Keras_Models_predictions.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "DmcNFGyW9dAB"
   },
   "source": [
    "# <h1 align = \"center\"> Keras Models</h1>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "KAbdXGyT-vAN"
   },
   "source": [
    "The goal for this analysis is to predict if a review rates the movie positively or negatively. Inside this dataset, there are 25,000 labelled movies reviews for training, 50,000 unlabeled reviews for training, and 25,000 reviews for testing.\n",
    "<a href=\"https://imgur.com/FfdEBRz\"><img src=\"https://i.imgur.com/FfdEBRzm.png\" title=\"source: imgur.com\" align=\"right\"></a>\n",
    "- IMDB movie reviews dataset\n",
    "- http://ai.stanford.edu/~amaas/data/sentiment\n",
    "- https://www.kaggle.com/lakshmi25npathi/imdb-dataset-of-50k-movie-reviews\n",
    "- Contains 25000 positive and 25000 negative reviews\n",
    "- Contains at most reviews per movie\n",
    "- At least 7 stars out of 10 $\\rightarrow$ positive (label = 1)\n",
    "- At most 4 stars out of 10 $\\rightarrow$ negative (label = 0)\n",
    "\n",
    "\n",
    "This notebook uses [tf.keras](https://www.tensorflow.org/guide/keras), a high-level API to build and train models in TensorFlow\n",
    "> **Here, we use Keras Sequential API to train and test simple neural networks to classify sentiments of reviews by inputting tfidf features**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "Zw2OdXqR97oQ"
   },
   "source": [
    "## <h2 align = \"center\">Dependencies</h2>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 123
    },
    "colab_type": "code",
    "id": "CFDhgYmH9bmc",
    "outputId": "878c5789-a5e6-4510-ba33-6961ab1d9bd1"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\framework\\dtypes.py:526: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\n",
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\framework\\dtypes.py:527: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\n",
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\framework\\dtypes.py:528: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\n",
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\framework\\dtypes.py:529: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\n",
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\framework\\dtypes.py:530: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\n",
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\framework\\dtypes.py:535: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Version:  1.13.1\n",
      "Eager mode:  False\n",
      "GPU is NOT AVAILABLE\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "import seaborn as sns\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from sklearn.metrics import accuracy_score, confusion_matrix\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "from sklearn.feature_selection import SelectKBest, f_classif\n",
    "\n",
    "import tensorflow as tf\n",
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.python.keras.callbacks import LambdaCallback, EarlyStopping\n",
    "from tensorflow.keras.layers import *\n",
    "from tensorflow.keras.preprocessing import text\n",
    "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
    "\n",
    "import warnings\n",
    "warnings.filterwarnings('ignore')\n",
    "\n",
    "print(\"Version: \", tf.__version__)\n",
    "print(\"Eager mode: \", tf.executing_eagerly())\n",
    "print(\"GPU is\", \"available\" if tf.test.is_gpu_available() else \"NOT AVAILABLE\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "7rxVlB-SMbNd",
    "outputId": "63fd2786-1a78-4dda-b1b9-c785be5f8b0e"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "\"\\ndevice_name = tf.test.gpu_device_name()\\nif device_name != '/device:GPU:0':\\n  raise SystemError('GPU device not found')\\nprint('Found GPU at: {}'.format(device_name))\\n\""
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "'''\n",
    "device_name = tf.test.gpu_device_name()\n",
    "if device_name != '/device:GPU:0':\n",
    "  raise SystemError('GPU device not found')\n",
    "print('Found GPU at: {}'.format(device_name))\n",
    "'''"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "bfcIlFC4F41g",
    "outputId": "59d36e18-9e6d-470e-b80f-048a92ce2257"
   },
   "outputs": [],
   "source": [
    "# for importing data from colab\n",
    "# from google.colab import drive \n",
    "# drive.mount('/content/drive')\n",
    "\n",
    "# movies = pd.read_csv('/content/drive/My Drive/Colab Notebooks/imdb_data.csv')\n",
    "# movies.sample(7)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 263
    },
    "colab_type": "code",
    "id": "dYAkdt8TF4nF",
    "outputId": "6d4ec9d4-b5e6-4635-caf2-2ad8b2e864c3"
   },
   "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>review</th>\n",
       "      <th>sentiment</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>36327</th>\n",
       "      <td>I watched this film not knowing anything about...</td>\n",
       "      <td>positive</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>40150</th>\n",
       "      <td>Unfortunately for myself - I stumbled onto thi...</td>\n",
       "      <td>positive</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13388</th>\n",
       "      <td>I really am shocked to see the number of revie...</td>\n",
       "      <td>positive</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>28839</th>\n",
       "      <td>Pretty poor Firestarter clone that seems more ...</td>\n",
       "      <td>negative</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>40919</th>\n",
       "      <td>i searched video store everywhere to find this...</td>\n",
       "      <td>negative</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>32992</th>\n",
       "      <td>This show sucks. it was put on fridays on roll...</td>\n",
       "      <td>negative</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9195</th>\n",
       "      <td>The world is made up two different types of mo...</td>\n",
       "      <td>positive</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                  review sentiment\n",
       "36327  I watched this film not knowing anything about...  positive\n",
       "40150  Unfortunately for myself - I stumbled onto thi...  positive\n",
       "13388  I really am shocked to see the number of revie...  positive\n",
       "28839  Pretty poor Firestarter clone that seems more ...  negative\n",
       "40919  i searched video store everywhere to find this...  negative\n",
       "32992  This show sucks. it was put on fridays on roll...  negative\n",
       "9195   The world is made up two different types of mo...  positive"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# read from local\n",
    "movies = pd.read_csv('data/imdb_data.csv')\n",
    "movies.sample(7)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "IstLZZ5OG3U8"
   },
   "source": [
    "> **Binarize the sentiment column**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 263
    },
    "colab_type": "code",
    "id": "fsRxNWaDG2sJ",
    "outputId": "8829dc08-c350-441c-a052-7838acd38e18"
   },
   "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>review</th>\n",
