Training.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "source": [
    "# Data Generator\n",
    "Data Generator -> only load batches into GPU, not total dataset"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "outputs": [],
   "source": [
    "import os\n",
    "import pandas as pd\n",
    "from copy import deepcopy\n",
    "import numpy as np\n",
    "import random\n",
    "from tensorflow import keras\n",
    "import tensorflow as tf\n",
    "\n",
    "class ASLDataGenerator(keras.utils.Sequence):\n",
    "\n",
    "    def __init__(self, dataset_path: str, max_phrase_length: int, max_sign_length: int, batch_size: int, train: bool = True):\n",
    "        super().__init__()\n",
    "        self.ds_path = dataset_path\n",
    "        self.files = os.listdir(dataset_path)\n",
    "        self.max_phrase_length = max_phrase_length\n",
    "        self.max_sign_length = max_sign_length\n",
    "        self.unwanted_columns = ['sequence_id', 'frame', 'participant_id', 'phrase']\n",
    "        self.train = train\n",
    "        self.batch_size = batch_size\n",
    "\n",
    "    def __len__(self):\n",
    "        return len(self.files)//self.batch_size\n",
    "\n",
    "    def shuffle(self):\n",
    "        random.shuffle(self.files)\n",
    "\n",
    "    def on_epoch_end(self):\n",
    "        if self.train:\n",
    "            self.shuffle()\n",
    "\n",
    "    def __getitem__(self, idx: int):\n",
    "        batch_signs = []\n",
    "        batch_contexts = []\n",
    "        batch_phrases = []\n",
    "        for i in range(idx, idx+self.batch_size):\n",
    "            df_phrase = pd.read_pickle(self.ds_path + '/' + self.files[idx])\n",
    "            phrase = df_phrase.iloc[0].phrase\n",
    "            if len(phrase) > self.max_phrase_length-1:\n",
    "                # check if phrase length longer than max_phrase_length\n",
    "                print('Houston we have a problem!')\n",
    "\n",
    "            # stop symbol is 61\n",
    "            phrase.append(61)\n",
    "\n",
    "            while len(phrase) < self.max_phrase_length:\n",
    "                #padding value for remaining phrase is 59, to fill up missing phrase values until max_phrase_length\n",
    "                phrase.append(59)\n",
    "\n",
    "            context = deepcopy(phrase)\n",
    "            # start symbol is 60\n",
    "            context.insert(0, 60)\n",
    "            context = context[:-1]\n",
    "\n",
    "            df_phrase = df_phrase.drop(self.unwanted_columns, axis=1)\n",
    "            signs = df_phrase.to_numpy(copy=True)\n",
    "\n",
    "            if signs.shape[0] > self.max_sign_length:\n",
    "                # check if length of sign data over max_sign_length\n",
    "                print('Hooooouuuuusteeeeeen!')\n",
    "\n",
    "            # fill up missing values of mediapipe data until max_sign_length with padding value 0\n",
    "            signs = np.pad(signs, [(0, self.max_sign_length-signs.shape[0]), (0,0)], 'edge')\n",
    "            batch_signs.append(signs)\n",
    "            batch_contexts.append(context)\n",
    "            batch_phrases.append(phrase)\n",
    "\n",
    "        batch_signs = np.array(batch_signs)\n",
    "        batch_contexts = np.array(batch_contexts)\n",
    "        batch_phrases = np.array(batch_phrases)\n",
    "\n",
    "        return [batch_signs, batch_contexts], batch_phrases"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Execute this cell for version with simple preprocessing (-1):"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "outputs": [],
   "source": [
    "# Data with -1 as preprocessing\n",
    "# get output from Data Generator\n",
    "# for validation and test data we did a participant Wise Split!\n",
    "\n",
    "MAX_PHRASE_LENGTH = 100  # experimentally determined untill no errors occured\n",
    "MAX_SIGN_LENGTH = 900  # experimentally determined untill no errors occured\n",
