: 1.0000e-04 - val_loss: 0.2228 - val_accuracy: 0.8942 - val_lr: 1.0000e-04\n",
      "Epoch 22/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1795 - accuracy: 0.9039 - lr: 1.0000e-04 - val_loss: 0.2395 - val_accuracy: 0.8830 - val_lr: 1.0000e-04\n",
      "Epoch 23/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1755 - accuracy: 0.9070 - lr: 1.0000e-04 - val_loss: 0.2411 - val_accuracy: 0.8825 - val_lr: 1.0000e-04\n",
      "Epoch 24/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1720 - accuracy: 0.9081 - lr: 1.0000e-04 - val_loss: 0.2217 - val_accuracy: 0.8948 - val_lr: 1.0000e-04\n",
      "Epoch 25/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1685 - accuracy: 0.9112 - lr: 9.6307e-05 - val_loss: 0.2330 - val_accuracy: 0.8871 - val_lr: 7.5000e-05\n",
      "Epoch 26/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1669 - accuracy: 0.9122 - lr: 7.5000e-05 - val_loss: 0.2029 - val_accuracy: 0.9087 - val_lr: 7.5000e-05\n",
      "Epoch 27/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1648 - accuracy: 0.9130 - lr: 7.5000e-05 - val_loss: 0.2167 - val_accuracy: 0.8962 - val_lr: 7.5000e-05\n",
      "Epoch 28/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1624 - accuracy: 0.9132 - lr: 7.5000e-05 - val_loss: 0.1992 - val_accuracy: 0.9084 - val_lr: 7.5000e-05\n",
      "Epoch 29/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1599 - accuracy: 0.9143 - lr: 7.5000e-05 - val_loss: 0.1905 - val_accuracy: 0.9127 - val_lr: 7.5000e-05\n",
      "Epoch 30/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1584 - accuracy: 0.9159 - lr: 7.5000e-05 - val_loss: 0.2361 - val_accuracy: 0.8849 - val_lr: 7.5000e-05\n",
      "Epoch 31/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1566 - accuracy: 0.9166 - lr: 7.5000e-05 - val_loss: 0.2098 - val_accuracy: 0.9021 - val_lr: 7.5000e-05\n",
      "Epoch 32/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1555 - accuracy: 0.9165 - lr: 7.5000e-05 - val_loss: 0.2011 - val_accuracy: 0.9054 - val_lr: 7.5000e-05\n",
      "Epoch 33/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1536 - accuracy: 0.9183 - lr: 7.5000e-05 - val_loss: 0.1869 - val_accuracy: 0.9146 - val_lr: 7.5000e-05\n",
      "Epoch 34/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1525 - accuracy: 0.9187 - lr: 7.5000e-05 - val_loss: 0.2166 - val_accuracy: 0.8974 - val_lr: 7.5000e-05\n",
      "Epoch 35/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1503 - accuracy: 0.9208 - lr: 7.5000e-05 - val_loss: 0.2094 - val_accuracy: 0.9004 - val_lr: 7.5000e-05\n",
      "Epoch 36/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1476 - accuracy: 0.9226 - lr: 7.5000e-05 - val_loss: 0.1987 - val_accuracy: 0.9044 - val_lr: 7.5000e-05\n",
      "Epoch 37/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1482 - accuracy: 0.9206 - lr: 7.5000e-05 - val_loss: 0.1972 - val_accuracy: 0.9060 - val_lr: 7.5000e-05\n",
      "Epoch 38/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1456 - accuracy: 0.9229 - lr: 7.5000e-05 - val_loss: 0.1857 - val_accuracy: 0.9106 - val_lr: 7.5000e-05\n",
      "Epoch 39/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.1442 - accuracy: 0.9225 - lr: 7.5000e-05 - val_loss: 0.1547 - val_accuracy: 0.9268 - val_lr: 7.5000e-05\n",
      "Epoch 40/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1432 - accuracy: 0.9249 - lr: 7.5000e-05 - val_loss: 0.1862 - val_accuracy: 0.9096 - val_lr: 7.5000e-05\n",
      "Epoch 41/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1420 - accuracy: 0.9244 - lr: 7.5000e-05 - val_loss: 0.2064 - val_accuracy: 0.8977 - val_lr: 7.5000e-05\n",
      "Epoch 42/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1400 - accuracy: 0.9249 - lr: 7.5000e-05 - val_loss: 0.1831 - val_accuracy: 0.9134 - val_lr: 7.5000e-05\n",
      "Epoch 43/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1387 - accuracy: 0.9269 - lr: 7.5000e-05 - val_loss: 0.1728 - val_accuracy: 0.9183 - val_lr: 7.5000e-05\n",
      "Epoch 44/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1378 - accuracy: 0.9259 - lr: 7.5000e-05 - val_loss: 0.1791 - val_accuracy: 0.9123 - val_lr: 7.5000e-05\n",
      "Epoch 45/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1361 - accuracy: 0.9278 - lr: 7.5000e-05 - val_loss: 0.1929 - val_accuracy: 0.9082 - val_lr: 7.5000e-05\n",
      "Epoch 46/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1357 - accuracy: 0.9280 - lr: 7.5000e-05 - val_loss: 0.1894 - val_accuracy: 0.9064 - val_lr: 7.5000e-05\n",
      "Epoch 47/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1350 - accuracy: 0.9284 - lr: 7.5000e-05 - val_loss: 0.1663 - val_accuracy: 0.9211 - val_lr: 7.5000e-05\n",
      "Epoch 48/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1324 - accuracy: 0.9294 - lr: 7.5000e-05 - val_loss: 0.1637 - val_accuracy: 0.9216 - val_lr: 7.5000e-05\n",
      "Epoch 49/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1327 - accuracy: 0.9301 - lr: 7.5000e-05 - val_loss: 0.1782 - val_accuracy: 0.9138 - val_lr: 7.5000e-05\n",
      "Epoch 50/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.1309 - accuracy: 0.9308 - lr: 6.9567e-05 - val_loss: 0.1638 - val_accuracy: 0.9202 - val_lr: 5.6250e-05\n",
      "Epoch 51/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.1300 - accuracy: 0.9303 - lr: 5.6250e-05 - val_loss: 0.1651 - val_accuracy: 0.9193 - val_lr: 5.6250e-05\n",