       "      <th>sentiment</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>38016</th>\n",
       "      <td>The Cure uses voice over to create an intense ...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>21165</th>\n",
       "      <td>I know a lot of people like this show and i ap...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>26137</th>\n",
       "      <td>Me being of Irish origins, loved this movie, N...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>21556</th>\n",
       "      <td>You know I only watched 15 minutes of this fil...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>29370</th>\n",
       "      <td>Direction must be the problem here. I recently...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>43634</th>\n",
       "      <td>This is one of the best TV movies I have ever ...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8939</th>\n",
       "      <td>'Wicked Little Things' really separates itself...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                  review  sentiment\n",
       "38016  The Cure uses voice over to create an intense ...          1\n",
       "21165  I know a lot of people like this show and i ap...          0\n",
       "26137  Me being of Irish origins, loved this movie, N...          1\n",
       "21556  You know I only watched 15 minutes of this fil...          0\n",
       "29370  Direction must be the problem here. I recently...          0\n",
       "43634  This is one of the best TV movies I have ever ...          1\n",
       "8939   'Wicked Little Things' really separates itself...          1"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "le = LabelEncoder()\n",
    "movies['sentiment'] = le.fit_transform(movies['sentiment'])\n",
    "movies.sample(7)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "BSby0Ik2IGxT"
   },
   "source": [
    " > **Vectorize the reviews text**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "_zYXdJ8N90wx",
    "outputId": "1401d0f9-cf0a-4dd8-c490-f6dd3ef9874e"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(50000, 20000)"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "NGRAM_RANGE = (1, 2)\n",
    "TOP_K = 20000\n",
    "TOKEN_MODE = 'word'\n",
    "MIN_DOC_FREQ = 2\n",
    "\n",
    "def ngram_vectorize(texts, labels):\n",
    "    kwargs = {\n",
    "        'ngram_range' : NGRAM_RANGE,\n",
    "        'dtype' : 'int32',\n",
    "        'strip_accents' : 'unicode',\n",
    "        'decode_error' : 'replace',\n",
    "        'analyzer' : TOKEN_MODE,\n",
    "        'min_df' : MIN_DOC_FREQ\n",
    "    }\n",
    "\n",
    "    transformed_texts = TfidfVectorizer(**kwargs).fit_transform(texts)\n",
    "    \n",
    "    # Select best k features, with feature importance measured by f_classif\n",
    "    # Set k as 20000 or (if number of ngrams is less) number of ngrams   \n",
    "    selector = SelectKBest(f_classif, k=min(TOP_K, transformed_texts.shape[1]))\n",
    "    selector.fit(transformed_texts, labels)\n",
    "    transformed_texts = selector.transform(transformed_texts).astype('float32')\n",
    "    return transformed_texts\n",
    "\n",
    "vect_reviews = ngram_vectorize(movies['review'], movies['sentiment'])\n",
    "vect_reviews.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "zpRD5_kdIMW3"
   },
   "source": [
    "> **Split the data**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 52
    },
    "colab_type": "code",
    "id": "awjc0YmM90-W",
    "outputId": "9e452aeb-4e1f-4a46-87e3-e44cf9bc1d6b"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train dataset shape: (40000, 20000), \n",
      "Test dataset shape: (10000, 20000)\n"
     ]
    }
   ],
   "source": [
    "X = vect_reviews.toarray()\n",
    "y = (np.array(movies['sentiment']))\n",
    "\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=7)\n",
    "print(f\"Train dataset shape: {X_train.shape}, \\nTest dataset shape: {X_test.shape}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "ER6jEWJ0I4Te"
   },
   "source": [
    "## <h2 align = \"center\">Modeling the data</h2>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "simple_log = LambdaCallback(\n",
    "    on_epoch_end = lambda e, l: print(f\" ~| Epoch: {e+1} | Validation Loss: {l['val_loss']:.5f}\", end =\" >|> \\n\" ))\n",
    "\n",
    "early_stop = EarlyStopping(monitor='val_loss',\n",
    "                              min_delta=0,\n",
    "                              patience=1,\n",
    "                              verbose=0, \n",
    "                              restore_best_weights=True)\n",
    "sns.set()\n",
    "def plot_learning_curve(hist):\n",
    "    plt.figure(figsize=(5,5))\n",
    "    train = hist.history['acc']\n",
    "    val = hist.history['val_acc']\n",
    "    epochs_run = range(1,len(train) + 1)\n",
    "    sns.lineplot(epochs_run, train, marker = 'o', color = 'coral', label = 'Training Accuracy')\n",
    "    sns.lineplot(epochs_run, val,  marker = '>', color = 'green', label = 'Validation Accuracy')\n",
    "    plt.title(\"Accuracies vs. Epochs\", fontsize = 20)\n",
    "    plt.legend()\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 229
    },
    "colab_type": "code",
    "id": "7bLEaoSw91Xl",
    "outputId": "24b0393a-bfd5-43de-8d51-008b4081352e"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_118 (Dense)            (None, 64)                1280064   \n",
      "_________________________________________________________________\n",
      "dense_119 (Dense)            (None, 1)                 65        \n",
      "=================================================================\n",
      "Total params: 1,280,129\n",
      "Trainable params: 1,280,129\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "input_shape = X_train.shape[1:]\n",
    "\n",
    "model = Sequential([    \n",
    "    Dense(64, activation = 'relu', input_shape = input_shape),\n",
    "    Dense(1, activation = 'sigmoid')\n",
    "])\n",
    "\n",
    "model.compile(\n",
    "    loss = 'binary_crossentropy',\n",
    "    optimizer = 'adam',\n",
    "    metrics = ['accuracy']\n",
    ")\n",
    "\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1000
    },
    "colab_type": "code",
    "id": "9L2C8tbQ91V5",
    "outputId": "a90a2aac-fab3-45a1-e0fa-9756841fefa4"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " ~| Epoch: 1 | Validation Loss: 0.18305 >|> \n",
      " ~| Epoch: 2 | Validation Loss: 0.17350 >|> \n",
      " ~| Epoch: 3 | Validation Loss: 0.19166 >|> \n",
      " ~| Epoch: 4 | Validation Loss: 0.20891 >|> \n",
      "\n",
      "Done.\n"
     ]
    }
   ],
   "source": [
    "epochs = 20\n",
    "\n",
    "h = model.fit(\n",
    "    X_train, y_train,\n",
    "    validation_split = 0.2,\n",
    "    epochs = epochs,\n",
    "    callbacks = [early_stop, simple_log],\n",
    "    verbose = False)\n",
    "print(\"\\nDone.\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 205
    },
    "colab_type": "code",
    "id": "CRZ12LUt91Rp",
    "outputId": "bd2f33b3-3679-4f08-c6eb-9092cbf4aa9a"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x360 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_learning_curve(h)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "dhAFmV-lSRnI"
