Category_Analysis/models_tf.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "***\n",
    "# Tensorflow models\n",
    "In this notebook a simle Fully connected neural network and a bidirectional LSTM are tested.\n",
    "\n",
    "Both models show good training results, but fail at regularization.  This concludes to either not having enough training data, or some other error in the data."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "***\n",
    "## Simple Neural Network with embedding layer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# import models\n",
    "import sys\n",
    "sys.path.append('tf_models')\n",
    "from train import train\n",
    "# Add functions path\n",
    "sys.path.append('../Functions')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "from datasets import load_stratified_dataset\n",
    "from sklearn.model_selection import train_test_split\n",
    "df = load_stratified_dataset(path='../Datasets/dataset_categories/dataset_categories_train.csv', labels='category', samples_per_label=1000, random_seed=10)\n",
    "X_train, X_test, y_train, y_test = train_test_split(df.text, df.category)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(0.9842134118080139, 0.6979695558547974, 0.661142885684967)"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train(train_data=X_train, train_labels=y_train, test_data=X_test, test_labels=y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test different sequence lengths"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true,
    "jupyter": {
     "outputs_hidden": true,
     "source_hidden": true
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n"
     ]
    }
   ],
   "source": [
    "train_acc = []\n",
    "val_acc = []\n",
    "test_acc = []\n",
    "x_value = []\n",
    "for d in np.linspace(10,1000,10):\n",
    "    d = int(d)\n",
    "    print(d, end='\\r')\n",
    "    tr, v, te = train(sequence_length=d, train_data=X_train, train_labels=y_train, test_data=X_test, test_labels=y_test)\n",
    "    train_acc.append(tr)\n",
    "    val_acc.append(v)\n",
    "    test_acc.append(te)\n",
    "    x_value.append(d)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "jupyter": {
     "source_hidden": true
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(x_value, train_acc, label='train')\n",
    "plt.plot(x_value, val_acc, label='val')\n",
    "plt.plot(x_value, test_acc, label='test')\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test different embedding dimensions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true,
    "jupyter": {
     "outputs_hidden": true,
     "source_hidden": true
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n"
     ]
    }
   ],
   "source": [
    "train_acc = []\n",
    "val_acc = []\n",
    "test_acc = []\n",
    "x_value = []\n",
    "for d in np.linspace(6,30,10):\n",
    "    d = int(d)\n",
    "    print(d, end='\\r')\n",
    "    tr, v, te = train(embedding_dim=d, train_data=X_train, train_labels=y_train, test_data=X_test, test_labels=y_test)\n",
    "    train_acc.append(tr)\n",
    "    val_acc.append(v)\n",
    "    test_acc.append(te)\n",
    "    x_value.append(d)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "jupyter": {
     "source_hidden": true
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(x_value, train_acc, label='train')\n",
    "plt.plot(x_value, val_acc, label='val')\n",
    "plt.plot(x_value, test_acc, label='test')\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test different vocab sizes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true,
    "jupyter": {
     "outputs_hidden": true,
     "source_hidden": true
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n"
     ]
    }
   ],
   "source": [
    "train_acc = []\n",
    "val_acc = []\n",
    "test_acc = []\n",
    "x_value = []\n",
    "for d in np.linspace(5000,20000,10):\n",
    "    d = int(d)\n",
    "    print(d, end='\\r')\n",
    "    tr, v, te = train(vocab_size=d, train_data=X_train, train_labels=y_train, test_data=X_test, test_labels=y_test)\n",
    "    train_acc.append(tr)\n",
    "    val_acc.append(v)\n",
    "    test_acc.append(te)\n",
    "    x_value.append(d)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "jupyter": {
     "source_hidden": true
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(x_value, train_acc, label='train')\n",