      "Epoch 52/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.1292 - accuracy: 0.9316 - lr: 5.6250e-05 - val_loss: 0.1960 - val_accuracy: 0.9043 - val_lr: 5.6250e-05\n",
      "Epoch 53/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.1278 - accuracy: 0.9323 - lr: 5.6250e-05 - val_loss: 0.1997 - val_accuracy: 0.9040 - val_lr: 5.6250e-05\n",
      "Epoch 54/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1276 - accuracy: 0.9332 - lr: 5.6250e-05 - val_loss: 0.1599 - val_accuracy: 0.9237 - val_lr: 5.6250e-05\n",
      "Epoch 55/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1266 - accuracy: 0.9334 - lr: 5.6250e-05 - val_loss: 0.1693 - val_accuracy: 0.9176 - val_lr: 5.6250e-05\n",
      "Epoch 56/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.1262 - accuracy: 0.9331 - lr: 5.6250e-05 - val_loss: 0.1528 - val_accuracy: 0.9261 - val_lr: 5.6250e-05\n",
      "Epoch 57/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1252 - accuracy: 0.9339 - lr: 5.6250e-05 - val_loss: 0.1972 - val_accuracy: 0.9033 - val_lr: 5.6250e-05\n",
      "Epoch 58/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1240 - accuracy: 0.9345 - lr: 5.6250e-05 - val_loss: 0.1813 - val_accuracy: 0.9117 - val_lr: 5.6250e-05\n",
      "Epoch 59/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1242 - accuracy: 0.9328 - lr: 5.6250e-05 - val_loss: 0.2111 - val_accuracy: 0.8964 - val_lr: 5.6250e-05\n",
      "Epoch 60/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1241 - accuracy: 0.9336 - lr: 5.6250e-05 - val_loss: 0.1832 - val_accuracy: 0.9108 - val_lr: 5.6250e-05\n",
      "Epoch 61/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1236 - accuracy: 0.9344 - lr: 5.6250e-05 - val_loss: 0.1788 - val_accuracy: 0.9099 - val_lr: 5.6250e-05\n",
      "Epoch 62/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1221 - accuracy: 0.9354 - lr: 5.6250e-05 - val_loss: 0.1680 - val_accuracy: 0.9179 - val_lr: 5.6250e-05\n",
      "Epoch 63/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1210 - accuracy: 0.9361 - lr: 5.6250e-05 - val_loss: 0.1516 - val_accuracy: 0.9251 - val_lr: 5.6250e-05\n",
      "Epoch 64/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1210 - accuracy: 0.9361 - lr: 5.6250e-05 - val_loss: 0.1678 - val_accuracy: 0.9167 - val_lr: 5.6250e-05\n",
      "Epoch 65/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1204 - accuracy: 0.9362 - lr: 5.6250e-05 - val_loss: 0.1732 - val_accuracy: 0.9155 - val_lr: 5.6250e-05\n",
      "Epoch 66/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1190 - accuracy: 0.9375 - lr: 5.6250e-05 - val_loss: 0.1610 - val_accuracy: 0.9216 - val_lr: 5.6250e-05\n",
      "Epoch 67/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1185 - accuracy: 0.9368 - lr: 5.6250e-05 - val_loss: 0.1720 - val_accuracy: 0.9145 - val_lr: 5.6250e-05\n",
      "Epoch 68/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.1190 - accuracy: 0.9354 - lr: 5.6250e-05 - val_loss: 0.1496 - val_accuracy: 0.9246 - val_lr: 5.6250e-05\n",
      "Epoch 69/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.1163 - accuracy: 0.9391 - lr: 5.6250e-05 - val_loss: 0.1676 - val_accuracy: 0.9177 - val_lr: 5.6250e-05\n",
      "Epoch 70/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1179 - accuracy: 0.9375 - lr: 5.6250e-05 - val_loss: 0.1809 - val_accuracy: 0.9088 - val_lr: 5.6250e-05\n",
      "Epoch 71/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1156 - accuracy: 0.9386 - lr: 5.6250e-05 - val_loss: 0.1548 - val_accuracy: 0.9252 - val_lr: 5.6250e-05\n",
      "Epoch 72/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1164 - accuracy: 0.9388 - lr: 5.6250e-05 - val_loss: 0.1805 - val_accuracy: 0.9099 - val_lr: 5.6250e-05\n",
      "Epoch 73/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1157 - accuracy: 0.9377 - lr: 5.6250e-05 - val_loss: 0.1884 - val_accuracy: 0.9048 - val_lr: 5.6250e-05\n",
      "Epoch 74/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1147 - accuracy: 0.9391 - lr: 5.6250e-05 - val_loss: 0.1496 - val_accuracy: 0.9237 - val_lr: 5.6250e-05\n",
      "Epoch 75/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1147 - accuracy: 0.9401 - lr: 5.0178e-05 - val_loss: 0.1456 - val_accuracy: 0.9274 - val_lr: 4.2188e-05\n",
      "Epoch 76/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1128 - accuracy: 0.9404 - lr: 4.2188e-05 - val_loss: 0.1526 - val_accuracy: 0.9250 - val_lr: 4.2188e-05\n",
      "Epoch 77/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1127 - accuracy: 0.9406 - lr: 4.2188e-05 - val_loss: 0.1594 - val_accuracy: 0.9213 - val_lr: 4.2188e-05\n",
      "Epoch 78/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1133 - accuracy: 0.9401 - lr: 4.2188e-05 - val_loss: 0.1572 - val_accuracy: 0.9214 - val_lr: 4.2188e-05\n",
      "Epoch 79/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1118 - accuracy: 0.9406 - lr: 4.2188e-05 - val_loss: 0.1606 - val_accuracy: 0.9205 - val_lr: 4.2188e-05\n",
      "Epoch 80/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1123 - accuracy: 0.9415 - lr: 4.2188e-05 - val_loss: 0.1407 - val_accuracy: 0.9279 - val_lr: 4.2188e-05\n",
      "Epoch 81/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1109 - accuracy: 0.9416 - lr: 4.2188e-05 - val_loss: 0.1624 - val_accuracy: 0.9177 - val_lr: 4.2188e-05\n",
      "Epoch 82/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1117 - accuracy: 0.9411 - lr: 4.2188e-05 - val_loss: 0.1430 - val_accuracy: 0.9273 - val_lr: 4.2188e-05\n",