   },
   "source": [
    "> **Classic case of overfitting**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 52
    },
    "colab_type": "code",
    "id": "RyBP5TaR91Pt",
    "outputId": "62b39ff9-4908-4858-a39c-c7843bffcb87"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10000/10000 [==============================] - 4s 362us/sample - loss: 0.1924 - acc: 0.9250\n",
      "Testing Accuracy:  92.5000011920929\n"
     ]
    }
   ],
   "source": [
    "loss, acc = model.evaluate(X_test, y_test)\n",
    "print(\"Testing Accuracy: \", acc*100)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "MrB6H3alTJDf"
   },
   "source": [
    "> **Trying dropout regularization**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 264
    },
    "colab_type": "code",
    "id": "6odmBWBI91NX",
    "outputId": "83bfff58-9ae1-48ed-cfae-84ccbe0309ba"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From C:\\ProgramData\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\keras\\layers\\core.py:143: calling dropout (from tensorflow.python.ops.nn_ops) with keep_prob is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use `rate` instead of `keep_prob`. Rate should be set to `rate = 1 - keep_prob`.\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_2 (Dense)              (None, 20)                400020    \n",
      "_________________________________________________________________\n",
      "dropout (Dropout)            (None, 20)                0         \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 1)                 21        \n",
      "=================================================================\n",
      "Total params: 400,041\n",
      "Trainable params: 400,041\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "input_shape = X_train.shape[1:]\n",
    "from tensorflow.keras.regularizers import l2\n",
    "\n",
    "model2 = Sequential([    \n",
    "    Dense(20, activation = 'relu', input_shape = input_shape, kernel_regularizer=l2()),\n",
    "    Dropout(0.2),\n",
    "    Dense(1, activation = 'sigmoid')\n",
    "])\n",
    "\n",
    "model2.compile(\n",
    "    loss = 'binary_crossentropy',\n",
    "    optimizer = 'adam',\n",
    "    metrics = ['accuracy']\n",
    ")\n",
    "\n",
    "model2.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 229
    },
    "colab_type": "code",
    "id": "Ek9m9Zdm91Jh",
    "outputId": "08b4c0e6-deea-4bff-f615-a3f8e59f1967"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " ~| Epoch: 1 | Validation Loss: 0.68045 >|> \n",
      " ~| Epoch: 2 | Validation Loss: 0.64146 >|> \n",
      " ~| Epoch: 3 | Validation Loss: 0.60731 >|> \n",
      " ~| Epoch: 4 | Validation Loss: 0.58412 >|> \n",
      " ~| Epoch: 5 | Validation Loss: 0.57064 >|> \n",
      " ~| Epoch: 6 | Validation Loss: 0.54559 >|> \n",
      " ~| Epoch: 7 | Validation Loss: 0.52379 >|> \n",
      " ~| Epoch: 8 | Validation Loss: 0.51261 >|> \n",
      " ~| Epoch: 9 | Validation Loss: 0.49817 >|> \n",
      " ~| Epoch: 10 | Validation Loss: 0.48691 >|> \n",
      "Done.\n"
     ]
    }
   ],
   "source": [
    "epochs = 10\n",
    "\n",
    "h2 = model2.fit(\n",
    "    X_train, y_train,\n",
    "    validation_split = 0.2,\n",
    "    epochs = epochs,\n",
    "    callbacks = [early_stop, simple_log],\n",
    "    verbose = False)\n",
    "print(\"Done.\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 390
    },
    "colab_type": "code",
    "id": "GSY_LcHB91Hj",
    "outputId": "05eedbaf-0d02-4a81-c7e5-eb6d0f4e9abe"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x360 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_learning_curve(h2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 52
    },
    "colab_type": "code",
    "id": "4jm33krA91GH",
    "outputId": "cbacb1c2-32e3-417c-c034-451934a872b6"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10000/10000 [==============================] - 6s 600us/sample - loss: 0.4948 - acc: 0.8817\n",
      "Testing Accuracy:  88.16999793052673\n"
     ]
    }
   ],
   "source": [
    "loss, acc = model2.evaluate(X_test, y_test)\n",
    "print(\"Testing Accuracy: \", acc*100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 229
    },
    "colab_type": "code",
    "id": "a1nOazZE91Eq",
    "outputId": "fca5508b-3eb0-4c06-c44a-a6f2e05ff5d1"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_4 (Dense)              (None, 64)                1280064   \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_5 (Dense)              (None, 16)                1040      \n",
      "_________________________________________________________________\n",
      "dropout_2 (Dropout)          (None, 16)                0         \n",
      "_________________________________________________________________\n",
      "dense_6 (Dense)              (None, 1)                 17        \n",
      "=================================================================\n",
      "Total params: 1,281,121\n",
      "Trainable params: 1,281,121\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "input_shape = X_train.shape[1:]\n",
    "from tensorflow.keras.regularizers import l2\n",
    "\n",
    "model3 = Sequential([   \n",
    "    Dense(64, activation = 'relu', input_shape = input_shape),\n",
    "    Dropout(0.2),\n",
    "    Dense(16, activation = 'relu'),\n",
    "    Dropout(0.2),\n",
    "    Dense(1, activation = 'sigmoid')\n",
    "])\n",
    "\n",
    "model3.compile(\n",
    "    loss = 'binary_crossentropy',\n",
    "    optimizer = 'adam',\n",
    "    metrics = ['accuracy']\n",
    ")\n",
    "\n",
    "model3.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "0Qb6uj-G9087",
    "outputId": "47bab320-edb0-4c14-cc0a-707ac9858ad2",
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " ~| Epoch: 1 | Validation Loss: 0.18744 >|> \n",
      " ~| Epoch: 2 | Validation Loss: 0.23030 >|> \n",
      " ~| Epoch: 3 | Validation Loss: 0.28712 >|> \n",
      "\n",
      "Done.\n"
     ]
    }
   ],
   "source": [
    "epochs = 20\n",
    "\n",
    "h3 = model3.fit(\n",
    "    X_train, y_train,\n",
    "    validation_split = 0.2,\n",
    "    epochs = epochs,\n",
    "    callbacks = [early_stop, simple_log],\n",
    "    verbose = False\n",
    ")\n",
    "print(\"\\nDone.\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "vweDHdyT906e",
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x360 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_learning_curve(h3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "Yzr7oiLo903n"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10000/10000 [==============================] - 3s 344us/sample - loss: 0.2063 - acc: 0.9227\n",
      "Testing Accuracy:  92.26999878883362\n"
     ]
    }
   ],
   "source": [
    "loss, acc = model3.evaluate(X_test, y_test)\n",
    "print(\"Testing Accuracy: \", acc*100)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### <h3 align = \"center\">Optimize activations and learning rate</h3>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "xd1tJdu19016"
   },
   "outputs": [],
   "source": [
