Name_Generator.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "xMKZ2op-8__1"
   },
   "source": [
    "# Name Generation using RNN\n",
    "Generate new names using GRU network. Here a character level model has been used for this task."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "id": "cM5QcVU39HMX"
   },
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "from numpy.random import shuffle\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import pandas as pd\n",
    "from tqdm import tqdm\n",
    "from sklearn.model_selection import train_test_split\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "e4yxpDtx9tq7"
   },
   "source": [
    "##  Data loading"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "id": "443q_yW79aKI"
   },
   "outputs": [],
   "source": [
    "# for loading data\n",
    "def load_data(filename):\n",
    "    with open(filename, 'r') as f:\n",
    "        data = f.read().lower().split('\\n')\n",
    "    return np.array(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 429,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "base_uri": "https://localhost:8080/",
     "height": 36
    },
    "colab_type": "code",
    "executionInfo": {
     "elapsed": 2213,
     "status": "ok",
     "timestamp": 1530253300369,
     "user": {
      "displayName": "Susanta Biswas",
      "photoUrl": "//lh4.googleusercontent.com/-yoEv-8bwEKc/AAAAAAAAAAI/AAAAAAAAAAo/EHyV9CKwDzY/s50-c-k-no/photo.jpg",
      "userId": "101556937833327431006"
     },
     "user_tz": -330
    },
    "id": "EK99gahE-AKi",
    "outputId": "63cfdb4d-1d83-496d-dec9-d1ca8ee1fd7b"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No. of names: 4095\n"
     ]
    }
   ],
   "source": [
    "data = load_data('drive/Colab Notebooks/name_list.txt')\n",
    "print('No. of names: ' + str(len(data)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "yXZwdCE-c19D"
   },
   "source": [
    "Split data into training and validation set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 430,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "base_uri": "https://localhost:8080/",
     "height": 55
    },
    "colab_type": "code",
    "executionInfo": {
     "elapsed": 757,
     "status": "ok",
     "timestamp": 1530253303308,
     "user": {
      "displayName": "Susanta Biswas",
      "photoUrl": "//lh4.googleusercontent.com/-yoEv-8bwEKc/AAAAAAAAAAI/AAAAAAAAAAo/EHyV9CKwDzY/s50-c-k-no/photo.jpg",
      "userId": "101556937833327431006"
     },
     "user_tz": -330
    },
    "id": "MVtv_zQ7cz9M",
    "outputId": "309db46f-a97f-4a69-f038-1d8a5633eeb7"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3890\n",
      "205\n"
     ]
    }
   ],
   "source": [
    "# split data\n",
    "train_data, val_data = train_test_split(data, test_size=0.05)\n",
    "print(len(train_data))\n",
    "print(len(val_data))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "BzX9VyBzTw4P"
   },
   "source": [
    "## Text Processing\n",
    "We will use a character level model for this task. So first we will create a vocabulary dictionary. Since this is a character level model so the only the vocabulary size is fixed. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 431,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "base_uri": "https://localhost:8080/",
     "height": 36
    },
    "colab_type": "code",
    "executionInfo": {
     "elapsed": 1070,
     "status": "ok",
     "timestamp": 1530253305503,
     "user": {
      "displayName": "Susanta Biswas",
      "photoUrl": "//lh4.googleusercontent.com/-yoEv-8bwEKc/AAAAAAAAAAI/AAAAAAAAAAo/EHyV9CKwDzY/s50-c-k-no/photo.jpg",
      "userId": "101556937833327431006"
     },
     "user_tz": -330
    },
    "id": "dHXpk2zuTfei",
    "outputId": "c78ae952-619b-4df4-d968-57f1eea33856"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "29\n"
     ]
    }
   ],
   "source": [
    "# find the unique characters for the vocabulary\n",
    "#  Here '#' means End of Sentence\n",
    "vocab_tokens = ['#'] + list(set(''.join(data)))\n",
    "vocab_tokens_len = len(vocab_tokens)\n",
    "print(vocab_tokens_len)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 432,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "base_uri": "https://localhost:8080/",
     "height": 36
    },
    "colab_type": "code",
    "executionInfo": {
     "elapsed": 1020,
     "status": "ok",
     "timestamp": 1530253306825,