      "Epoch 83/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1103 - accuracy: 0.9420 - lr: 4.2188e-05 - val_loss: 0.1629 - val_accuracy: 0.9179 - val_lr: 4.2188e-05\n",
      "Epoch 84/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1099 - accuracy: 0.9419 - lr: 4.2188e-05 - val_loss: 0.1561 - val_accuracy: 0.9221 - val_lr: 4.2188e-05\n",
      "Epoch 85/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1097 - accuracy: 0.9420 - lr: 4.2188e-05 - val_loss: 0.1471 - val_accuracy: 0.9268 - val_lr: 4.2188e-05\n",
      "Epoch 86/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1091 - accuracy: 0.9429 - lr: 4.2188e-05 - val_loss: 0.1743 - val_accuracy: 0.9123 - val_lr: 4.2188e-05\n",
      "Epoch 87/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1089 - accuracy: 0.9430 - lr: 4.2188e-05 - val_loss: 0.1575 - val_accuracy: 0.9208 - val_lr: 4.2188e-05\n",
      "Epoch 88/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1093 - accuracy: 0.9423 - lr: 4.2188e-05 - val_loss: 0.1612 - val_accuracy: 0.9180 - val_lr: 4.2188e-05\n",
      "Epoch 89/100\n",
      "176/176 [==============================] - 1s 8ms/step - loss: 0.1088 - accuracy: 0.9424 - lr: 4.2188e-05 - val_loss: 0.1594 - val_accuracy: 0.9188 - val_lr: 4.2188e-05\n",
      "Epoch 90/100\n",
      "176/176 [==============================] - 2s 9ms/step - loss: 0.1091 - accuracy: 0.9423 - lr: 4.2188e-05 - val_loss: 0.1651 - val_accuracy: 0.9163 - val_lr: 4.2188e-05\n",
      "Epoch 91/100\n",
      "176/176 [==============================] - 2s 10ms/step - loss: 0.1080 - accuracy: 0.9427 - lr: 4.2188e-05 - val_loss: 0.1711 - val_accuracy: 0.9145 - val_lr: 4.2188e-05\n",
      "Epoch 92/100\n",
      "176/176 [==============================] - 2s 10ms/step - loss: 0.1079 - accuracy: 0.9428 - lr: 4.2188e-05 - val_loss: 0.1502 - val_accuracy: 0.9242 - val_lr: 4.2188e-05\n",
      "Epoch 93/100\n",
      "176/176 [==============================] - 2s 9ms/step - loss: 0.1071 - accuracy: 0.9443 - lr: 4.2188e-05 - val_loss: 0.1514 - val_accuracy: 0.9241 - val_lr: 4.2188e-05\n",
      "Epoch 94/100\n",
      "176/176 [==============================] - 2s 9ms/step - loss: 0.1069 - accuracy: 0.9438 - lr: 4.2188e-05 - val_loss: 0.1462 - val_accuracy: 0.9258 - val_lr: 4.2188e-05\n",
      "Epoch 95/100\n",
      "176/176 [==============================] - 2s 9ms/step - loss: 0.1069 - accuracy: 0.9438 - lr: 4.2188e-05 - val_loss: 0.1643 - val_accuracy: 0.9168 - val_lr: 4.2188e-05\n",
      "Epoch 96/100\n",
      "176/176 [==============================] - 2s 9ms/step - loss: 0.1054 - accuracy: 0.9446 - lr: 4.2188e-05 - val_loss: 0.1735 - val_accuracy: 0.9122 - val_lr: 4.2188e-05\n",
      "Epoch 97/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1056 - accuracy: 0.9442 - lr: 4.2188e-05 - val_loss: 0.1581 - val_accuracy: 0.9196 - val_lr: 4.2188e-05\n",
      "Epoch 98/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1055 - accuracy: 0.9441 - lr: 4.2188e-05 - val_loss: 0.1611 - val_accuracy: 0.9184 - val_lr: 4.2188e-05\n",
      "Epoch 99/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1053 - accuracy: 0.9439 - lr: 4.2188e-05 - val_loss: 0.1562 - val_accuracy: 0.9208 - val_lr: 4.2188e-05\n",
      "Epoch 100/100\n",
      "176/176 [==============================] - 1s 7ms/step - loss: 0.1045 - accuracy: 0.9451 - lr: 3.6135e-05 - val_loss: 0.1509 - val_accuracy: 0.9227 - val_lr: 3.1641e-05\n"
     ]
    }
   ],
   "source": [
    "from keras.regularizers import l2\n",
    "\n",
    "def get_lr_metric(optimizer):\n",
    "    def lr(y_true, y_pred):\n",
    "        return optimizer._decayed_lr(tf.float32) # I use ._decayed_lr method instead of .lr\n",
    "    return lr\n",
    "\n",
    "batch_size = 256\n",
    "\n",
    "input_shape = x_train.shape[1]\n",
    "\n",
    "input = keras.Input(shape=input_shape,name='Input_layer')\n",
    "\n",
    "regularizer = l2(0.)\n",
    "\n",
    "dense1 = layers.Dense(512,activation='relu',bias_regularizer=regularizer, activity_regularizer=regularizer)(input)\n",
    "dropout1 = layers.Dropout(0.1,name='Dropout_1')(dense1)\n",
    "dense2 = layers.Dense(512,activation='relu',bias_regularizer=regularizer, activity_regularizer=regularizer)(dropout1)\n",
    "dropout2 = layers.Dropout(0.1)(dense2)\n",
    "out = layers.Dense(3, activation='softmax')(dropout2)\n",
    "\n",
    "classifier = keras.Model(input, out)\n",
    "\n",
    "initial_learning_rate = 1e-4\n",
    "\n",
    "lr_schedule = tf.keras.optimizers.schedules.ExponentialDecay(\n",
    "    initial_learning_rate,\n",
    "    decay_steps=int(x_train.shape[0]/batch_size)*25,\n",
    "    decay_rate=0.75,\n",
    "    staircase=True)\n",
    "\n",
    "\n",
    "opt = tf.keras.optimizers.Adam(learning_rate=lr_schedule)\n",
    "lr_metric = get_lr_metric(opt)\n",
    "\n",
    "classifier.compile(loss='binary_crossentropy', optimizer=opt, metrics=['accuracy',lr_metric])\n",
    "history = classifier.fit(x_train, y_train,\n",
    "                epochs=100,\n",
    "                batch_size=batch_size,\n",
    "                shuffle=True,\n",
    "                validation_data=(x_valid, y_valid))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history.history[\"loss\"], label=\"Training Loss\")\n",