    "def create_model(learning_rate=0.01, activation='relu'):\n",
    "    opt = tf.keras.optimizers.Adam(lr=learning_rate)\n",
    "    model = Sequential()\n",
    "    model.add(Dense(64, input_shape=input_shape, activation=activation)),\n",
    "    Dropout(0.2),\n",
    "    model.add(Dense(16, activation=activation))\n",
    "    model.add(Dense(1, activation='sigmoid'))\n",
    "    \n",
    "    model.compile(optimizer=opt, loss='binary_crossentropy', metrics=['accuracy'])\n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "yvkpAZii90tP"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "# Random Search is better than exhaustive search for hyperparameter tuning\n",
    "# Import KerasClassifier from keras wrappers\n",
    "from keras.wrappers.scikit_learn import KerasClassifier\n",
    "from sklearn.model_selection import RandomizedSearchCV, KFold, GridSearchCV\n",
    "\n",
    "\n",
    "# Create a KerasClassifier\n",
    "model = KerasClassifier(build_fn = create_model)\n",
    "\n",
    "# Define the parameters to try out\n",
    "params = {'activation': ['relu'],'epochs': [3, 5, 10], \n",
    "          'learning_rate': [0.1, 0.01, 0.001], 'batch_size': [64, 128, 256]}\n",
    "\n",
    "# Create a randomize search cv object passing in the parameters to try\n",
    "random_search = RandomizedSearchCV(model, param_distributions = params, cv = KFold(2))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "jYnnju8H90pt"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "20000/20000 [==============================] - 10s 510us/sample - loss: 0.2570 - acc: 0.8898 - loss: 0\n",
      "Epoch 2/10\n",
      "20000/20000 [==============================] - 9s 458us/sample - loss: 0.0723 - acc: 0.9743\n",
      "Epoch 3/10\n",
      "20000/20000 [==============================] - 9s 452us/sample - loss: 0.0295 - acc: 0.9893\n",
      "Epoch 4/10\n",
      "20000/20000 [==============================] - 9s 465us/sample - loss: 0.0162 - acc: 0.9942\n",
      "Epoch 5/10\n",
      "20000/20000 [==============================] - 12s 590us/sample - loss: 0.0097 - acc: 0.9968\n",
      "Epoch 6/10\n",
      "20000/20000 [==============================] - 10s 501us/sample - loss: 0.0105 - acc: 0.9961\n",
      "Epoch 7/10\n",
      "20000/20000 [==============================] - 10s 483us/sample - loss: 0.0111 - acc: 0.9959\n",
      "Epoch 8/10\n",
      "20000/20000 [==============================] - 10s 479us/sample - loss: 0.0062 - acc: 0.9980\n",
      "Epoch 9/10\n",
      "20000/20000 [==============================] - 9s 457us/sample - loss: 0.0035 - acc: 0.9987\n",
      "Epoch 10/10\n",
      "20000/20000 [==============================] - 9s 454us/sample - loss: 0.0038 - acc: 0.9989\n",
      "20000/20000 [==============================] - 4s 220us/sample - loss: 0.7403 - acc: 0.9007\n",
      "Epoch 1/10\n",
      "20000/20000 [==============================] - 13s 648us/sample - loss: 0.2608 - acc: 0.8937\n",
      "Epoch 2/10\n",
      "20000/20000 [==============================] - 9s 451us/sample - loss: 0.0780 - acc: 0.9728\n",
      "Epoch 3/10\n",
      "20000/20000 [==============================] - 9s 450us/sample - loss: 0.0342 - acc: 0.9868s -\n",
      "Epoch 4/10\n",
      "20000/20000 [==============================] - 9s 475us/sample - loss: 0.0281 - acc: 0.9893\n",
      "Epoch 5/10\n",
      "20000/20000 [==============================] - 10s 478us/sample - loss: 0.0153 - acc: 0.9949\n",
      "Epoch 6/10\n",
      "20000/20000 [==============================] - 9s 450us/sample - loss: 0.0102 - acc: 0.9967\n",
      "Epoch 7/10\n",
      "20000/20000 [==============================] - 9s 454us/sample - loss: 0.0065 - acc: 0.9977\n",
      "Epoch 8/10\n",
      "20000/20000 [==============================] - 9s 460us/sample - loss: 0.0044 - acc: 0.9984\n",
      "Epoch 9/10\n",
      "20000/20000 [==============================] - 9s 456us/sample - loss: 0.0071 - acc: 0.9975s - \n",
      "Epoch 10/10\n",
      "20000/20000 [==============================] - 9s 462us/sample - loss: 0.0109 - acc: 0.9961\n",
      "20000/20000 [==============================] - 5s 241us/sample - loss: 0.5632 - acc: 0.8997\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 8s 385us/sample - loss: 0.3782 - acc: 0.8224\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 5s 263us/sample - loss: 0.1151 - acc: 0.9617\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 5s 264us/sample - loss: 0.0460 - acc: 0.9872\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 5s 261us/sample - loss: 0.0297 - acc: 0.9908\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 5s 264us/sample - loss: 0.0220 - acc: 0.9926\n",
      "20000/20000 [==============================] - 3s 165us/sample - loss: 0.4734 - acc: 0.9010\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 9s 427us/sample - loss: 0.2775 - acc: 0.8797\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 5s 267us/sample - loss: 0.0734 - acc: 0.9733\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 5s 261us/sample - loss: 0.0328 - acc: 0.9880\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 5s 263us/sample - loss: 0.0278 - acc: 0.9902\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 5s 263us/sample - loss: 0.0203 - acc: 0.9928\n",
      "20000/20000 [==============================] - 3s 170us/sample - loss: 0.5503 - acc: 0.8981\n",
      "Epoch 1/3\n",
      "20000/20000 [==============================] - 11s 556us/sample - loss: 0.2549 - acc: 0.8952\n",
      "Epoch 2/3\n",
      "20000/20000 [==============================] - 9s 444us/sample - loss: 0.0796 - acc: 0.9683\n",
      "Epoch 3/3\n",
      "20000/20000 [==============================] - 9s 443us/sample - loss: 0.0472 - acc: 0.9820\n",
      "20000/20000 [==============================] - 4s 217us/sample - loss: 0.5058 - acc: 0.8995\n",
      "Epoch 1/3\n",
      "20000/20000 [==============================] - 11s 565us/sample - loss: 0.2669 - acc: 0.8888\n",
      "Epoch 2/3\n",
      "20000/20000 [==============================] - 9s 462us/sample - loss: 0.0881 - acc: 0.9649\n",
      "Epoch 3/3\n",
      "20000/20000 [==============================] - 9s 448us/sample - loss: 0.0563 - acc: 0.9795\n",
      "20000/20000 [==============================] - 4s 212us/sample - loss: 0.3878 - acc: 0.9046\n",
      "Epoch 1/3\n",
      "20000/20000 [==============================] - 10s 503us/sample - loss: 0.3426 - acc: 0.8693\n",
      "Epoch 2/3\n",
      "20000/20000 [==============================] - 9s 455us/sample - loss: 0.1179 - acc: 0.9592\n",
      "Epoch 3/3\n",
      "20000/20000 [==============================] - 10s 484us/sample - loss: 0.0529 - acc: 0.9862\n",
      "20000/20000 [==============================] - 4s 220us/sample - loss: 0.2361 - acc: 0.9171\n",
      "Epoch 1/3\n",
      "20000/20000 [==============================] - 10s 507us/sample - loss: 0.3342 - acc: 0.8871\n",
      "Epoch 2/3\n",
      "20000/20000 [==============================] - 9s 447us/sample - loss: 0.1212 - acc: 0.9583\n",
      "Epoch 3/3\n",
      "20000/20000 [==============================] - 9s 448us/sample - loss: 0.0569 - acc: 0.9850\n",