    "plt.plot(x_value, val_acc, label='val')\n",
    "plt.plot(x_value, test_acc, label='test')\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test different dropouts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true,
    "jupyter": {
     "outputs_hidden": true,
     "source_hidden": true
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n"
     ]
    }
   ],
   "source": [
    "train_acc = []\n",
    "val_acc = []\n",
    "test_acc = []\n",
    "x_value = []\n",
    "for d in np.linspace(0.1, 0.5, 5):\n",
    "    print(d, end='\\r')\n",
    "    tr, v, te = train(dropout=d, train_data=X_train, train_labels=y_train, test_data=X_test, test_labels=y_test)\n",
    "    train_acc.append(tr)\n",
    "    val_acc.append(v)\n",
    "    test_acc.append(te)\n",
    "    x_value.append(d)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "jupyter": {
     "source_hidden": true
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(x_value, train_acc, label='train')\n",
    "plt.plot(x_value, val_acc, label='val')\n",
    "plt.plot(x_value, test_acc, label='test')\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test different regularizations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true,
    "jupyter": {
     "outputs_hidden": true,
     "source_hidden": true
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n"
     ]
    }
   ],
   "source": [
    "train_acc = []\n",
    "val_acc = []\n",
    "test_acc = []\n",
    "x_value = []\n",
    "for d in np.linspace(0, 3, 5):\n",
    "    print(d, end='\\r')\n",
    "    tr, v, te = train(regularization=d, train_data=X_train, train_labels=y_train, test_data=X_test, test_labels=y_test)\n",
    "    train_acc.append(tr)\n",
    "    val_acc.append(v)\n",
    "    test_acc.append(te)\n",
    "    x_value.append(d)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "jupyter": {
     "source_hidden": true
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(x_value, train_acc, label='train')\n",
    "plt.plot(x_value, val_acc, label='val')\n",
    "plt.plot(x_value, test_acc, label='test')\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test different node sizes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "collapsed": true,
    "jupyter": {
     "outputs_hidden": true,
     "source_hidden": true
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n"
     ]
    }
   ],
   "source": [
    "train_acc = []\n",
    "val_acc = []\n",
    "test_acc = []\n",
    "n_value = []\n",
    "d_value = []\n",
    "for n in [16, 64, 128]:\n",
    "    for d in [0.2, 0.4, 0.6]:\n",
    "    \n",
    "        print(str(n) + str(d), end='\\r')\n",
    "        tr, v, te = train(tmp=n, dropout=d, train_data=X_train, train_labels=y_train, test_data=X_test, test_labels=y_test)\n",
    "        train_acc.append(tr)\n",
    "        val_acc.append(v)\n",
    "        test_acc.append(te)\n",
    "        n_value.append(n)\n",
    "        d_value.append(d)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "jupyter": {
     "source_hidden": true
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "x = [str(i) for i in zip(d_value, n_value)]\n",
    "plt.figure(figsize=(12,6))\n",
    "plt.plot(x, train_acc, label='train')\n",
    "plt.plot(x, val_acc, label='val')\n",
    "plt.plot(x, test_acc, label='test')\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test different training sizes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": true,
    "jupyter": {
     "outputs_hidden": true,
     "source_hidden": true
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.1\n",
      "Load Datasets\n"
     ]
    }
   ],
   "source": [
    "train_acc = []\n",
    "val_acc = []\n",
    "test_acc = []\n",
    "x_value = []\n",
    "for d in [10, 100, 500, 1000, 5000, 10000]:\n",
    "    print(d, end='\\r')\n",
    "    df = load_stratified_dataset(path='../Datasets/dataset_categories/dataset_big.csv', labels='category', samples_per_label=d, random_seed=10)\n",
    "    X_train, X_test, y_train, y_test = train_test_split(df.text, df.category)\n",
    "\n",
    "    tr, v, te = train(train_data=X_train, train_labels=y_train, test_data=X_test, test_labels=y_test)\n",
    "    train_acc.append(tr)\n",
    "    val_acc.append(v)\n",
    "    test_acc.append(te)\n",