    "BATCH_SIZE= 32\n",
    "\n",
    "data_gen_train = ASLDataGenerator('./dataset complete/preprocessed_files_data_generator/train_ones', MAX_PHRASE_LENGTH, MAX_SIGN_LENGTH, batch_size=BATCH_SIZE)\n",
    "data_gen_train.shuffle()\n",
    "\n",
    "data_gen_test = ASLDataGenerator('./dataset complete/preprocessed_files_data_generator/test_ones', MAX_PHRASE_LENGTH, MAX_SIGN_LENGTH, train=False, batch_size=BATCH_SIZE)\n",
    "\n",
    "data_gen_val = ASLDataGenerator('./dataset complete/preprocessed_files_data_generator/val_ones', MAX_PHRASE_LENGTH, MAX_SIGN_LENGTH, train=False, batch_size=BATCH_SIZE)"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-17T17:47:07.026061800Z",
     "start_time": "2023-07-17T17:47:06.906712500Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "outputs": [],
   "source": [
    "# get input and phrase from one example of the output of the Data Generator, to later get shape\n",
    "input, phrase = data_gen_train[0]"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-17T17:47:09.638393200Z",
     "start_time": "2023-07-17T17:47:09.583612400Z"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Execute this cell for version with more complex preprocessing (mirroring):"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "outputs": [],
   "source": [
    "# mirrored complex preprocessing data\n",
    "# get output from Data generator\n",
    "\n",
    "import tensorflow as tf\n",
    "MAX_PHRASE_LENGTH = 100\n",
    "MAX_SIGN_LENGTH = 900\n",
    "BATCH_SIZE=32\n",
    "\n",
    "data_gen_train = ASLDataGenerator('./dataset complete/preprocessed_files_data_generator/train', MAX_PHRASE_LENGTH, MAX_SIGN_LENGTH, batch_size=BATCH_SIZE)\n",
    "data_gen_train.shuffle()\n",
    "\n",
    "data_gen_test = ASLDataGenerator('./dataset complete/preprocessed_files_data_generator/test', MAX_PHRASE_LENGTH, MAX_SIGN_LENGTH, train=False, batch_size=BATCH_SIZE)\n",
    "\n",
    "data_gen_val = ASLDataGenerator('./dataset complete/preprocessed_files_data_generator/validation_files', MAX_PHRASE_LENGTH, MAX_SIGN_LENGTH, train=False, batch_size=BATCH_SIZE)"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-14T20:55:02.889516200Z",
     "start_time": "2023-07-14T20:55:02.817453400Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "outputs": [],
   "source": [
    "# get input and phrase from one example of the output of the Data Generator, to later get shape\n",
    "input, phrase = data_gen_train[0]"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# Train Transformer using Generator"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "outputs": [],
   "source": [
    "# We used the Parameter values returned from the Hyperparameter tuning\n",
    "\n",
    "import json\n",
    "\n",
    "with open (\"./dataset complete/character_to_prediction_index.json\", \"r\") as f:\n",
    "    characters = json.load(f)\n",
    "\n",
    "# Tuned Hyperparameters\n",
    "EPOCHS = 40\n",
    "d_model = 141  # hidden layer(s) dimensionality\n",
    "num_layers = 1 # how many encoders and decoders to stack\n",
    "num_heads = 3 # how many attention heads should every mha have\n",
    "ff_dim = 389 # how many neurons shall feed-forward layers have\n",
    "dropout_rate = 0.2867\n",
    "\n",
    "output_vocab_size = len(characters) + 3"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-17T17:47:15.128397100Z",
     "start_time": "2023-07-17T17:47:15.119388500Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "outputs": [
    {
     "data": {
      "text/plain": "62"
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "output_vocab_size"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-17T15:46:36.935443Z",
     "start_time": "2023-07-17T15:46:36.923427700Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "outputs": [],
   "source": [
    "from transformer_update import get_compiled_transformer\n",
    "\n",
    "transformer = get_compiled_transformer(\n",
    "    sign_shape=(MAX_SIGN_LENGTH, input[0].shape[-1]),\n",