      "20000/20000 [==============================] - 4s 224us/sample - loss: 0.2253 - acc: 0.9172\n",
      "Epoch 1/10\n",
      "20000/20000 [==============================] - 10s 525us/sample - loss: 0.3408 - acc: 0.8726\n",
      "Epoch 2/10\n",
      "20000/20000 [==============================] - 9s 447us/sample - loss: 0.1173 - acc: 0.9582\n",
      "Epoch 3/10\n",
      "20000/20000 [==============================] - 9s 445us/sample - loss: 0.0519 - acc: 0.9864\n",
      "Epoch 4/10\n",
      "20000/20000 [==============================] - 9s 451us/sample - loss: 0.0204 - acc: 0.9972\n",
      "Epoch 5/10\n",
      "20000/20000 [==============================] - 9s 474us/sample - loss: 0.0077 - acc: 0.9995\n",
      "Epoch 6/10\n",
      "20000/20000 [==============================] - 10s 490us/sample - loss: 0.0032 - acc: 0.9998\n",
      "Epoch 7/10\n",
      "20000/20000 [==============================] - 9s 461us/sample - loss: 0.0016 - acc: 1.0000\n",
      "Epoch 8/10\n",
      "20000/20000 [==============================] - 9s 460us/sample - loss: 9.6778e-04 - acc: 1.0000\n",
      "Epoch 9/10\n",
      "20000/20000 [==============================] - 9s 459us/sample - loss: 6.5906e-04 - acc: 1.0000\n",
      "Epoch 10/10\n",
      "20000/20000 [==============================] - 9s 459us/sample - loss: 4.6826e-04 - acc: 1.0000s - loss: 5.00 - ETA: 3s - loss: 4 - E\n",
      "20000/20000 [==============================] - 4s 225us/sample - loss: 0.3992 - acc: 0.9119\n",
      "Epoch 1/10\n",
      "20000/20000 [==============================] - 10s 494us/sample - loss: 0.3436 - acc: 0.8703\n",
      "Epoch 2/10\n",
      "20000/20000 [==============================] - 9s 455us/sample - loss: 0.1241 - acc: 0.9563s - los\n",
      "Epoch 3/10\n",
      "20000/20000 [==============================] - 9s 460us/sample - loss: 0.0561 - acc: 0.9847\n",
      "Epoch 4/10\n",
      "20000/20000 [==============================] - 9s 460us/sample - loss: 0.0226 - acc: 0.9969\n",
      "Epoch 5/10\n",
      "20000/20000 [==============================] - 9s 461us/sample - loss: 0.0088 - acc: 0.9993s\n",
      "Epoch 6/10\n",
      "20000/20000 [==============================] - 10s 492us/sample - loss: 0.0042 - acc: 0.9997\n",
      "Epoch 7/10\n",
      "20000/20000 [==============================] - 10s 484us/sample - loss: 0.0020 - acc: 0.9998\n",
      "Epoch 8/10\n",
      "20000/20000 [==============================] - 9s 469us/sample - loss: 0.0011 - acc: 1.0000\n",
      "Epoch 9/10\n",
      "20000/20000 [==============================] - 9s 464us/sample - loss: 6.8518e-04 - acc: 1.0000\n",
      "Epoch 10/10\n",
      "20000/20000 [==============================] - 9s 469us/sample - loss: 2.9483e-04 - acc: 1.0000\n",
      "20000/20000 [==============================] - 5s 232us/sample - loss: 0.4245 - acc: 0.9122\n",
      "Epoch 1/3\n",
      "20000/20000 [==============================] - 7s 329us/sample - loss: 0.5613 - acc: 0.8494\n",
      "Epoch 2/3\n",
      "20000/20000 [==============================] - 5s 268us/sample - loss: 0.2480 - acc: 0.9223\n",
      "Epoch 3/3\n",
      "20000/20000 [==============================] - 5s 266us/sample - loss: 0.1462 - acc: 0.9513\n",
      "20000/20000 [==============================] - 3s 167us/sample - loss: 0.1937 - acc: 0.9243\n",
      "Epoch 1/3\n",
      "20000/20000 [==============================] - 7s 327us/sample - loss: 0.5252 - acc: 0.8303\n",
      "Epoch 2/3\n",
      "20000/20000 [==============================] - 5s 267us/sample - loss: 0.2197 - acc: 0.9291\n",
      "Epoch 3/3\n",
      "20000/20000 [==============================] - 5s 268us/sample - loss: 0.1329 - acc: 0.9573\n",
      "20000/20000 [==============================] - 5s 227us/sample - loss: 0.1892 - acc: 0.9232\n",
      "Epoch 1/5\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "20000/20000 [==============================] - 10s 500us/sample - loss: 0.2976 - acc: 0.8616\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 9s 447us/sample - loss: 0.0964 - acc: 0.9657\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 9s 449us/sample - loss: 0.0620 - acc: 0.9799\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 9s 451us/sample - loss: 0.0378 - acc: 0.9879\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 9s 463us/sample - loss: 0.0232 - acc: 0.9925s - loss:\n",
      "20000/20000 [==============================] - 5s 239us/sample - loss: 0.5668 - acc: 0.9018s - loss: 0.5708 - acc: 0.\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 11s 538us/sample - loss: 0.2822 - acc: 0.8873\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 9s 458us/sample - loss: 0.0985 - acc: 0.9667\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 9s 455us/sample - loss: 0.0505 - acc: 0.9857\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 9s 467us/sample - loss: 0.0483 - acc: 0.9870\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 10s 480us/sample - loss: 0.0415 - acc: 0.9886\n",
      "20000/20000 [==============================] - 5s 242us/sample - loss: 0.4363 - acc: 0.9013\n",
      "Epoch 1/3\n",
      "20000/20000 [==============================] - 8s 402us/sample - loss: 0.2668 - acc: 0.8854\n",
      "Epoch 2/3\n",
      "20000/20000 [==============================] - 7s 328us/sample - loss: 0.0726 - acc: 0.9759\n",
      "Epoch 3/3\n",
      "20000/20000 [==============================] - 6s 324us/sample - loss: 0.0235 - acc: 0.9926\n",
      "20000/20000 [==============================] - 4s 194us/sample - loss: 0.4124 - acc: 0.9023\n",
      "Epoch 1/3\n",
      "20000/20000 [==============================] - 8s 384us/sample - loss: 0.2752 - acc: 0.8856\n",
      "Epoch 2/3\n",
      "20000/20000 [==============================] - 7s 339us/sample - loss: 0.0803 - acc: 0.9734\n",
      "Epoch 3/3\n",
      "20000/20000 [==============================] - 7s 331us/sample - loss: 0.0282 - acc: 0.9905\n",
      "20000/20000 [==============================] - 4s 201us/sample - loss: 0.3692 - acc: 0.9064\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 9s 464us/sample - loss: 0.4361 - acc: 0.8518\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 6s 324us/sample - loss: 0.1540 - acc: 0.9454\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 7s 325us/sample - loss: 0.0833 - acc: 0.9746\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 7s 329us/sample - loss: 0.0443 - acc: 0.9909\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 7s 327us/sample - loss: 0.0217 - acc: 0.9979\n",
      "20000/20000 [==============================] - 4s 189us/sample - loss: 0.2542 - acc: 0.9158\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 8s 417us/sample - loss: 0.4030 - acc: 0.8669\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 7s 333us/sample - loss: 0.1479 - acc: 0.9481\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 6s 320us/sample - loss: 0.0800 - acc: 0.9770\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 7s 331us/sample - loss: 0.0411 - acc: 0.9922\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 7s 329us/sample - loss: 0.0191 - acc: 0.9981\n",