    "    x_value.append(d*7)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "jupyter": {
     "source_hidden": true
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(x_value, train_acc, label='train')\n",
    "plt.plot(x_value, val_acc, label='val')\n",
    "plt.plot(x_value, test_acc, label='test')\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "***\n",
    "## Bidirectional LSTM with embedding layer\n",
    "Calculated on Google Colab"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "executionInfo": {
     "elapsed": 670,
     "status": "ok",
     "timestamp": 1605964580712,
     "user": {
      "displayName": "Fabian März",
      "photoUrl": "",
      "userId": "12627298575665245863"
     },
     "user_tz": -60
    },
    "id": "m8wmHF_zcK_C",
    "outputId": "87f4d30d-37c8-425d-f79b-44853e1a4197"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount(\"/content/drive\", force_remount=True).\n"
     ]
    }
   ],
   "source": [
    "from google.colab import drive\n",
    "drive.mount('/content/drive')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "executionInfo": {
     "elapsed": 2434,
     "status": "ok",
     "timestamp": 1605964582484,
     "user": {
      "displayName": "Fabian März",
      "photoUrl": "",
      "userId": "12627298575665245863"
     },
     "user_tz": -60
    },
    "id": "G0RFa4pQb-OY"
   },
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# import models\n",
    "import sys\n",
    "sys.path.append('/content/drive/MyDrive/Colab Notebooks/my_notebook/tf_models')\n",
    "from train import train\n",
    "# Add functions path\n",
    "sys.path.append('/content/drive/MyDrive/Colab Notebooks/my_notebook/Functions')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "executionInfo": {
     "elapsed": 13649,
     "status": "ok",
     "timestamp": 1605964593704,
     "user": {
      "displayName": "Fabian März",
      "photoUrl": "",
      "userId": "12627298575665245863"
     },
     "user_tz": -60
    },
    "id": "sljTrfmzb-OY",
    "outputId": "1bc9b627-ad62-4a8e-add1-1d10db8358f1"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Smallest sample size in dataset is 12026 samples!\n"
     ]
    }
   ],
   "source": [
    "from datasets import load_stratified_dataset\n",
    "from sklearn.model_selection import train_test_split\n",
    "df = load_stratified_dataset(path='/content/drive/MyDrive/Colab Notebooks/my_notebook/dataset_categories_train.csv', labels='category', samples_per_label=300000, random_seed=10)\n",
    "X_train, X_test, y_train, y_test = train_test_split(df.text_lem, df.category)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1000
    },
    "executionInfo": {
     "elapsed": 490533,
     "status": "ok",
     "timestamp": 1605965070597,
     "user": {
      "displayName": "Fabian März",
      "photoUrl": "",
      "userId": "12627298575665245863"
     },
     "user_tz": -60
    },
    "id": "h8iNEAIIdBm3",
    "outputId": "e7a0fa4a-9a4f-4069-a570-84a9cc5e07bf"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Epoch 1/30\n",
      "87/87 [==============================] - 17s 196ms/step - loss: 1.8206 - categorical_accuracy: 0.3234 - val_loss: 1.9474 - val_categorical_accuracy: 0.1625\n",
      "Epoch 2/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 1.0086 - categorical_accuracy: 0.6674 - val_loss: 1.8408 - val_categorical_accuracy: 0.1952\n",
      "Epoch 3/30\n",
      "87/87 [==============================] - 15s 174ms/step - loss: 0.7808 - categorical_accuracy: 0.7550 - val_loss: 1.5982 - val_categorical_accuracy: 0.3715\n",
      "Epoch 4/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.6591 - categorical_accuracy: 0.7932 - val_loss: 1.2496 - val_categorical_accuracy: 0.5658\n",
      "Epoch 5/30\n",
      "87/87 [==============================] - 15s 172ms/step - loss: 0.5698 - categorical_accuracy: 0.8211 - val_loss: 0.9438 - val_categorical_accuracy: 0.6936\n",
      "Epoch 6/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.5021 - categorical_accuracy: 0.8442 - val_loss: 0.9568 - val_categorical_accuracy: 0.7244\n",
      "Epoch 7/30\n",
      "87/87 [==============================] - 15s 174ms/step - loss: 0.4445 - categorical_accuracy: 0.8592 - val_loss: 0.9353 - val_categorical_accuracy: 0.7503\n",