    "    context_shape=(input[1].shape[-1]),\n",
    "    d_model=d_model,\n",
    "    num_layers=num_layers,\n",
    "    num_heads=num_heads,\n",
    "    ff_dim=ff_dim,\n",
    "    dropout_rate=dropout_rate,\n",
    "    output_vocab_size=output_vocab_size\n",
    ")"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-17T17:47:19.266639200Z",
     "start_time": "2023-07-17T17:47:17.864291900Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model\"\n",
      "__________________________________________________________________________________________________\n",
      " Layer (type)                   Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      " input_1 (InputLayer)           [(None, 900, 144)]   0           []                               \n",
      "                                                                                                  \n",
      " input_2 (InputLayer)           [(None, 100)]        0           []                               \n",
      "                                                                                                  \n",
      " transformer (Transformer)      (None, 100, 62)      979803      ['input_1[0][0]',                \n",
      "                                                                  'input_2[0][0]']                \n",
      "                                                                                                  \n",
      "==================================================================================================\n",
      "Total params: 979,803\n",
      "Trainable params: 979,803\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "transformer.summary()"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-17T15:46:43.426922800Z",
     "start_time": "2023-07-17T15:46:43.400836300Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Num GPUs Available:  1\n"
     ]
    }
   ],
   "source": [
    "print(\"Num GPUs Available: \", len(tf.config.list_physical_devices('GPU')))"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-16T09:16:21.641810800Z",
     "start_time": "2023-07-16T09:16:21.631801200Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 3.0283 - masked_accuracy: 0.1593\n",
      "Epoch 1: saving model to model_checkpoints\\cp-0001.ckpt\n",
      "2394/2394 [==============================] - 240s 99ms/step - loss: 3.0283 - masked_accuracy: 0.1593 - val_loss: 2.5257 - val_masked_accuracy: 0.2330\n",
      "Epoch 2/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.6653 - masked_accuracy: 0.2088\n",
      "Epoch 2: saving model to model_checkpoints\\cp-0002.ckpt\n",
      "2394/2394 [==============================] - 236s 99ms/step - loss: 2.6653 - masked_accuracy: 0.2088 - val_loss: 2.4890 - val_masked_accuracy: 0.2696\n",
      "Epoch 3/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.5596 - masked_accuracy: 0.2384\n",
      "Epoch 3: saving model to model_checkpoints\\cp-0003.ckpt\n",
      "2394/2394 [==============================] - 237s 99ms/step - loss: 2.5596 - masked_accuracy: 0.2384 - val_loss: 2.3701 - val_masked_accuracy: 0.2870\n",
      "Epoch 4/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.4795 - masked_accuracy: 0.2538\n",
      "Epoch 4: saving model to model_checkpoints\\cp-0004.ckpt\n",
      "2394/2394 [==============================] - 236s 99ms/step - loss: 2.4795 - masked_accuracy: 0.2538 - val_loss: 2.2713 - val_masked_accuracy: 0.3128\n",
      "Epoch 5/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.4291 - masked_accuracy: 0.2723\n",
      "Epoch 5: saving model to model_checkpoints\\cp-0005.ckpt\n",
      "2394/2394 [==============================] - 235s 98ms/step - loss: 2.4291 - masked_accuracy: 0.2723 - val_loss: 2.1898 - val_masked_accuracy: 0.3362\n",
      "Epoch 6/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.4002 - masked_accuracy: 0.2766\n",
      "Epoch 6: saving model to model_checkpoints\\cp-0006.ckpt\n",