      "20000/20000 [==============================] - 4s 192us/sample - loss: 0.2515 - acc: 0.9179\n",
      "Epoch 1/3\n",
      "20000/20000 [==============================] - ETA: 0s - loss: 0.2641 - acc: 0.883 - 7s 340us/sample - loss: 0.2640 - acc: 0.8838\n",
      "Epoch 2/3\n",
      "20000/20000 [==============================] - 6s 278us/sample - loss: 0.0620 - acc: 0.9770\n",
      "Epoch 3/3\n",
      "20000/20000 [==============================] - 6s 281us/sample - loss: 0.0239 - acc: 0.9926\n",
      "20000/20000 [==============================] - 4s 184us/sample - loss: 0.3565 - acc: 0.9036\n",
      "Epoch 1/3\n",
      "20000/20000 [==============================] - 7s 339us/sample - loss: 0.3286 - acc: 0.8712\n",
      "Epoch 2/3\n",
      "20000/20000 [==============================] - 5s 267us/sample - loss: 0.1037 - acc: 0.9678\n",
      "Epoch 3/3\n",
      "20000/20000 [==============================] - 5s 266us/sample - loss: 0.0424 - acc: 0.9893\n",
      "20000/20000 [==============================] - 4s 189us/sample - loss: 0.5163 - acc: 0.9040\n",
      "Epoch 1/3\n",
      "40000/40000 [==============================] - 15s 382us/sample - loss: 0.3989 - acc: 0.8617\n",
      "Epoch 2/3\n",
      "40000/40000 [==============================] - 11s 268us/sample - loss: 0.1567 - acc: 0.9438\n",
      "Epoch 3/3\n",
      "40000/40000 [==============================] - 11s 265us/sample - loss: 0.1050 - acc: 0.9642\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "RandomizedSearchCV(cv=KFold(n_splits=2, random_state=None, shuffle=False),\n",
       "                   error_score=nan,\n",
       "                   estimator=<keras.wrappers.scikit_learn.KerasClassifier object at 0x000001D6C8D142C8>,\n",
       "                   iid='deprecated', n_iter=10, n_jobs=None,\n",
       "                   param_distributions={'activation': ['relu'],\n",
       "                                        'batch_size': [64, 128, 256],\n",
       "                                        'epochs': [3, 5, 10],\n",
       "                                        'learning_rate': [0.1, 0.01, 0.001]},\n",
       "                   pre_dispatch='2*n_jobs', random_state=None, refit=True,\n",
       "                   return_train_score=False, scoring=None, verbose=0)"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# takes a lot of time ...\n",
    "random_search.fit(X_train, y_train,\n",
    "                  verbose = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "npgSXLyG90m_"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'learning_rate': 0.001, 'epochs': 3, 'batch_size': 256, 'activation': 'relu'}"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "random_search.best_params_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training set Accuracy:  98.2175\n",
      "Test set Accuracy:  92.73\n"
     ]
    }
   ],
   "source": [
    "best_random = random_search.best_estimator_ \n",
    "pred = best_random.predict(X_test)\n",
    "pred_train = best_random.predict(X_train)\n",
    "print(\"Training set Accuracy: \", accuracy_score(pred_train, y_train)*100)\n",
    "print(\"Test set Accuracy: \", accuracy_score(pred, y_test)*100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "def create_model(drop_out=0.2, units=64, learning_rate = 0.001):\n",
    "    model = Sequential()\n",
    "    model.add(Dropout(drop_out, input_shape=input_shape))\n",
    "    model.add(Dense(units, activation = 'relu'))\n",
    "    model.add(Dropout(drop_out))\n",
    "    model.add(Dense(1, activation = 'sigmoid'))\n",
    "    model.compile(optimizer=tf.keras.optimizers.Adam(lr=learning_rate), loss='binary_crossentropy', metrics=['accuracy'])\n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_params = KerasClassifier(build_fn = create_model)\n",
    "\n",
    "params = {'drop_out': [0.2, 0.15, 0.1, 0.01], 'units': [60, 70, 90],\n",
    "          'learning_rate': [0.001], 'batch_size': [128], 'epochs' : [5]}\n",
    "\n",
    "# Create a randomize search cv object passing in the parameters to try\n",
    "random_params = RandomizedSearchCV(model_params, param_distributions = params, cv = KFold(2))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 12s 611us/sample - loss: 0.5862 - acc: 0.8242\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 7s 374us/sample - loss: 0.3813 - acc: 0.8921\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 7s 370us/sample - loss: 0.2797 - acc: 0.9103\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 7s 369us/sample - loss: 0.2300 - acc: 0.9219\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 7s 371us/sample - loss: 0.1941 - acc: 0.9374\n",
      "20000/20000 [==============================] - 4s 187us/sample - loss: 0.2163 - acc: 0.9189\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 9s 446us/sample - loss: 0.5949 - acc: 0.8191s - loss: 0.6082 - \n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 7s 367us/sample - loss: 0.3908 - acc: 0.8887\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 7s 365us/sample - loss: 0.2878 - acc: 0.9068\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 7s 372us/sample - loss: 0.2350 - acc: 0.9217s\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 7s 373us/sample - loss: 0.2008 - acc: 0.9320\n",
      "20000/20000 [==============================] - 3s 169us/sample - loss: 0.2154 - acc: 0.9192\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 9s 455us/sample - loss: 0.5889 - acc: 0.8260\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 7s 367us/sample - loss: 0.3830 - acc: 0.8921\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 7s 373us/sample - loss: 0.2767 - acc: 0.9143\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 8s 386us/sample - loss: 0.2234 - acc: 0.9262\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 8s 391us/sample - loss: 0.1889 - acc: 0.9370\n",
      "20000/20000 [==============================] - 3s 170us/sample - loss: 0.2152 - acc: 0.9200\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 9s 471us/sample - loss: 0.5847 - acc: 0.8285\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 7s 369us/sample - loss: 0.3739 - acc: 0.8909\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 7s 373us/sample - loss: 0.2722 - acc: 0.9121\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 7s 372us/sample - loss: 0.2223 - acc: 0.9269\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 7s 372us/sample - loss: 0.1881 - acc: 0.9385\n",
      "20000/20000 [==============================] - 4s 179us/sample - loss: 0.2112 - acc: 0.9203\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 10s 488us/sample - loss: 0.5336 - acc: 0.8608\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 9s 428us/sample - loss: 0.2876 - acc: 0.9133\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 9s 427us/sample - loss: 0.1993 - acc: 0.9378\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 9s 438us/sample - loss: 0.1544 - acc: 0.9516\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 9s 428us/sample - loss: 0.1235 - acc: 0.9632\n",