      "Epoch 8/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.4011 - categorical_accuracy: 0.8741 - val_loss: 0.9364 - val_categorical_accuracy: 0.7567\n",
      "Epoch 9/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.3554 - categorical_accuracy: 0.8881 - val_loss: 0.9611 - val_categorical_accuracy: 0.7635\n",
      "Epoch 10/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.3280 - categorical_accuracy: 0.8972 - val_loss: 1.0059 - val_categorical_accuracy: 0.7601\n",
      "Epoch 11/30\n",
      "87/87 [==============================] - 15s 175ms/step - loss: 0.2948 - categorical_accuracy: 0.9051 - val_loss: 1.0504 - val_categorical_accuracy: 0.7614\n",
      "Epoch 12/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.2697 - categorical_accuracy: 0.9136 - val_loss: 1.1036 - val_categorical_accuracy: 0.7604\n",
      "Epoch 13/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.2519 - categorical_accuracy: 0.9185 - val_loss: 1.0818 - val_categorical_accuracy: 0.7640\n",
      "Epoch 14/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.2363 - categorical_accuracy: 0.9226 - val_loss: 1.0967 - val_categorical_accuracy: 0.7632\n",
      "Epoch 15/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.2186 - categorical_accuracy: 0.9290 - val_loss: 1.1640 - val_categorical_accuracy: 0.7568\n",
      "Epoch 16/30\n",
      "87/87 [==============================] - 15s 172ms/step - loss: 0.2028 - categorical_accuracy: 0.9350 - val_loss: 1.1576 - val_categorical_accuracy: 0.7542\n",
      "Epoch 17/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.1952 - categorical_accuracy: 0.9364 - val_loss: 1.1962 - val_categorical_accuracy: 0.7566\n",
      "Epoch 18/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.1840 - categorical_accuracy: 0.9395 - val_loss: 1.2049 - val_categorical_accuracy: 0.7603\n",
      "Epoch 19/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.1655 - categorical_accuracy: 0.9452 - val_loss: 1.3132 - val_categorical_accuracy: 0.7539\n",
      "Epoch 20/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.1562 - categorical_accuracy: 0.9486 - val_loss: 1.3103 - val_categorical_accuracy: 0.7559\n",
      "Epoch 21/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.1484 - categorical_accuracy: 0.9509 - val_loss: 1.3268 - val_categorical_accuracy: 0.7589\n",
      "Epoch 22/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.1438 - categorical_accuracy: 0.9526 - val_loss: 1.3669 - val_categorical_accuracy: 0.7563\n",
      "Epoch 23/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.1466 - categorical_accuracy: 0.9504 - val_loss: 1.3426 - val_categorical_accuracy: 0.7567\n",
      "Epoch 24/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.1354 - categorical_accuracy: 0.9542 - val_loss: 1.3353 - val_categorical_accuracy: 0.7578\n",
      "Epoch 25/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.1330 - categorical_accuracy: 0.9556 - val_loss: 1.3202 - val_categorical_accuracy: 0.7604\n",
      "Epoch 26/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.1269 - categorical_accuracy: 0.9574 - val_loss: 1.3922 - val_categorical_accuracy: 0.7596\n",
      "Epoch 27/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.1210 - categorical_accuracy: 0.9589 - val_loss: 1.3786 - val_categorical_accuracy: 0.7593\n",
      "Epoch 28/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.1227 - categorical_accuracy: 0.9590 - val_loss: 1.3701 - val_categorical_accuracy: 0.7601\n",
      "Epoch 29/30\n",
      "87/87 [==============================] - 15s 174ms/step - loss: 0.1151 - categorical_accuracy: 0.9610 - val_loss: 1.3836 - val_categorical_accuracy: 0.7597\n",
      "Epoch 30/30\n",
      "87/87 [==============================] - 15s 173ms/step - loss: 0.1129 - categorical_accuracy: 0.9618 - val_loss: 1.3990 - val_categorical_accuracy: 0.7611\n",
      "42/42 [==============================] - 3s 62ms/step - loss: 1.4135 - categorical_accuracy: 0.7587\n",
      "Loss:  1.413507342338562\n",
      "Accuracy:  0.7587189674377441\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x432 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "(0.9618273973464966, 0.7611002326011658, 0.7587189674377441)"
      ]
     },
     "execution_count": 4,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train(train_data=X_train, train_labels=y_train, test_data=X_test, test_labels=y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test different node sizes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "collapsed": true,