      "2394/2394 [==============================] - 236s 98ms/step - loss: 2.4002 - masked_accuracy: 0.2766 - val_loss: 2.1750 - val_masked_accuracy: 0.3430\n",
      "Epoch 7/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.3680 - masked_accuracy: 0.2884\n",
      "Epoch 7: saving model to model_checkpoints\\cp-0007.ckpt\n",
      "2394/2394 [==============================] - 236s 99ms/step - loss: 2.3680 - masked_accuracy: 0.2884 - val_loss: 2.1397 - val_masked_accuracy: 0.3486\n",
      "Epoch 8/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.3411 - masked_accuracy: 0.2947\n",
      "Epoch 8: saving model to model_checkpoints\\cp-0008.ckpt\n",
      "2394/2394 [==============================] - 236s 99ms/step - loss: 2.3411 - masked_accuracy: 0.2947 - val_loss: 2.1188 - val_masked_accuracy: 0.3640\n",
      "Epoch 9/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.2986 - masked_accuracy: 0.3057\n",
      "Epoch 9: saving model to model_checkpoints\\cp-0009.ckpt\n",
      "2394/2394 [==============================] - 236s 99ms/step - loss: 2.2986 - masked_accuracy: 0.3057 - val_loss: 2.0944 - val_masked_accuracy: 0.3586\n",
      "Epoch 10/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.2907 - masked_accuracy: 0.3071\n",
      "Epoch 10: saving model to model_checkpoints\\cp-0010.ckpt\n",
      "2394/2394 [==============================] - 236s 98ms/step - loss: 2.2907 - masked_accuracy: 0.3071 - val_loss: 2.0453 - val_masked_accuracy: 0.3751\n",
      "Epoch 11/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.2984 - masked_accuracy: 0.3095\n",
      "Epoch 11: saving model to model_checkpoints\\cp-0011.ckpt\n",
      "2394/2394 [==============================] - 236s 99ms/step - loss: 2.2984 - masked_accuracy: 0.3095 - val_loss: 2.0165 - val_masked_accuracy: 0.3862\n",
      "Epoch 12/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.2657 - masked_accuracy: 0.3155\n",
      "Epoch 12: saving model to model_checkpoints\\cp-0012.ckpt\n",
      "2394/2394 [==============================] - 236s 98ms/step - loss: 2.2657 - masked_accuracy: 0.3155 - val_loss: 2.0115 - val_masked_accuracy: 0.3810\n",
      "Epoch 13/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.2525 - masked_accuracy: 0.3187\n",
      "Epoch 13: saving model to model_checkpoints\\cp-0013.ckpt\n",
      "2394/2394 [==============================] - 236s 98ms/step - loss: 2.2525 - masked_accuracy: 0.3187 - val_loss: 2.0451 - val_masked_accuracy: 0.3794\n",
      "Epoch 14/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.2381 - masked_accuracy: 0.3241\n",
      "Epoch 14: saving model to model_checkpoints\\cp-0014.ckpt\n",
      "2394/2394 [==============================] - 237s 99ms/step - loss: 2.2381 - masked_accuracy: 0.3241 - val_loss: 1.9784 - val_masked_accuracy: 0.4057\n",
      "Epoch 15/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.2129 - masked_accuracy: 0.3317\n",
      "Epoch 15: saving model to model_checkpoints\\cp-0015.ckpt\n",
      "2394/2394 [==============================] - 236s 99ms/step - loss: 2.2129 - masked_accuracy: 0.3317 - val_loss: 1.8998 - val_masked_accuracy: 0.4320\n",
      "Epoch 16/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.1561 - masked_accuracy: 0.3496\n",
      "Epoch 16: saving model to model_checkpoints\\cp-0016.ckpt\n",
      "2394/2394 [==============================] - 236s 99ms/step - loss: 2.1561 - masked_accuracy: 0.3496 - val_loss: 1.8092 - val_masked_accuracy: 0.4549\n",
      "Epoch 17/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.1211 - masked_accuracy: 0.3614\n",
      "Epoch 17: saving model to model_checkpoints\\cp-0017.ckpt\n",
      "2394/2394 [==============================] - 235s 98ms/step - loss: 2.1211 - masked_accuracy: 0.3614 - val_loss: 1.7762 - val_masked_accuracy: 0.4567\n",
      "Epoch 18/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.0968 - masked_accuracy: 0.3654\n",
      "Epoch 18: saving model to model_checkpoints\\cp-0018.ckpt\n",