      "20000/20000 [==============================] - 5s 260us/sample - loss: 0.1964 - acc: 0.9223\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 10s 496us/sample - loss: 0.5299 - acc: 0.8562\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 9s 430us/sample - loss: 0.2870 - acc: 0.9122\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 9s 431us/sample - loss: 0.2008 - acc: 0.9366\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 9s 433us/sample - loss: 0.1566 - acc: 0.9512\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 9s 474us/sample - loss: 0.1272 - acc: 0.9614\n",
      "20000/20000 [==============================] - 4s 196us/sample - loss: 0.1927 - acc: 0.9223\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 10s 481us/sample - loss: 0.5511 - acc: 0.8323\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 9s 428us/sample - loss: 0.3178 - acc: 0.9027\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 9s 434us/sample - loss: 0.2297 - acc: 0.9232\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 10s 477us/sample - loss: 0.1833 - acc: 0.9390\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 9s 426us/sample - loss: 0.1554 - acc: 0.9481\n",
      "20000/20000 [==============================] - 4s 182us/sample - loss: 0.1990 - acc: 0.9227\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 10s 497us/sample - loss: 0.5580 - acc: 0.8432\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 9s 443us/sample - loss: 0.3218 - acc: 0.8992\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 9s 459us/sample - loss: 0.2319 - acc: 0.9226\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 9s 429us/sample - loss: 0.1885 - acc: 0.9347\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 9s 429us/sample - loss: 0.1579 - acc: 0.9478\n",
      "20000/20000 [==============================] - 4s 192us/sample - loss: 0.1961 - acc: 0.9228\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 13s 651us/sample - loss: 0.5101 - acc: 0.8445\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 11s 536us/sample - loss: 0.2559 - acc: 0.9201\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 10s 520us/sample - loss: 0.1748 - acc: 0.9438\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 10s 522us/sample - loss: 0.1327 - acc: 0.9600\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 10s 523us/sample - loss: 0.1033 - acc: 0.9706\n",
      "20000/20000 [==============================] - 5s 248us/sample - loss: 0.1915 - acc: 0.9238\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 12s 592us/sample - loss: 0.5039 - acc: 0.8494\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 11s 530us/sample - loss: 0.2542 - acc: 0.9182\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 11s 531us/sample - loss: 0.1772 - acc: 0.9438\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 10s 517us/sample - loss: 0.1357 - acc: 0.9585\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 11s 555us/sample - loss: 0.1074 - acc: 0.9689\n",
      "20000/20000 [==============================] - 4s 216us/sample - loss: 0.1888 - acc: 0.9239\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 9s 469us/sample - loss: 0.5798 - acc: 0.8067\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 8s 377us/sample - loss: 0.3658 - acc: 0.8982s - loss: 0.3699 - acc:\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 8s 376us/sample - loss: 0.2622 - acc: 0.9180\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 8s 416us/sample - loss: 0.2109 - acc: 0.9326\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 8s 394us/sample - loss: 0.1759 - acc: 0.9444\n",
      "20000/20000 [==============================] - 4s 190us/sample - loss: 0.2112 - acc: 0.9208\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 9s 457us/sample - loss: 0.5838 - acc: 0.8389\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 7s 371us/sample - loss: 0.3708 - acc: 0.8946\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 8s 379us/sample - loss: 0.2658 - acc: 0.9172\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 8s 422us/sample - loss: 0.2151 - acc: 0.9308\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 8s 383us/sample - loss: 0.1807 - acc: 0.9427\n",
      "20000/20000 [==============================] - 4s 187us/sample - loss: 0.2094 - acc: 0.9209s - loss: 0.2137 - - ETA: 1s - los\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 10s 500us/sample - loss: 0.5467 - acc: 0.8538\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 9s 430us/sample - loss: 0.3081 - acc: 0.9049\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 9s 462us/sample - loss: 0.2202 - acc: 0.9286\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 9s 460us/sample - loss: 0.1758 - acc: 0.9417\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 9s 425us/sample - loss: 0.1471 - acc: 0.9513\n",
      "20000/20000 [==============================] - 4s 192us/sample - loss: 0.1978 - acc: 0.9226\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 10s 517us/sample - loss: 0.5417 - acc: 0.8480\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 9s 473us/sample - loss: 0.3076 - acc: 0.9039s - loss: 0.35 - ETA:  - ETA: 3s - loss:\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 9s 438us/sample - loss: 0.2206 - acc: 0.9283\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 9s 427us/sample - loss: 0.1760 - acc: 0.9426\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 9s 431us/sample - loss: 0.1484 - acc: 0.9520\n",
      "20000/20000 [==============================] - 4s 194us/sample - loss: 0.1941 - acc: 0.9236\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 19s 927us/sample - loss: 0.5213 - acc: 0.8367\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 10s 522us/sample - loss: 0.2749 - acc: 0.9122\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 10s 517us/sample - loss: 0.1952 - acc: 0.9351\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 10s 519us/sample - loss: 0.1536 - acc: 0.9485\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 11s 527us/sample - loss: 0.1265 - acc: 0.9596\n",
      "20000/20000 [==============================] - 7s 340us/sample - loss: 0.1926 - acc: 0.9238\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 13s 656us/sample - loss: 0.5166 - acc: 0.8526\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 11s 532us/sample - loss: 0.2751 - acc: 0.9117\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 10s 517us/sample - loss: 0.1981 - acc: 0.9335\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 11s 563us/sample - loss: 0.1573 - acc: 0.9484\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 10s 519us/sample - loss: 0.1298 - acc: 0.9571\n",
      "20000/20000 [==============================] - 5s 236us/sample - loss: 0.1892 - acc: 0.9243\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 16s 783us/sample - loss: 0.5534 - acc: 0.8316\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 9s 450us/sample - loss: 0.3038 - acc: 0.9085\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 9s 427us/sample - loss: 0.2113 - acc: 0.9319\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 9s 429us/sample - loss: 0.1675 - acc: 0.9457\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 9s 425us/sample - loss: 0.1375 - acc: 0.9565\n",