    "executionInfo": {
     "elapsed": 1561974,
     "status": "ok",
     "timestamp": 1605966885910,
     "user": {
      "displayName": "Fabian März",
      "photoUrl": "",
      "userId": "12627298575665245863"
     },
     "user_tz": -60
    },
    "id": "HM_AFCG-b-OY",
    "jupyter": {
     "outputs_hidden": true,
     "source_hidden": true
    },
    "outputId": "b13d6428-00bf-4830-ee78-a2e690af6f58"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n"
     ]
    }
   ],
   "source": [
    "train_acc = []\n",
    "val_acc = []\n",
    "test_acc = []\n",
    "x_value = []\n",
    "for d in np.linspace(8,100,5):\n",
    "    d = int(d)\n",
    "    print(d, end='\\r')\n",
    "    tr, v, te = train(tmp=d, train_data=X_train, train_labels=y_train, test_data=X_test, test_labels=y_test)\n",
    "    train_acc.append(tr)\n",
    "    val_acc.append(v)\n",
    "    test_acc.append(te)\n",
    "    x_value.append(d)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 265
    },
    "executionInfo": {
     "elapsed": 1560883,
     "status": "ok",
     "timestamp": 1605966885931,
     "user": {
      "displayName": "Fabian März",
      "photoUrl": "",
      "userId": "12627298575665245863"
     },
     "user_tz": -60
    },
    "id": "b_o1zkBOb-Oa",
    "jupyter": {
     "source_hidden": true
    },
    "outputId": "08595ee4-c148-4ffd-fda6-8e61a4f1638f"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(x_value, train_acc, label='train')\n",
    "plt.plot(x_value, val_acc, label='val')\n",
    "plt.plot(x_value, test_acc, label='test')\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test different dropouts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "collapsed": true,
    "executionInfo": {
     "elapsed": 5953670,
     "status": "ok",
     "timestamp": 1605974756188,
     "user": {
      "displayName": "Fabian März",
      "photoUrl": "",
      "userId": "12627298575665245863"
     },
     "user_tz": -60
    },
    "id": "ufLas2rRb-Oa",
    "jupyter": {
     "outputs_hidden": true,
     "source_hidden": true
    },
    "outputId": "cc99ac4d-f41e-44d6-94de-cdf74bdd4511"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n"
     ]
    }
   ],
   "source": [
    "train_acc = []\n",
    "val_acc = []\n",
    "test_acc = []\n",
    "x_value = []\n",
    "for d in np.linspace(0.1, 0.8, 5):\n",
    "    print(d, end='\\r')\n",
    "    tr, v, te = train(dropout=d, train_data=X_train, train_labels=y_train, test_data=X_test, test_labels=y_test)\n",
    "    train_acc.append(tr)\n",
    "    val_acc.append(v)\n",
    "    test_acc.append(te)\n",
    "    x_value.append(d)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 267
    },
    "executionInfo": {
     "elapsed": 5952869,
     "status": "ok",
     "timestamp": 1605974756194,
     "user": {
      "displayName": "Fabian März",
      "photoUrl": "",
      "userId": "12627298575665245863"
     },
     "user_tz": -60
    },
    "id": "zlb4H0hh_CXR",
    "jupyter": {
     "source_hidden": true
    },
    "outputId": "19d140ce-c70b-4700-deb7-80d8b7833858"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(x_value, train_acc, label='train')\n",
    "plt.plot(x_value, val_acc, label='val')\n",
    "plt.plot(x_value, test_acc, label='test')\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test different Sequence length"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "collapsed": true,
    "executionInfo": {
     "elapsed": 1900834,
     "status": "ok",
     "timestamp": 1605976657046,
     "user": {
      "displayName": "Fabian März",
      "photoUrl": "",
      "userId": "12627298575665245863"
     },
     "user_tz": -60
    },
    "id": "6cn8bZfG_i_g",
    "jupyter": {
     "outputs_hidden": true,
     "source_hidden": true
    },
    "outputId": "819e6fa6-f82b-4fa6-dd16-5df4e0828386"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n"
     ]
    }
   ],
   "source": [
    "train_acc = []\n",
    "val_acc = []\n",
    "test_acc = []\n",
    "x_value = []\n",
    "for d in np.linspace(10,3000,5):\n",
    "    d = int(d)\n",
    "    print(d, end='\\r')\n",
    "    tr, v, te = train(sequence_length=d, train_data=X_train, train_labels=y_train, test_data=X_test, test_labels=y_test)\n",