    "plt.plot(history.history[\"val_loss\"], label=\"Validation Loss\")\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "validation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 576x576 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "y_pred = classifier.predict(x_valid)\n",
    "plot_confusion_matrix(y_valid, y_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       0.99      0.93      0.96     20924\n",
      "           1       0.54      0.85      0.66      1032\n",
      "           2       0.61      0.89      0.73      1703\n",
      "\n",
      "    accuracy                           0.92     23659\n",
      "   macro avg       0.71      0.89      0.78     23659\n",
      "weighted avg       0.95      0.92      0.93     23659\n",
      "\n"
     ]
    }
   ],
   "source": [
    "y_pred_ = np.argmax(y_pred,axis=1)\n",
    "y_valid_ = np.argmax(y_valid,axis=1)\n",
    "\n",
    "print(classification_report(y_valid_, y_pred_))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2> Without autoencoder"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h3>SVC"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 44955/44955 [00:02<00:00, 21659.82it/s]\n",
      "100%|██████████| 23659/23659 [00:01<00:00, 22089.86it/s]\n",
      "100%|██████████| 24866/24866 [00:01<00:00, 21749.31it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(44955, 350, 1)\n",
      "[1. 0. 0.]\n",
      "Sv trained..\n",
      "Scoring:\n"
     ]
    }
   ],
   "source": [
    "from sklearn.svm import SVC\n",
    "\n",
    "\n",
    "x_train,y_train = create_input(training_patches_df)\n",
    "x_valid,y_valid = create_input(validation_patches_df)\n",
    "x_test,y_test = create_input(test_patches_df)\n",
    "\n",
    "print(x_train.shape)\n",
    "print(y_train[0])\n",
    "\n",
    "y_train = np.argmax(y_train,axis=1)\n",
    "y_valid = np.argmax(y_valid,axis=1)\n",
    "y_test = np.argmax(y_test,axis=1)\n",
    "\n",
    "\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "\n",
    "sv = SVC(C=1, kernel='rbf', degree=3,   # C is a regularization parameter -> the higher and the more we weights points outside the margin \n",
    "         gamma='scale', coef0=0.0, shrinking=True, \n",
    "         probability=False, tol=0.001, cache_size=200, \n",
    "         class_weight=None, verbose=False, max_iter=- 1, \n",
    "         decision_function_shape='ovr', break_ties=False, \n",
    "         random_state=SEED)\n",
    "\n",
    "print(\"training ...\",end=\"\\r\")\n",
    "sv.fit(x_train[:,:,0],y_train)\n",
    "print(\"Sv trained\",end=\"\\n\")\n",
    "print(\"Scoring:\")\n",
    "#sv.score(x_valid[:,:,0],y_valid)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "import seaborn as sns \n",
    "def plot_confusion_matrix(y_true, y_pred):\n",
    "  \"\"\"\n",
    "    :param ndarray y_true: contains the ground truth labels in the shape (n_samples,)\n",
    "    :param ndarray y_pred: contains the predicted labels in the shape (n_samples,)\n",
    "  \"\"\"\n",
    "  #y_true = np.argmax(y_true,axis=1) already 1dimensional\n",
    "  #y_pred = np.argmax(y_pred,axis=1)\n",
    "  \n",
    "  con_mat = tf.math.confusion_matrix(labels=y_true, predictions=y_pred).numpy()\n",
    "  con_mat_norm = np.around(con_mat.astype('float') / con_mat.sum(axis=1)[:, np.newaxis], decimals=2)\n",
    "  con_mat_df = pd.DataFrame(con_mat_norm,\n",
    "                  index = classes, \n",
    "                  columns = classes)\n",
    "  figure = plt.figure(figsize=(8, 8))\n",
    "  sns.heatmap(con_mat_df, annot=True,cmap=plt.cm.Blues)\n",
    "  plt.tight_layout()\n",
    "  plt.ylabel('True label')\n",
    "  plt.xlabel('Predicted label')\n",
    "  plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 576x576 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "y_pred = sv.predict(x_valid[:,:,0])\n",
    "plot_confusion_matrix(y_valid,y_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       1.00      0.93      0.96     20924\n",
      "           1       0.46      0.85      0.60      1032\n",
      "           2       0.69      0.93      0.79      1703\n",
      "\n",
      "    accuracy                           0.92     23659\n",
      "   macro avg       0.72      0.90      0.78     23659\n",
      "weighted avg       0.95      0.92      0.93     23659\n",
      "\n"
     ]
    }
   ],
   "source": [
    "\n",
    "print(classification_report(y_valid, y_pred))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_pred = sv.predict(x_test[:,:,0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 24866/24866 [00:01<00:00, 21323.05it/s]\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 576x576 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "x_test,y_test = create_input(test_patches_df)\n",
    "y_test = np.argmax(y_test,axis=1)\n",
    "plot_confusion_matrix(y_test,y_pred)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h3> KNN "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "KNN trained.\n"
     ]
    }
   ],
   "source": [
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "\n",
    "KNN = KNeighborsClassifier(n_neighbors=3,weights='distance',)\n",
    "print(\"training ...\",end=\"\\r\")\n",
    "KNN.fit(x_train[:,:,0],y_train)\n",