      "2394/2394 [==============================] - 235s 98ms/step - loss: 2.0968 - masked_accuracy: 0.3654 - val_loss: 1.7711 - val_masked_accuracy: 0.4585\n",
      "Epoch 19/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.0925 - masked_accuracy: 0.3714\n",
      "Epoch 19: saving model to model_checkpoints\\cp-0019.ckpt\n",
      "2394/2394 [==============================] - 235s 98ms/step - loss: 2.0925 - masked_accuracy: 0.3714 - val_loss: 1.7208 - val_masked_accuracy: 0.4746\n",
      "Epoch 20/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.0857 - masked_accuracy: 0.3712\n",
      "Epoch 20: saving model to model_checkpoints\\cp-0020.ckpt\n",
      "2394/2394 [==============================] - 236s 98ms/step - loss: 2.0857 - masked_accuracy: 0.3712 - val_loss: 1.7347 - val_masked_accuracy: 0.4724\n",
      "Epoch 21/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.0540 - masked_accuracy: 0.3800\n",
      "Epoch 21: saving model to model_checkpoints\\cp-0021.ckpt\n",
      "2394/2394 [==============================] - 236s 99ms/step - loss: 2.0540 - masked_accuracy: 0.3800 - val_loss: 1.7116 - val_masked_accuracy: 0.4775\n",
      "Epoch 22/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.0621 - masked_accuracy: 0.3779\n",
      "Epoch 22: saving model to model_checkpoints\\cp-0022.ckpt\n",
      "2394/2394 [==============================] - 236s 98ms/step - loss: 2.0621 - masked_accuracy: 0.3779 - val_loss: 1.6963 - val_masked_accuracy: 0.4916\n",
      "Epoch 23/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.0436 - masked_accuracy: 0.3818\n",
      "Epoch 23: saving model to model_checkpoints\\cp-0023.ckpt\n",
      "2394/2394 [==============================] - 235s 98ms/step - loss: 2.0436 - masked_accuracy: 0.3818 - val_loss: 1.6881 - val_masked_accuracy: 0.4849\n",
      "Epoch 24/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.0335 - masked_accuracy: 0.3857\n",
      "Epoch 24: saving model to model_checkpoints\\cp-0024.ckpt\n",
      "2394/2394 [==============================] - 235s 98ms/step - loss: 2.0335 - masked_accuracy: 0.3857 - val_loss: 1.6888 - val_masked_accuracy: 0.4894\n",
      "Epoch 25/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.0138 - masked_accuracy: 0.3910\n",
      "Epoch 25: saving model to model_checkpoints\\cp-0025.ckpt\n",
      "2394/2394 [==============================] - 235s 98ms/step - loss: 2.0138 - masked_accuracy: 0.3910 - val_loss: 1.6730 - val_masked_accuracy: 0.4951\n",
      "Epoch 26/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.0127 - masked_accuracy: 0.3918\n",
      "Epoch 26: saving model to model_checkpoints\\cp-0026.ckpt\n",
      "2394/2394 [==============================] - 235s 98ms/step - loss: 2.0127 - masked_accuracy: 0.3918 - val_loss: 1.6236 - val_masked_accuracy: 0.5047\n",
      "Epoch 27/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.0208 - masked_accuracy: 0.3890\n",
      "Epoch 27: saving model to model_checkpoints\\cp-0027.ckpt\n",
      "2394/2394 [==============================] - 235s 98ms/step - loss: 2.0208 - masked_accuracy: 0.3890 - val_loss: 1.6518 - val_masked_accuracy: 0.4896\n",
      "Epoch 28/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 2.0192 - masked_accuracy: 0.3888\n",
      "Epoch 28: saving model to model_checkpoints\\cp-0028.ckpt\n",
      "2394/2394 [==============================] - 236s 99ms/step - loss: 2.0192 - masked_accuracy: 0.3888 - val_loss: 1.6772 - val_masked_accuracy: 0.4900\n",
      "Epoch 29/40\n",
      "2394/2394 [==============================] - ETA: 0s - loss: 1.9907 - masked_accuracy: 0.4004\n",
      "Epoch 29: saving model to model_checkpoints\\cp-0029.ckpt\n",
      "2394/2394 [==============================] - 236s 99ms/step - loss: 1.9907 - masked_accuracy: 0.4004 - val_loss: 1.6351 - val_masked_accuracy: 0.5028\n",
      "482/482 [==============================] - 22s 45ms/step - loss: 1.6007 - masked_accuracy: 0.5128\n"