      "20000/20000 [==============================] - 9s 434us/sample - loss: 0.1961 - acc: 0.9234\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 10s 499us/sample - loss: 0.5437 - acc: 0.8378\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 9s 430us/sample - loss: 0.3026 - acc: 0.9096\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 9s 430us/sample - loss: 0.2144 - acc: 0.9297\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 9s 428us/sample - loss: 0.1715 - acc: 0.9441\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 9s 436us/sample - loss: 0.1413 - acc: 0.9564\n",
      "20000/20000 [==============================] - 6s 308us/sample - loss: 0.1937 - acc: 0.9232\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 15s 746us/sample - loss: 0.5234 - acc: 0.8429\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 10s 520us/sample - loss: 0.2794 - acc: 0.9088\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 11s 537us/sample - loss: 0.2013 - acc: 0.9329\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 11s 556us/sample - loss: 0.1607 - acc: 0.9453\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 11s 528us/sample - loss: 0.1350 - acc: 0.9552\n",
      "20000/20000 [==============================] - 5s 256us/sample - loss: 0.1943 - acc: 0.9248\n",
      "Epoch 1/5\n",
      "20000/20000 [==============================] - 12s 624us/sample - loss: 0.5160 - acc: 0.8392\n",
      "Epoch 2/5\n",
      "20000/20000 [==============================] - 11s 559us/sample - loss: 0.2791 - acc: 0.9100\n",
      "Epoch 3/5\n",
      "20000/20000 [==============================] - 10s 520us/sample - loss: 0.2040 - acc: 0.9294\n",
      "Epoch 4/5\n",
      "20000/20000 [==============================] - 10s 520us/sample - loss: 0.1637 - acc: 0.9445\n",
      "Epoch 5/5\n",
      "20000/20000 [==============================] - 10s 525us/sample - loss: 0.1409 - acc: 0.9513\n",
      "20000/20000 [==============================] - 7s 327us/sample - loss: 0.1902 - acc: 0.9244\n",
      "Epoch 1/5\n",
      "40000/40000 [==============================] - 28s 704us/sample - loss: 0.4105 - acc: 0.8709\n",
      "Epoch 2/5\n",
      "40000/40000 [==============================] - 21s 522us/sample - loss: 0.2088 - acc: 0.9221\n",
      "Epoch 3/5\n",
      "40000/40000 [==============================] - 22s 544us/sample - loss: 0.1633 - acc: 0.9401\n",
      "Epoch 4/5\n",
      "40000/40000 [==============================] - 21s 524us/sample - loss: 0.1365 - acc: 0.9505\n",
      "Epoch 5/5\n",
      "40000/40000 [==============================] - 21s 520us/sample - loss: 0.1195 - acc: 0.9565\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "RandomizedSearchCV(cv=KFold(n_splits=2, random_state=None, shuffle=False),\n",
       "                   error_score=nan,\n",
       "                   estimator=<keras.wrappers.scikit_learn.KerasClassifier object at 0x000001D6A62ECC48>,\n",
       "                   iid='deprecated', n_iter=10, n_jobs=None,\n",
       "                   param_distributions={'batch_size': [128],\n",
       "                                        'drop_out': [0.2, 0.15, 0.1, 0.01],\n",
       "                                        'epochs': [5], 'learning_rate': [0.001],\n",
       "                                        'units': [20, 40, 60]},\n",
       "                   pre_dispatch='2*n_jobs', random_state=None, refit=True,\n",
       "                   return_train_score=False, scoring=None, verbose=0)"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# takes a lot of time ...\n",
    "random_params.fit(X_train, y_train, verbose = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<keras.wrappers.scikit_learn.KerasClassifier at 0x1d916913148>"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "random_params.best_estimator_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'units': 60,\n",
       " 'learning_rate': 0.001,\n",
       " 'epochs': 5,\n",
       " 'drop_out': 0.2,\n",
       " 'batch_size': 128}"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "random_params.best_params_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training set Accuracy:  97.46000000000001\n",
      "Test set Accuracy:  92.83\n"
     ]
    }
   ],
   "source": [
    "best_random = random_params.best_estimator_ \n",
    "pred = best_random.predict(X_test)\n",
    "pred_train = best_random.predict(X_train)\n",
    "print(\"Training set Accuracy: \", accuracy_score(pred_train, y_train)*100)\n",
    "print(\"Test set Accuracy: \", accuracy_score(pred, y_test)*100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "lsvl7vs090is"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dropout_78 (Dropout)         (None, 20000)             0         \n",
      "_________________________________________________________________\n",
      "dense_128 (Dense)            (None, 64)                1280064   \n",
      "_________________________________________________________________\n",
      "dropout_79 (Dropout)         (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_129 (Dense)            (None, 1)                 65        \n",
      "=================================================================\n",
      "Total params: 1,280,129\n",
      "Trainable params: 1,280,129\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "input_shape = X_train.shape[1:]\n",
    "\n",
    "model4 = Sequential([ \n",
    "    Dropout(0.2, input_shape=input_shape),\n",
    "    Dense(64, activation = 'relu'),\n",
    "    Dropout(0.2),\n",
    "    Dense(1, activation = 'sigmoid'),\n",
    "])\n",
    "\n",
    "optimizer = tf.keras.optimizers.Adam(lr=0.002)\n",
    "model4.compile(optimizer=optimizer, loss='binary_crossentropy', metrics=['acc'])\n",
    "\n",
    "model4.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " ~| Epoch: 1 | Validation Loss: 0.19927 >|> \n",
      " ~| Epoch: 2 | Validation Loss: 0.17206 >|> \n",
      " ~| Epoch: 3 | Validation Loss: 0.17024 >|> \n",
      " ~| Epoch: 4 | Validation Loss: 0.17179 >|> \n",
      "Done.\n"
     ]
    }
   ],
   "source": [
    "epochs = 150\n",
    "batch_size = 128\n",
    "\n",
    "\n",
    "h4 = model4.fit(X_train, y_train, epochs=epochs,\n",
    "               validation_split = 0.2,\n",
    "               callbacks=[early_stop, simple_log],\n",
    "               verbose=False,\n",
    "               batch_size=batch_size)\n",
    "print(\"Done.\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x360 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_learning_curve(h4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10000/10000 [==============================] - 7s 668us/sample - loss: 0.1904 - acc: 0.9243s - loss: 0.1883\n",
      "Testing Accuracy:  92.43000149726868\n"
     ]
    }
   ],
   "source": [
    "loss, acc = model4.evaluate(X_test, y_test)\n",
    "print(\"Testing Accuracy: \", acc*100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "gMIV84-h90ex"
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "name": "Keras_Models_predictions.ipynb",
   "provenance": []
  },
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}