    "    train_acc.append(tr)\n",
    "    val_acc.append(v)\n",
    "    test_acc.append(te)\n",
    "    x_value.append(d)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 265
    },
    "executionInfo": {
     "elapsed": 1900855,
     "status": "ok",
     "timestamp": 1605976657085,
     "user": {
      "displayName": "Fabian März",
      "photoUrl": "",
      "userId": "12627298575665245863"
     },
     "user_tz": -60
    },
    "id": "UYZyAqaCJUif",
    "jupyter": {
     "source_hidden": true
    },
    "outputId": "6d8b95da-b0f6-4936-be62-9a562bf929d6"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(x_value, train_acc, label='train')\n",
    "plt.plot(x_value, val_acc, label='val')\n",
    "plt.plot(x_value, test_acc, label='test')\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test different embedding dimensions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "collapsed": true,
    "executionInfo": {
     "elapsed": 3039971,
     "status": "ok",
     "timestamp": 1605971554058,
     "user": {
      "displayName": "Fabian März",
      "photoUrl": "",
      "userId": "12627298575665245863"
     },
     "user_tz": -60
    },
    "id": "1Gz8lGtWJWyv",
    "jupyter": {
     "outputs_hidden": true,
     "source_hidden": true
    },
    "outputId": "c65133e4-3952-4838-8b05-5e5df44a3e9c"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n"
     ]
    }
   ],
   "source": [
    "train_acc = []\n",
    "val_acc = []\n",
    "test_acc = []\n",
    "x_value = []\n",
    "for d in np.linspace(6,30,5):\n",
    "    d = int(d)\n",
    "    print(d, end='\\r')\n",
    "    tr, v, te = train(embedding_dim=d, train_data=X_train, train_labels=y_train, test_data=X_test, test_labels=y_test)\n",
    "    train_acc.append(tr)\n",
    "    val_acc.append(v)\n",
    "    test_acc.append(te)\n",
    "    x_value.append(d)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 265
    },
    "executionInfo": {
     "elapsed": 3039973,
     "status": "ok",
     "timestamp": 1605971554064,
     "user": {
      "displayName": "Fabian März",
      "photoUrl": "",
      "userId": "12627298575665245863"
     },
     "user_tz": -60
    },
    "id": "qKDWAnbuJaC7",
    "jupyter": {
     "source_hidden": true
    },
    "outputId": "7c528a26-80e8-41cb-b503-58c8a7bc3e25"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(x_value, train_acc, label='train')\n",
    "plt.plot(x_value, val_acc, label='val')\n",
    "plt.plot(x_value, test_acc, label='test')\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test different vocabulary sizes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "collapsed": true,
    "executionInfo": {
     "elapsed": 4638174,
     "status": "ok",
     "timestamp": 1605973152269,
     "user": {
      "displayName": "Fabian März",
      "photoUrl": "",
      "userId": "12627298575665245863"
     },
     "user_tz": -60
    },
    "id": "z3ZBZYt9JcZp",
    "jupyter": {
     "outputs_hidden": true,
     "source_hidden": true
    },
    "outputId": "f5ea6462-8d7a-4e07-f745-029737c2508a"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n",
      "Tensorflow version 2.3.0\n",
      "Load Datasets\n"
     ]
    }
   ],
   "source": [
    "train_acc = []\n",
    "val_acc = []\n",
    "test_acc = []\n",
    "x_value = []\n",
    "for d in np.linspace(5000,20000,5):\n",
    "    d = int(d)\n",
    "    print(d, end='\\r')\n",
    "    tr, v, te = train(vocab_size=d, train_data=X_train, train_labels=y_train, test_data=X_test, test_labels=y_test)\n",
    "    train_acc.append(tr)\n",
    "    val_acc.append(v)\n",
    "    test_acc.append(te)\n",
    "    x_value.append(d)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 265
    },
    "executionInfo": {
     "elapsed": 4638172,
     "status": "ok",
     "timestamp": 1605973152271,
     "user": {
      "displayName": "Fabian März",
      "photoUrl": "",
      "userId": "12627298575665245863"
     },
     "user_tz": -60
    },
    "id": "4KaA6NadJe_1",
    "jupyter": {
     "source_hidden": true
    },
    "outputId": "25e1b826-9288-4b77-d7c0-2f4eab03fa3c"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(x_value, train_acc, label='train')\n",
    "plt.plot(x_value, val_acc, label='val')\n",
    "plt.plot(x_value, test_acc, label='test')\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}