     ]
    },
    {
     "data": {
      "text/plain": "[1.600693702697754, 0.5128329992294312]"
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cp_callback = tf.keras.callbacks.ModelCheckpoint(\n",
    "    filepath=\"model_checkpoints/cp-{epoch:04d}.ckpt\", save_weights_only=True, verbose=1\n",
    ")\n",
    "es_callback = tf.keras.callbacks.EarlyStopping(\n",
    "    monitor='val_loss',\n",
    "    patience=3,\n",
    "    restore_best_weights=True\n",
    ")\n",
    "#shuffle_callback = ShuffleCallback()\n",
    "\n",
    "history = transformer.fit(data_gen_train, epochs=EPOCHS, batch_size=BATCH_SIZE, validation_data=data_gen_val, callbacks=[cp_callback, es_callback])\n",
    "\n",
    "transformer.evaluate(data_gen_test, batch_size=BATCH_SIZE)"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-17T19:41:51.051248Z",
     "start_time": "2023-07-17T17:47:26.219986600Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "outputs": [],
   "source": [
    "# saving weights to file, so we can integrate these in the python server of our application\n",
    "transformer.save_weights('./savedModelWeights/savedModelWeights')"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-17T19:41:51.159428600Z",
     "start_time": "2023-07-17T19:41:51.050247400Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "outputs": [
    {
     "data": {
      "text/plain": "<tensorflow.python.checkpoint.checkpoint.CheckpointLoadStatus at 0x237a0c13c40>"
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "restored = get_compiled_transformer(\n",
    "    sign_shape=(MAX_SIGN_LENGTH, input[0].shape[-1]),\n",
    "    context_shape=(input[1].shape[-1]),\n",
    "    d_model=d_model,\n",
    "    num_layers=num_layers,\n",
    "    num_heads=num_heads,\n",
    "    ff_dim=ff_dim,\n",
    "    dropout_rate=dropout_rate,\n",
    "    output_vocab_size=output_vocab_size\n",
    ")\n",
    "\n",
    "restored.load_weights('./savedModelWeights')"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-16T09:21:34.382750800Z",
     "start_time": "2023-07-16T09:21:33.853244100Z"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Show Plots for Training history\n",
    "Saved history to file to be able to display plots with upgraded matplotlib version"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dict_keys(['loss', 'masked_accuracy', 'val_loss', 'val_masked_accuracy'])\n"
     ]
    }
   ],
   "source": [
    "print(history.history.keys())"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-11T17:18:55.254085300Z",
     "start_time": "2023-07-11T17:18:55.250080200Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "outputs": [],
   "source": [
    "import pickle\n",
    "\n",
    "# Saved the history to a file using pickle\n",
    "with open(\"training_history.pkl\", \"wb\") as f:\n",
    "    pickle.dump(history.history, f)"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-17T20:58:37.146962200Z",
     "start_time": "2023-07-17T20:58:37.132965200Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "outputs": [],
   "source": [
    "import pickle\n",
    "\n",
    "# Load the history from the file\n",
    "with open(\"training_history.pkl\", \"rb\") as f:\n",
    "    loaded_history = pickle.load(f)\n",
    "loaded_history"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "plt.plot(loaded_history['masked_accuracy'])\n",
    "plt.plot(loaded_history['val_masked_accuracy'])\n",
    "plt.title('Accuracy over Epochs')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper left')\n",
    "plt.show()"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-17T21:03:05.038912700Z",
     "start_time": "2023-07-17T21:03:04.392783600Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(loaded_history['loss'])\n",
    "plt.plot(loaded_history['val_loss'])\n",
    "plt.title('Loss over Epochs')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper left')\n",
    "plt.show()"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-17T21:03:43.672586300Z",