    "print(\"KNN trained\",end=\"\\n\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAicAAAJGCAYAAACA+CUiAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/YYfK9AAAACXBIWXMAAAsTAAALEwEAmpwYAAAyHklEQVR4nO3deZxVdf3H8ddnGHABARUYEHDHDM3SXCsFKRVcAFFTtMWyaLN9UX8mGaWlaZqFFZqlZVpqKiqK5pJLKqC5gaJEKpuACq4kzPD9/TEXmkGYGUbn3HPuvJ49zuNxzznf+73fQ1f5+D7f872RUkKSJCkvqso9AEmSpIYsTiRJUq5YnEiSpFyxOJEkSblicSJJknKlutwDWJeNdj3Rx4jUakum/qrcQ1BBrfQJRrXSxh0jsvqsLP+OXPavX2V2XauYnEiSpFzJbXIiSZLWISo7W6jsq5MkSYVjcSJJknLF2zqSJBVNdnNvy8LkRJIk5YrJiSRJReOEWEmSpOxYnEiSVDQR2W3NDiWGRsTMiJgVESev5fxWEXF7RDwWEXdFRL/m+rQ4kSRJrRIRHYDxwDBgIDA6Igau0ewc4LKU0i7AOOAnzfVrcSJJUtFEVXZb0/YEZqWUZqeUlgNXAiPWaDMQuKP0+s61nH8bixNJktRafYE5Dfbnlo419CgwqvT6cGCTiNi8qU4tTiRJKpoM55xExJiImNZgG7Oeo/0OMCgi/gUMAuYBdU29wUeJJUnSOqWUJgAT1nF6HtC/wX6/0rGG759PKTmJiC7AESmlpU19psWJJElFk591TqYCAyJiG+qLkmOAYxs2iIgewMsppZXAKcAlzXWam6uTJEnFklKqBU4EJgNPAn9NKU2PiHERMbzUbDAwMyKeBmqAM5rr1+REkqSiydFv66SUJgGT1jg2tsHrq4Gr16dPkxNJkpQrJieSJBVNfuactInKvjpJklQ4JieSJBVNjuactAWTE0mSlCsmJ5IkFY1zTiRJkrJjcSJJknLF2zqSJBWNE2IlSZKyY3IiSVLROCFWkiQpOyYnkiQVjcmJJElSdkxOJEkqmiqf1pEkScqMyYkkSUXjnBNJkqTsmJxIklQ0rhArSZKUHZMTSZKKxjknkiRJ2TE5kSSpaJxzIkmSlB2TE0mSisY5J5IkSdkxOZEkqWiccyJJkpQdixNJkpQr3taRJKlonBArSZKUHZMTSZKKxgmxkiRJ2TE5kSSpaJxzIkmSlB2TE0mSisY5J5IkSdkxOZEkqWiccyJJkpQdkxNJkorG5ESSJCk7JieSJBWNT+tIkiRlx+REkqSicc6JJElSdkxOJEkqGuecSJIkZcfkRJKkonHOiSRJUnYsTiRJUq54W0eSpKJxQqwkSVJ2LE4kSSqYiMhsa8FYhkbEzIiYFREnr+X8lhFxZ0T8KyIei4iDm+vT4kSSJLVKRHQAxgPDgIHA6IgYuEaz7wN/TSntChwDXNhcv845kSSpYFqSaGRkT2BWSmk2QERcCYwAZjRok4CupdfdgPnNdWpxIkmSWqsvMKfB/lxgrzXanA7cGhFfBToDH2uuU2/rSJJUNJHdFhFjImJag23Meo52NPCHlFI/4GDgjxFNryJnciJJktYppTQBmLCO0/OA/g32+5WONXQCMLTU1/0RsSHQA1i0rs80OZEkqWBy9LTOVGBARGwTEZ2on/A6cY02zwMfLY37vcCGwOKmOrU4kSRJrZJSqgVOBCYDT1L/VM70iBgXEcNLzb4NfD4iHgWuAI5PKaWm+vW2jiRJBZOjp3VIKU0CJq1xbGyD1zOAD69PnyYnkiQpV0xOJEkqmDwlJ23B5ESSJOWKyYkkSQVjciJJkpQhkxNJkoqmsoMTkxNJkpQvFic5c8CH3suj157GE9f/gO985oC3nd+yz6ZM+s1XmfKXU5h80dfp26s7ALvs0Je7Lv02D119KlP+cgpHHrhbxiNXHtx3z90MP+QgDh16AL+76O2rTS9fvpzvfvsbHDr0AI475ijmzZvb6PyC+fPZe/ddufT3v8tqyMqJ++69h5GHDmX4sAO55OK1f3dO+vY3GT7sQD45+uPML3135s+by94ffD9HHzGSo48YyY9/+IOsh94u5WiF2DbhbZ0cqaoKzj/54xzypV8xb+FS7r38u9z4j8d5avYLq9v85JuHc/lNU7j8hgcZtMcOjPvqcE447TLe/O8KTjjtMv79/GL69OzGfZd/j9v++SSvvL6sjFekLNXV1XHmGeP47UW/p6amhmOPPpLB+w9hu+23X93m2muuomvXrtx4y23cPOkmzv/5Ofzs3PNXnz/n7J/ykX33LcPoVU51dXX89Mfj+PVFl1DTu4bjjj6KQfsPYbvt/vfdue5vV7NJ165MvPlWbpl0E7/4+bmcde55APTrvyV/uea6Mo1elcjkJEf22Hlr/j3nRZ6d9xIrauu4avLDHDp4l0Ztdty2D/+YMhOAf0x9mkMHvw+AWc8v4t/P1/9UwYLFr7B4yWv02KxLthegsnri8cfo338r+vXvT8dOnRh68CHcdeftjdrceccdDB9xOAAHHHgQUx64n1WrSN9x+9/p268v220/IPOxq7yeePwx+m+5Zf13p2MnDhp2MHfd0fi7c9cdt3PYiJEAfOzAg5jy4P++O9K7zeIkR7bo1Y25C5es3p+3cAl9e3Zr1Obxp+cxYsgHABgx5P107bIRm3Xr3KjN7jttRafqambPebHNx6z8WLRwIb379F6936umhoULFzZus2ghvXv3AaC6upoum2zC0qVLePONN/j97y7ii186MdMxKx8WLVpITel7AVBT05vFi9b87ixq/N3psglLly4FYN68uRxz5OGccPwnePihaZmNuz3ztk4rRMTYJk6nlNKP1vG+McAYgOp+g6nusVNbDK/QTjnvWs476Sg+MXwv7nt4FvMWLqGubuXq8717dOV3P/4Unx/7R/+rRi326wt/xSc+9Wk27ty5+cZSAz169uLm2+6ge/dNmTH9Cb71tRO5+vob6dLF5Fat11ZzTt5Yy7GNgc8BmwNrLU5SShOACQAb7Xpiu/ubdf6iV+hXs+nq/b41mzJv8SuN2ixY/ArHfOdiADpv1ImRH/3A6nklm3TekL9d8CVOH38DUx5/NrNxKx961dTwwoL/zU9atHAhNTU1jdv0quGFFxZQ07s3tbW1vP7aa3TvvimPP/Yof791Muefew6vvfYqEVV06rQBo4/7RNaXoTLo1auGhS8sWL2/cOEL9Oy15nenV+Pvzuuv0b17dyKCTp06ATBwp53p178/zz37H3ba+X2ZXkN74yJsrZBSOnfVRn2xsRHwWeBKYNu2+MxKMG36c2y/ZU+22mJzOlZ34KiDduOmux5r1Gbz7p1Xfym/+9mDuPT6BwDoWN2Bv5z7ef5844Nc+/dHsh66cmCnnd/H888/y9y5c1ixfDm3TLqJQfsPadRm8P5DmHj9tQDcdutk9txrbyKCP/zxz9x82x3cfNsdHPfJT/O5MV+wMGlH6r87zzFv7lxWrFjO5JsnMXiN786g/Ydww/XXAfD3WyezR+m78/LLL1NXVwfA3DlzeP755+jXv3/Wl6AK02ZP60TEZsC3gOOAS4HdUkpLmn5X+1ZXt5JvnvVXbrjwK3SoCi69/gGenP0Cp33pEB6e8Tw3/eNx9tt9AOO+OpyU4N6HZ/GNn/wVgCMO3I2P7LY9m3XvzCeG7w3AmLF/5LGn55XzkpSh6upqTjl1LF8a8zlWrqxj5OFHsP32Axj/y1+w0047M3jIRzn8iCM59eTvcujQA+jarRtnn3NeuYetHKiuruak/zuNL3/hBFbWrWTE4Uew3fYDuPBXFzBwp50ZvP8QRo46ku+f8j2GDzuQrt268dOf/RyAhx+ayq9/9Uuqq6upqqri1LGn061b9/JeUDtQ6clJtMW8hIj4GTCK+tRkfErp9fXtoz3e1tG7Z8nUX5V7CCqolc7VUitt3DG7imHzT12R2Rf1pctGZ14JtdXTOt8GtgC+D8yPiFdL22sR8WobfaYkSe1DZLiVQZvc1kkp+YiyJElqFVeIlSSpYCp9zokJhyRJyhWTE0mSCsbkRJIkKUMmJ5IkFYzJiSRJUoZMTiRJKprKDk5MTiRJUr6YnEiSVDDOOZEkScqQyYkkSQVjciJJkpQhixNJkpQr3taRJKlgvK0jSZKUIZMTSZIKxuREkiQpQyYnkiQVTWUHJyYnkiQpX0xOJEkqGOecSJIkZcjkRJKkgjE5kSRJypDJiSRJBWNyIkmSlCGTE0mSiqaygxOTE0mSlC8mJ5IkFYxzTiRJkjJkciJJUsGYnEiSJGXI5ESSpIIxOZEkScqQxYkkScoVb+tIklQw3taRJElah4gYGhEzI2JWRJy8lvPnRcQjpe3piFjaXJ8mJ5IkFU1OgpOI6ACMBw4A5gJTI2JiSmnGqjYppW82aP9VYNfm+jU5kSRJrbUnMCulNDultBy4EhjRRPvRwBXNdWpyIklSwWQ55yQixgBjGhyakFKaUHrdF5jT4NxcYK919LMVsA1wR3OfaXEiSZLWqVSITGi2YfOOAa5OKdU119DiRJKkgsnR0zrzgP4N9vuVjq3NMcBXWtKpc04kSVJrTQUGRMQ2EdGJ+gJk4pqNImJHYFPg/pZ0anIiSVLB5CU4SSnVRsSJwGSgA3BJSml6RIwDpqWUVhUqxwBXppRSS/q1OJEkSa2WUpoETFrj2Ng19k9fnz4tTiRJKpgczTlpE845kSRJuWJyIklSwVR4cGJyIkmS8sXkRJKkgnHOiSRJUoZMTiRJKpgKD05MTiRJUr6YnEiSVDBVVZUdnZicSJKkXLE4kSRJueJtHUmSCsYJsZIkSRkyOZEkqWBchE2SJClDJieSJBVMhQcnJieSJClfTE4kSSoY55xIkiRlyOREkqSCMTmRJEnKkMmJJEkFU+HBicmJJEnKF5MTSZIKxjknkiRJGTI5kSSpYCo8ODE5kSRJ+WJyIklSwTjnRJIkKUMmJ5IkFUyFBycmJ5IkKV8sTiRJUq54W0eSpIJxQqwkSVKGTE4kSSqYCg9OTE4kSVK+mJxIklQwzjmRJEnKkMmJJEkFU+HBSX6Lk/n3/aLcQ1CBbfGZP5d7CCqoh88bVe4hqKC27rFhuYdQMXJbnEiSpLVzzokkSVKGTE4kSSqYCg9OTE4kSVK+mJxIklQwzjmRJEnKkMmJJEkFU+HBicmJJEnKF5MTSZIKxjknkiRJGTI5kSSpYExOJEmSMmRxIkmSWi0ihkbEzIiYFREnr6PNxyNiRkRMj4hmf5nV2zqSJBVMXu7qREQHYDxwADAXmBoRE1NKMxq0GQCcAnw4pbQkIno116/JiSRJaq09gVkppdkppeXAlcCINdp8HhifUloCkFJa1FynFieSJBVMRGS5jYmIaQ22MQ2G0heY02B/bulYQzsAO0TEfRHxQEQMbe76vK0jSZLWKaU0AZjwDrqoBgYAg4F+wN0R8b6U0tKm3iBJkgokL3NOgHlA/wb7/UrHGpoLPJhSWgH8JyKepr5YmbquTr2tI0mSWmsqMCAitomITsAxwMQ12lxHfWpCRPSg/jbP7KY6NTmRJKlg8rIIW0qpNiJOBCYDHYBLUkrTI2IcMC2lNLF07sCImAHUAd9NKb3UVL8WJ5IkqdVSSpOASWscG9vgdQK+VdpaxOJEkqSCyUlw0maccyJJknLF5ESSpIKpqvDoxOREkiTlismJJEkFU+HBicmJJEnKF5MTSZIKJi/rnLQVkxNJkpQrJieSJBVMVWUHJyYnkiQpX0xOJEkqGOecSJIkZcjiRJIk5Yq3dSRJKpgKv6tjciJJkvLF5ESSpIIJKjs6MTmRJEm5YnIiSVLBuAibJElShkxOJEkqGBdhkyRJypDJiSRJBVPhwYnJiSRJyheTE0mSCqaqwqMTkxNJkpQrJieSJBVMhQcnJieSJClfTE4kSSoY1zmRJEnKkMmJJEkFU+HBicmJJEnKF5MTSZIKxnVOJEmSMmRxIkmScsXbOpIkFUxl39QxOZEkSTljciJJUsG4CJskSVKGTE4kSSqYqsoOTkxOJElSvpicSJJUMM45kSRJypDJiSRJBVPhwYnJiSRJyheTE0mSCqbS55yssziJiF8CaV3nU0pfa5MRSZKkdq2p5GRaZqOQJEktVunrnKyzOEkpXdpwPyI2Tim92fZDkiRJ7VmzE2IjYp+ImAE8Vdp/f0Rc2OYjkyRJaxURmW3l0JKndc4HDgJeAkgpPQrs14ZjkiRJ7ViLntZJKc1Zo3qqa5vhSJKk5lT4lJMWFSdzIuJDQIqIjsDXgSfbdliSJKm9asltnS8CXwH6AvOBD5T2JUlSGVRFZLY1JyKGRsTMiJgVESev5fzxEbE4Ih4pbZ9rrs9mk5OU0ovAcc2OTpIktSsR0QEYDxwAzAWmRsTElNKMNZr+JaV0Ykv7bcnTOttGxA2lqmdRRFwfEduu1+glSVIl2hOYlVKanVJaDlwJjHinnbbkts6fgb8CfYAtgKuAK97pB0uSpNaJyHKLMRExrcE2psFQ+gJzGuzPLR1b0xER8VhEXB0R/Zu7vpYUJxunlP6YUqotbX8CNmzB+yRJUsGllCaklHZvsE1Yzy5uALZOKe0C3AZc2kz7Jn9bZ7PSy5tLE1yupP63do4GJq3nwCRJ0rskRz/8Nw9omIT0Kx1bLaX0UoPdi4Gzm+u0qQmxD1FfjKz6E/hCw88CTmmuc0mSVNGmAgMiYhvqi5JjgGMbNoiIPimlBaXd4bRgOZKmfltnm9aPVZIktZW8BCcppdqIOBGYDHQALkkpTY+IccC0lNJE4GsRMRyoBV4Gjm+u3xatEBsROwMDaTDXJKV02XpfhSRJqigppUmsMd0jpTS2wetTWM+7Lc0WJxHxA2Aw9cXJJGAYcC9gcSJJUhm0ZHG0ImvJ0zpHAh8FXkgpfQZ4P9CtTUclSZLarZbc1lmWUloZEbUR0RVYROOZuXoX3X/fPZz3s5+wcmUdw0ceyac++/lG55cvX84PTzuZmU9Op2u37vz4rJ+zxRZ9qV2xgjPHjWXmUzOoravj4EOG8+kTxqzjU1SpPvq+Ppz5yQ/SoSr4413/5hc3Nl6k8YzjduMj760BYKNOHejZdUO2+eLVfOS9vTjjuA+ubjegT1c+d+F9THpobqbjV/lMfeA+fnP+WdStXMmwww7n6E+e0Oj84488xG9+cTaz//0M//fDs9h3/wNWnxu2765sve0AAHrV9OaHZ1+Q6djbowoPTlpUnEyLiO7ARdQ/wfM6cH9bDqq9qqur45yf/pgLfn0xvWpq+MxxR7PvoP3ZZrvtV7eZeN01dN2kK1dPnMxtt0xi/C/O5Yyzfs7tf5/M8uXLufyq6/nvsmUcc8RhHDDsELbYYm1r4agSVUVw9qd3Z9RZdzD/5WXcPu4gbnl4LjPnv7q6zamXP7z69ecP2IFdttoUgHufXMSg798MQPfOnXjonMO48/EFqH2oq6tj/Lln8pPzf0uPXjV89XPHsvdHBrPVNtutbtOzpjffPvVHXH3F25eo6LTBBvz60r9mOWRVuGZv66SUvpxSWppS+g31a+d/unR7R++yGU88Tr/+W9K3X386duzEAQcN4+677mjU5p677uDgw0YCsP/HDmTalAdIKREEy/67jNraWt566y06duxI586dy3AVKpcPbrc5/1n4Os8tfoMVdSv52wPPMeyD/dbZ/oh9tuKaB5572/ERe/bn748tYNnyurYcrnJk5pNPsEW//vTp24+OHTsy+KNDuf+euxq16d2nL9tuvwNV0ZLZAGprEZHZVg7r/JZFxG5rbsBmQHXp9TpFxB4R0bvB/qdKv8lzQYPF3bSGxYsW0qtm9R8bvWp6s3jxore1qeld36a6upouXTbhlaVLGfKxA9low4049IBBjBj2UY771Gfo1q17lsNXmfXZdCPmvfzG6v35L79Jn003XmvbfptvzJY9u3D39IVvO3f43ltxzf3PttUwlUMvLV5Ez17/+3dPj169eHHx278b67J8+XJO/Oxovv75T/DPu+9o/g1SM5q6rXNuE+cSMKSJ878FPgYQEfsBPwW+CnwAmED9JNu3Ka3XPwbg57/8NcevMd9C6zZ9+uNUdajixlvv4tXXXuWLn/0ke+y1D337OT1Ibzdq762YOOV5VqbU6HhNtw0Z2K87d3hLR+vhj9fcTI+eNSyYN5eTvvZ5tt52AFv47542Ven5VVOLsO3/DvrtkFJ6ufT6aGBCSuka4JqIeKSJz5xAffHCkjfr0rraVaqevWpYtPCF1fuLFr5Az5693tZm4Qsv0KumN7W1tbz++mt0696dW39zE/t8aF+qO3Zks802Z5cP7MqTM56wOGlHFixZRt/N/ncrb4vNNmbBkjfX2nbU3lvxvcumve34yL224qaH5lLb/v7xa9c279mLxYv+9++eFxctokfPmha/f1XbPn37scuuu/PvZ56yONE70lbFV4eIWFX4fBRomPO1aOG39ui9O+3MnOefY/68uaxYsZzbJt/MvoMb14j7DtqfSTdcB8Cdf7+V3ffYi4igpncfpk19AIBly97kicceZautt836ElRGD89+iW17b8KWPTvTsUMVo/beilsenve2dgP6dKV7505MeebFt507Yh9v6bRH79lxJ+bNfZ4X5s9lxYoV3HX7Lez9kUEteu9rr77K8uXLAXhl6RKmP/4IW/rvnjZX6XNO2qpQuAL4R0S8CCwD7gGIiO2BV9roMwuvurqa75x0Kl//8udZuXIlh444nG23G8CEC3/JjgN3Yr/BQzhs5BH88PsnceTwg+jatTs/+uk5ABx59Gh+/INTGX3EYaSUOHTE4QzY4T1lviJlqW5l4nuXTePq7+5Ph6rg8rtn89S8Vzhl1Pv4139e5pZ/1Rcqo/beir+tZSJs/x6d2WKzjbnvqUVvO6fK1qG6mq988xT+71tfYmXdSg48dCRbb7s9l140nh123Il99h3MzCefYNwp3+S1117lgfv+wWUXX8hFl1/L88/N5oKzf0RUVZFWruToT3ym0VM+UmtESm0T30bE3kAf4NaU0hulYzsAXVJKDzf5ZtrnbR29e7b7wl/KPQQV1MPnjSr3EFRQW/fYMLOY4RvXP5XZ35Hnj9gx8/ikJcvXB3AcsG1KaVxEbAn0TilNaep9KaUH1nLs6VaPVJIktQstua1zIbCS+qdzxgGvAdcAe7ThuCRJ0jpUuUIse6WUdouIfwGklJZERKc2HpckSWqnWlKcrIiIDtSvbUJE9KQ+SZEkSWVQrqdostKSR4kvAK4FekXEGcC9wJltOipJktRuNZucpJQuj4iHqF+vJICRKaUn23xkkiSpXWrJ0zpbAm8CNzQ8llJ6vi0HJkmS1s4JsXAT9fNNAtgQ2AaYCezUhuOSJEntVEtu67yv4X7pF4m/3GYjkiRJTarw+bDr/9s6pdVd92qDsUiSJLVozsm3GuxWAbsB89tsRJIkqUlVFR6dtGTOySYNXtdSPwflmrYZjiRJau+aLE5Ki69tklL6TkbjkSRJzVjvORkFs87ri4jqlFId8OEMxyNJktq5ppKTKdTPL3kkIiYCVwFvrDqZUvpbG49NkiStRYVPOWnRnJMNgZeo/1XiVeudJMDiRJIkveuaKk56lZ7UeYL/FSWrpDYdlSRJWqf2/LROB6ALjYuSVSxOJElSm2iqOFmQUhqX2UgkSVKLVHhw0uTTSBV+6ZIkKY+aSk4+mtkoJElSi1X6rxKvMzlJKb2c5UAkSZKgZY8SS5KkHKn0p3UqfQVcSZJUMCYnkiQVTIUHJyYnkiQpXyxOJElSrnhbR5Kkgmm3jxJLkiSVg8mJJEkFExW+iLvJiSRJyhWTE0mSCsY5J5IkSRkyOZEkqWBMTiRJkjJkciJJUsFEha9fb3IiSZJyxeREkqSCcc6JJElShkxOJEkqmAqfcmJyIkmS8sXiRJKkgqmKyGxrTkQMjYiZETErIk5uot0REZEiYvdmr289/zwkSZIAiIgOwHhgGDAQGB0RA9fSbhPg68CDLenX4kSSpIKpiuy2ZuwJzEopzU4pLQeuBEaspd2PgLOA/7bo+tbjz0KSJLUzETEmIqY12MY0ON0XmNNgf27pWMP37wb0Tynd1NLP9GkdSZIKJsundVJKE4AJrXlvRFQBPweOX5/3mZxIkqTWmgf0b7Dfr3RslU2AnYG7IuJZYG9gYnOTYi1OJElSa00FBkTENhHRCTgGmLjqZErplZRSj5TS1imlrYEHgOEppWlNdeptHUmSCqaKfKzCllKqjYgTgclAB+CSlNL0iBgHTEspTWy6h7WzOJEkSa2WUpoETFrj2Nh1tB3ckj4tTiRJKhiXr5ckScqQyYkkSQXTgsXRCs3kRJIk5YrJiSRJBdOSH+QrMpMTSZKUKyYnkiQVTIUHJyYnkiQpX0xOJEkqGOecSJIkZcjkRJKkgqnw4MTkRJIk5YvJiSRJBVPpyUKlX58kSSoYkxNJkgomKnzSicmJJEnKFZMTSZIKprJzE5MTSZKUMxYnkiQpV7ytI0lSwbh8vSRJUoZMTiRJKpjKzk1MTiRJUs6YnEiSVDAVPuXE5ESSJOWLyYkkSQXj8vWSJEkZMjmRJKlgKj1ZqPTrkyRJBWNyIklSwTjnRJIkKUMmJ5IkFUxl5yYmJ5IkKWdMTiRJKhjnnEiSJGUot8lJxw7WTWq9xy44otxDUEENOP4P5R6CCmrZ9V/I7LMq/W/ISr8+SZJUMLlNTiRJ0to550SSJClDFieSJClXvK0jSVLBVPZNHZMTSZKUMyYnkiQVTIXPhzU5kSRJ+WJyIklSwVRV+KwTkxNJkpQrJieSJBWMc04kSZIyZHIiSVLBhHNOJEmSsmNyIklSwTjnRJIkKUMWJ5IkFUwVkdnWnIgYGhEzI2JWRJy8lvNfjIjHI+KRiLg3IgY2f32SJEmtEBEdgPHAMGAgMHotxcefU0rvSyl9ADgb+Hlz/TrnRJKkgsnRnJM9gVkppdkAEXElMAKYsapBSunVBu07A6m5Ti1OJEnSOkXEGGBMg0MTUkoTSq/7AnManJsL7LWWPr4CfAvoBAxp7jMtTiRJKpgsk5NSITKh2YZN9zEeGB8RxwLfBz7dVHvnnEiSpNaaB/RvsN+vdGxdrgRGNtepxYkkSQUTGf6vGVOBARGxTUR0Ao4BJjYaa8SABruHAM8016m3dSRJUquklGoj4kRgMtABuCSlND0ixgHTUkoTgRMj4mPACmAJzdzSAYsTSZL0DqSUJgGT1jg2tsHrr69vnxYnkiQVTFV+HiVuE845kSRJuWJyIklSwbRgomqhmZxIkqRcMTmRJKlgcrR8fZswOZEkSbliciJJUsE450SSJClDJieSJBWM65xIkiRlyOREkqSCcc6JJElShkxOJEkqGNc5kSRJypDJiSRJBVPhwYnJiSRJyheTE0mSCqaqwiedmJxIkqRcMTmRJKlgKjs3MTmRJEk5Y3EiSZJyxds6kiQVTYXf1zE5kSRJuWJyIklSwfjDf5IkSRkyOZEkqWAqfA02kxNJkpQvJieSJBVMhQcnJieSJClfTE4kSSqaCo9OTE4kSVKumJxIklQwrnMiSZKUIZMTSZIKxnVOJEmSMmRyIklSwVR4cGJyIkmS8sXkRJKkoqnw6MTkRJIk5YrJiSRJBeM6J5IkSRmyOJEkSbnibR1JkgrGRdgkSZIyZHIiSVLBVHhwYnIiSZLyxeREkqSiqfDoxOREkiTlismJJEkF4yJskiRJGTI5kSSpYFznRJIkKUMWJ5IkFUxkuDU7loihETEzImZFxMlrOf+tiJgREY9FxO0RsVVzfVqcSJKkVomIDsB4YBgwEBgdEQPXaPYvYPeU0i7A1cDZzfVrcSJJUtHkJzrZE5iVUpqdUloOXAmMaNggpXRnSunN0u4DQL/mOrU4kSRJ6xQRYyJiWoNtTIPTfYE5Dfbnlo6tywnAzc19pk/rSJJUMFmuc5JSmgBMeKf9RMQngN2BQc21tTiRJEmtNQ/o32C/X+lYIxHxMeBUYFBK6a3mOrU4kSSpYHK0zslUYEBEbEN9UXIMcGzDBhGxK/BbYGhKaVFLOrU4yZl/3nsP55x1BnUrVzJy1JF85oQxjc4vX76csaeexJMzptOtW3d++rOfs0Xf+rlFzzw9kzPGjeWNN94gIvjjFVezwQYblOMyVCZT7r+X8eedxcqVKzl4+ChGf+qERucf+9c0xp93NrP//Qzf/9FZDBpyIAALF8xn7EnfIKVEbW0thx81msNGfbwcl6AyOWDX/pzz+Q/RoSr4w21Pcc41jzQ6379HFy76xmC6dd6ADlXBaZc9yOSH5jDk/X350af2olN1FctrV/J/f3iAfzw+vzwXocyllGoj4kRgMtABuCSlND0ixgHTUkoTgZ8BXYCror6qej6lNLypfi1OcqSuro6fnjmOCydcQk1NDZ8cfRSDBg9h2+22X93mur9dTdeuXbn+pluZfPNNXHD+ufz0Z+dRW1vL90/5Lj8682x2eM+OLF26hOpq/+9tT+rq6rjgnDM5+4IJ9OxVw5c/M5p99h3M1ttst7pNr5o+fO+0H3PVn//Q6L2b9ejJLy/+E506dWLZm29ywrGj2GffwfTo2Svjq1A5VFUF53/hwxzyg5uY99Ib3HvOKG6c8ixPzVm6us1JH9+Na+6dzUW3zGDH/t257rSD2XHMn3np1f9y5Bm3sODlNxm45abccPohbPfZP5XvYtqJ/AQnkFKaBExa49jYBq8/tr59vutP60TE+Ij48Lvdb3sw/YnH6L/llvTr15+OHTtx4NCDuevO2xu1+cddt3Po8JEAfPSAg5jy4P2klHjg/vsYsMN72OE9OwLQvfumdOjQIetLUBk9NeMJ+vbbki369qNjx47sf8BQ/nn3nY3a9N6iL9sN2IGIxv/od+zYkU6dOgGwfMVyUlqZ2bhVfnsM6MW/X3iVZxe+xoralVx1zywO3XPrRm1SSnTduCMA3TbegAVL3gDg0f+8xIKX658SnfH8Ejbs1IFO1T4IqnemLb5BTwPnRMSzEXF26V6TWmDRwoXU1PRZvV9T05vFixY2arN44aLVbaqrq+nSZROWLl3K888+S0TwlS+ewLEfH8Wll1yc6dhVfi8uXkjPXjWr93v2quHFxS26vQvAooUv8LnjjmD08AM5+pOfNTVpR7bYfGPmvvj66v15L71B3807N2pzxpUPccygAcz63XFcO3YY35pw39v6OfxD2/DI7BdZXmtxq3fmXS9OUkq/SCntQ/2jQi8Bl0TEUxHxg4jYoan3NnyW+pKL3/FTS+1KbV0tjzz8ED/+yTn87tLLufOO25jywP3lHpYKpFdNby6+/Bouu/pGbp00kZdfeqncQ1KOfHzf7fjTHU+z/QmXc/i4m/ndN4c0mpT53v6b8uNP7cWJF95TvkG2J/lZhK1NtFn2llJ6LqV0VkppV2A0MBJ4spn3TEgp7Z5S2v2znxvTVNOK1KumhoULF6zeX7jwhUb/JQzQs6bX6ja1tbW8/vprdO/enZqa3uz6wd3ZdNNN2WijjfjwvoN46skZmY5f5dWjZ02jpG3xooWtSj969OzFNttuz+OPPvRuDk85Nv+lN+nXo8vq/b6bd2beS280avPpA3bkmvv+DcCDMxeyYccO9Oi64er2fznlQD53/p3854VXsxu4KlabFScRUR0Rh0XE5dSvBjcTGNVWn1cJBu70PuY89xzz5s5lxYrl3HrLJAYNHtKozaDBQ7hx4nUA3H7bZPbYc28ign0+/BFmPfMMy5Yto7a2loenTWWb7bZby6eoUu343p2YN+c5Fsyfy4oVK7jztlv40L6DW/TexYte4K3//heA1159lccf/Rf9t9y67QarXJn2zCK279ONrXptQsfqKo7ad3tumvJcozZzFr/O4F3qF/58T7/ubNipA4tf+S/dOnfib6cN47TLpnD/UwvX1r3aQGT4v7JcX0rp3e0w4gDqk5KDgSnUr7N/fUrpjSbfuIbX33qXB1YQ997zD849+0zq6lYyYuQRnDDmi/x6/AUMHLgzg/YfwltvvcVp//c9Zj71JN26dePMs39Ov371699MunEiv//dBILgw/vux9e/9d0yX035LH1zebmHUBYP/vMexp93NitX1jHs0JEc95kx/H7CeN6z40A+tN/+PDXjCX5w0jd4/bVX6dhpAzbbvAeXXHEt0x68n99ccA4RQUqJkUeN5tCRR5b7cspiwPF/KPcQyuKgD/bnZyfUP0p86e0zOfuqf3Hasbvz8KzF3DTlOXbs350LvzKIzht2JKXEqZc+yO2PzOWko3blu0fuyqz5r6zu67DTb2LxK/8t49WUx7Lrv5DZ3+RPLXgzs78jd+yzceYVSlsUJ3cAfwauSSktaW0/7bU40bujvRYneufaa3Gidy7L4mTmC9kVJ+/pnX1x8q4vhJFSGtJ8K0mSpLVzlS5JkgomT4uwtQVXypEkSbliciJJUtFUeHRiciJJknLF5ESSpIIp1/ojWTE5kSRJuWJyIklSwURlBycmJ5IkKV9MTiRJKpgKD05MTiRJUr6YnEiSVDQVHp2YnEiSpFwxOZEkqWBc50SSJClDFieSJClXvK0jSVLBuAibJElShkxOJEkqmAoPTkxOJElSvpicSJJUNBUenZicSJKkXDE5kSSpYFyETZIkKUMmJ5IkFYzrnEiSJGXI5ESSpIKp8ODE5ESSJOWLyYkkSQXjnBNJkqQMmZxIklQ4lR2dmJxIkqRcMTmRJKlgnHMiSZKUIZMTSZIKpsKDE5MTSZKULxYnkiQpV7ytI0lSwTghVpIkKUMmJ5IkFUxU+JRYkxNJkpQrJieSJBVNZQcnJieSJClfTE4kSSqYCg9OTE4kSVK+WJxIklQwEdltzY8lhkbEzIiYFREnr+X8fhHxcETURsSRLbk+ixNJktQqEdEBGA8MAwYCoyNi4BrNngeOB/7c0n6dcyJJUsHkaJ2TPYFZKaXZABFxJTACmLGqQUrp2dK5lS3t1OREkiStU0SMiYhpDbYxDU73BeY02J9bOvaOmJxIklQ0GQYnKaUJwITsPtHkRJIktd48oH+D/X6lY++IxYkkSQUTGW7NmAoMiIhtIqITcAww8Z1en8WJJElqlZRSLXAiMBl4EvhrSml6RIyLiOEAEbFHRMwFjgJ+GxHTm+vXOSeSJBVMS9YfyUpKaRIwaY1jYxu8nkr97Z4WMzmRJEm5YnIiSVLB5GidkzZhciJJknLF4kSSJOWKt3UkSSqYPE2IbQsmJ5IkKVcsTiRJUq5YnEiSpFxxzokkSQXjnBNJkqQMmZxIklQwLsImSZKUIZMTSZIKxjknkiRJGTI5kSSpYCo8ODE5kSRJ+WJyIklS0VR4dGJyIkmScsXkRJKkgnGdE0mSpAyZnEiSVDCucyJJkpQhkxNJkgqmwoMTkxNJkpQvFieSJClXvK0jSVLRVPh9HZMTSZKUKyYnkiQVjIuwSZIkZcjkRJKkgnERNkmSpAxFSqncY1ArRMSYlNKEco9DxeN3R63ld0dZMTkprjHlHoAKy++OWsvvjjJhcSJJknLF4kSSJOWKxUlxed9XreV3R63ld0eZcEKsJEnKFZMTSZKUKxYnkiQpVyxOCiQiUkSc22D/OxFxehmHpIKJiFMjYnpEPBYRj0TEXuUek/ItIu6MiIPWOPaNiPh1ucakymdxUixvAaMioke5B6LiiYh9gEOB3VJKuwAfA+aUd1QqgCuAY9Y4dkzpuNQmLE6KpZb62fLfLPdAVEh9gBdTSm8BpJReTCnNL/OYlH9XA4dERCeAiNga2AK4p5yDUmWzOCme8cBxEdGt3ANR4dwK9I+IpyPiwogYVO4BKf9SSi8DU4BhpUPHAH9NPuqpNmRxUjAppVeBy4CvlXssKpaU0uvAB6lfgnwx8JeIOL6sg1JRNLy14y0dtTnXOSmQiHg9pdQlIjYDHgZ+T/3/h6eXd2Qqoog4Evh0Sumwco9F+RYRXYDZwFDgypTSDmUekiqcyUkBlWLWvwInlHssKo6IeE9EDGhw6APAc2UajgqklLrdCVyCqYkyYHFSXOcCPrWj9dEFuDQiZkTEY8BA4PTyDkkFcgXwfixOlAFv60iSpFwxOZEkSblicSJJknLF4kSSJOWKxYkkScoVixNJkpQrFidSG4uIutIvAD8REVdFxMbvoK8/lBZPIyIujoiBTbQdHBEfasVnPLu2H5dc1/E12ry+np91ekR8Z33HKKmyWZxIbW9ZSukDKaWdgeXAFxuejIjq1nSaUvpcSmlGE00GA+tdnEhSuVmcSNm6B9i+lGrcExETgRkR0SEifhYRUyPisYj4AkDU+1VEzIyIvwO9VnUUEXdFxO6l10Mj4uGIeDQibi/9cuwXgW+WUpt9I6JnRFxT+oypEfHh0ns3j4hbI2J6RFwMRHMXERHXRcRDpfeMWePceaXjt0dEz9Kx7SLiltJ77omIHd+VP01JFalV/8Umaf2VEpJhwC2lQ7sBO6eU/lP6C/6VlNIeEbEBcF9E3ArsCryH+tVca4AZ1C8h3rDfnsBFwH6lvjZLKb0cEb8BXk8pnVNq92fgvJTSvRGxJTAZeC/wA+DelNK4iDiElv0swmdLn7ERMDUirkkpvQR0BqallL4ZEWNLfZ8ITAC+mFJ6JiL2Ai4EhrTij1FSO2BxIrW9jSLikdLre4DfUX+7ZUpK6T+l4wcCu6yaTwJ0AwYA+wFXpJTqgPkRccda+t8buHtVX6XfXlqbjwEDI1YHI11LP+i2HzCq9N6bImJJC67paxFxeOl1/9JYXwJWAn8pHf8T8LfSZ3wIuKrBZ2/Qgs+Q1E5ZnEhtb1lK6QMND5T+kn6j4SHgqymlyWu0O/hdHEcVsHdK6b9rGUuLRcRg6gudfVJKb0bEXcCG62ieSp+7dM0/A0laF+ecSPkwGfhSRHQEiIgdIqIzcDdwdGlOSh9g/7W89wFgv4jYpvTezUrHXwM2adDuVuCrq3Yi4gOll3cDx5aODQM2bWas3YAlpcJkR+qTm1WqgFXpz7HU3y56FfhPRBxV+oyIiPc38xmS2jGLEykfLqZ+PsnDEfEE8Fvqk81rgWdK5y4D7l/zjSmlxcAY6m+hPMr/bqvcABy+akIs8DVg99KE2xn876mhH1Jf3Eyn/vbO882M9RagOiKeBH5KfXG0yhvAnqVrGAKMKx0/DjihNL7pwIgW/JlIaqf8VWJJkpQrJieSJClXLE4kSVKuWJxIkqRcsTiRJEm5YnEiSZJyxeJEkiTlisWJJEnKlf8HGNmy8fRuGQsAAAAASUVORK5CYII=",
      "text/plain": [
       "<Figure size 576x576 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "y_pred = KNN.predict(x_valid[:,:,0])\n",
    "plot_confusion_matrix(y_valid,y_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       0.99      0.92      0.95     20924\n",
      "           1       0.44      0.77      0.56      1032\n",
      "           2       0.55      0.82      0.66      1703\n",
      "\n",
      "    accuracy                           0.90     23659\n",
      "   macro avg       0.66      0.83      0.72     23659\n",
      "weighted avg       0.94      0.90      0.91     23659\n",
      "\n"
     ]
    }
   ],
   "source": [
    "print(classification_report(y_valid, y_pred))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h3> RFC "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 44955/44955 [00:02<00:00, 22126.87it/s]\n",
      "100%|██████████| 23659/23659 [00:01<00:00, 22699.57it/s]\n",
      "100%|██████████| 23659/23659 [00:01<00:00, 22751.87it/s]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "RandomForestClassifier(n_estimators=200, random_state=123)"
      ]
     },
     "execution_count": 63,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.ensemble import RandomForestClassifier\n",
    "\n",
    "x_train,y_train = create_input(training_patches_df)\n",
    "x_valid,y_valid = create_input(validation_balanced_patches_df)\n",
    "x_test,y_test = create_input(validation_patches_df)\n",
    "\n",
    "\n",
    "rf = RandomForestClassifier(\n",
    "    n_estimators= 200,\n",
    "    random_state= SEED,\n",
    ")\n",
    "\n",
    "rf.fit(x_train[:,:,0],y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 576x576 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import seaborn as sns \n",
    "def plot_confusion_matrix(y_true, y_pred):\n",
    "  \"\"\"\n",
    "    :param ndarray y_true: contains the ground truth labels in the shape (n_samples,)\n",
    "    :param ndarray y_pred: contains the predicted labels in the shape (n_samples,)\n",
    "  \"\"\"\n",
    "  y_true = np.argmax(y_true,axis=1)\n",
    "  y_pred = np.argmax(y_pred,axis=1)\n",
    "  \n",
    "  con_mat = tf.math.confusion_matrix(labels=y_true, predictions=y_pred).numpy()\n",
    "  con_mat_norm = np.around(con_mat.astype('float') / con_mat.sum(axis=1)[:, np.newaxis], decimals=2)\n",
    "  con_mat_df = pd.DataFrame(con_mat_norm,\n",
    "                  index = classes, \n",
    "                  columns = classes)\n",
    "  figure = plt.figure(figsize=(8, 8))\n",
    "  sns.heatmap(con_mat_df, annot=True,cmap=plt.cm.Blues)\n",
    "  plt.tight_layout()\n",
    "  plt.ylabel('True label')\n",
    "  plt.xlabel('Predicted label')\n",
    "  plt.show()\n",
    "\n",
    "y_pred = rf.predict(x_valid[:,:,0])\n",
    "plot_confusion_matrix(y_valid,y_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       1.00      0.86      0.92     20924\n",
      "           1       0.65      0.70      0.68      1032\n",
      "           2       0.76      0.84      0.80      1703\n",
      "\n",
      "   micro avg       0.96      0.85      0.90     23659\n",
      "   macro avg       0.80      0.80      0.80     23659\n",
      "weighted avg       0.96      0.85      0.90     23659\n",
      " samples avg       0.85      0.85      0.85     23659\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\simon\\anaconda3\\envs\\tf-gpu\\lib\\site-packages\\sklearn\\metrics\\_classification.py:1308: UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in samples with no predicted labels. Use `zero_division` parameter to control this behavior.\n",
      "  _warn_prf(average, modifier, msg_start, len(result))\n"
     ]
    }
   ],
   "source": [
    "print(classification_report(y_valid, y_pred))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "95    S109\n",
       "4     S005\n",
       "90    S104\n",
       "27    S028\n",
       "32    S033\n",
       "73    S087\n",
       "5     S006\n",
       "87    S101\n",
       "54    S055\n",
       "63    S064\n",
       "Name: sigId, dtype: object"
      ]
     },
     "execution_count": 66,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "validation_df['sigId']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "interpreter": {
   "hash": "57032db1ad97fb6277226b3fe4aea1573f683021aa87982e2676dc61901b587a"
  },
  "kernelspec": {
   "display_name": "Python 3.9.7 64-bit ('tf-gpu': conda)",
   "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": "3.9.7"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}