     "start_time": "2023-07-17T21:03:43.560979900Z"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# Translate\n",
    "Test prediction and translation in Python, however only test data from Google available"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "outputs": [],
   "source": [
    "#test Translation within Python, however again based on test dataset\n",
    "\n",
    "import tensorflow as tf\n",
    "import numpy as np\n",
    "from copy import deepcopy\n",
    "\n",
    "MAX_PHRASE_LENGTH = 100\n",
    "\n",
    "class Translator:\n",
    "    def __init__(self, transformer):\n",
    "        self.transformer = transformer\n",
    "\n",
    "    def translate(self, signs, max_length=MAX_PHRASE_LENGTH):\n",
    "        print(signs.shape)\n",
    "        signs = tf.convert_to_tensor(signs, dtype=tf.float32, name='input_1')\n",
    "        signs = signs[tf.newaxis]\n",
    "\n",
    "        start = np.array([60])\n",
    "        end = np.array([59])\n",
    "\n",
    "        output = tf.TensorArray(dtype=tf.int64, size=0, dynamic_size=True)\n",
    "        output = output.write(0, start)\n",
    "\n",
    "        for i in tf.range(max_length):\n",
    "            context = deepcopy(output)\n",
    "            while context.size() < 100:\n",
    "                context = context.write(context.size(), end)\n",
    "            context = tf.transpose(context.stack())\n",
    "            context = tf.identity(context, name='input_2')\n",
    "            predictions = self.transformer([signs, context], training=False)\n",
    "\n",
    "            predictions = predictions[:, -1:, :]\n",
    "            predicted = tf.argmax(predictions, axis=-1)\n",
    "            output = output.write(i+1, predicted[0])\n",
    "\n",
    "            if predicted == end:\n",
    "                break\n",
    "\n",
    "        output_tensor = output.stack()  # Convert TensorArray to tensor\n",
    "        output_array = output_tensor.numpy()  # Convert tensor to NumPy array\n",
    "\n",
    "        return output_array"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-14T23:47:33.739614600Z",
     "start_time": "2023-07-14T23:47:33.731602600Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "outputs": [],
   "source": [
    "transformer = tf.saved_model.load('./savedModel')"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-11T21:29:52.769954500Z",
     "start_time": "2023-07-11T21:29:50.303822600Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "outputs": [],
   "source": [
    "translator = Translator(transformer)"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-14T23:47:36.186060200Z",
     "start_time": "2023-07-14T23:47:36.180054700Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(900, 144)\n",
      "[[60]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [46]\n",
      " [46]\n",
      " [46]\n",
      " [46]\n",
      " [46]\n",
      " [46]\n",
      " [46]\n",
      " [39]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [51]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [36]\n",
      " [ 0]]\n"
     ]
    }
   ],
   "source": [
    "output = translator.translate(input[0][0], MAX_PHRASE_LENGTH)\n",
    "print(output)"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-14T23:53:21.388385300Z",
     "start_time": "2023-07-14T23:53:18.353192200Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "outputs": [
    {
     "data": {
      "text/plain": "array([[33, 49, 32, ..., 59, 59, 59],\n       [33, 49, 32, ..., 59, 59, 59],\n       [33, 49, 32, ..., 59, 59, 59],\n       ...,\n       [33, 49, 32, ..., 59, 59, 59],\n       [33, 49, 32, ..., 59, 59, 59],\n       [33, 49, 32, ..., 59, 59, 59]])"
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "phrase"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-14T23:51:44.713819700Z",
     "start_time": "2023-07-14T23:51:44.705810900Z"
    }
   }
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}