     "user": {
      "displayName": "Susanta Biswas",
      "photoUrl": "//lh4.googleusercontent.com/-yoEv-8bwEKc/AAAAAAAAAAI/AAAAAAAAAAo/EHyV9CKwDzY/s50-c-k-no/photo.jpg",
      "userId": "101556937833327431006"
     },
     "user_tz": -330
    },
    "id": "RGmouiWVVUtX",
    "outputId": "f494283d-4364-4da6-95de-77b0e9f4e1df"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Max length:  15\n"
     ]
    }
   ],
   "source": [
    "# find the max length amongst names\n",
    "max_name_len = max(len(name) for name in train_data)\n",
    "print('Max length: ', max_name_len)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "A8R8xAmGYnSO"
   },
   "source": [
    "For training the model we need to represent the string characters in terms of numbers. For that we will create a dictionary mapping of characters and numerical indices and vice-versa."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 433,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "base_uri": "https://localhost:8080/",
     "height": 169
    },
    "colab_type": "code",
    "executionInfo": {
     "elapsed": 825,
     "status": "ok",
     "timestamp": 1530253308376,
     "user": {
      "displayName": "Susanta Biswas",
      "photoUrl": "//lh4.googleusercontent.com/-yoEv-8bwEKc/AAAAAAAAAAI/AAAAAAAAAAo/EHyV9CKwDzY/s50-c-k-no/photo.jpg",
      "userId": "101556937833327431006"
     },
     "user_tz": -330
    },
    "id": "zg5X_7SXXuoY",
    "outputId": "68780e03-ea71-41e1-be69-319778466fbe"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "29it [00:00, 64527.75it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Dict len:  29\n",
      "{'#': 0, 'y': 1, 'm': 2, 'z': 3, 'a': 4, 'u': 5, 'n': 6, 'g': 7, 'h': 8, 'r': 9, 'l': 10, 's': 11, 'k': 12, 'o': 13, ';': 14, 'b': 15, 'f': 16, 'p': 17, 'e': 18, 'w': 19, 'q': 20, 'c': 21, '(': 22, 'x': 23, 'i': 24, 'v': 25, 'j': 26, 'd': 27, 't': 28}\n",
      "\n",
      "Dict len:  29\n",
      "{0: '#', 1: 'y', 2: 'm', 3: 'z', 4: 'a', 5: 'u', 6: 'n', 7: 'g', 8: 'h', 9: 'r', 10: 'l', 11: 's', 12: 'k', 13: 'o', 14: ';', 15: 'b', 16: 'f', 17: 'p', 18: 'e', 19: 'w', 20: 'q', 21: 'c', 22: '(', 23: 'x', 24: 'i', 25: 'v', 26: 'j', 27: 'd', 28: 't'}\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "# create vocabulary dict \n",
    "# character to index\n",
    "ch_to_idx = {ch:idx for idx, ch in tqdm(enumerate(vocab_tokens))}\n",
    "print('\\nDict len: ', len(ch_to_idx))\n",
    "print(ch_to_idx)\n",
    "\n",
    "# index to chracter\n",
    "idx_to_ch = {ch_to_idx[ch]:ch for ch in ch_to_idx.keys()}\n",
    "print('\\nDict len: ', len(idx_to_ch))\n",
    "print(idx_to_ch)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "id": "Avryng4jZkVg"
   },
   "outputs": [],
   "source": [
    "# for converting string names to matrix representation consisting of numerical indices\n",
    "# output shape: (m, Tx)\n",
    "def to_num_matrix(names, ch_to_idx):\n",
    "    \n",
    "    max_len = max(map(len, names))\n",
    "    names_idx = np.zeros((len(names), max_len), dtype='int32')\n",
    "        \n",
    "    for i, name in enumerate(names):\n",
    "        for j, ch in enumerate(name):\n",
    "            names_idx[i][j] = ch_to_idx[ch]\n",
    "        \n",
    "    return names_idx"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 435,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "base_uri": "https://localhost:8080/",
     "height": 36
    },
    "colab_type": "code",
    "executionInfo": {
     "elapsed": 864,
     "status": "ok",
     "timestamp": 1530253311890,
     "user": {
      "displayName": "Susanta Biswas",
      "photoUrl": "//lh4.googleusercontent.com/-yoEv-8bwEKc/AAAAAAAAAAI/AAAAAAAAAAo/EHyV9CKwDzY/s50-c-k-no/photo.jpg",
      "userId": "101556937833327431006"
     },
     "user_tz": -330
    },
    "id": "scRd9ClObP8E",
    "outputId": "f73a76b9-a05c-4dfb-f425-1e786a18dcc2"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(3890, 15)"
      ]
     },
     "execution_count": 435,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "to_num_matrix(train_data, ch_to_idx).shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "WdvYAu9sp3dA"
   },
   "source": [
    "## Training\n",
    "This is a simple model using Vanilla Recurrent units."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "id": "BuPJPvBx9CBR"
   },
   "outputs": [],
   "source": [
    "from IPython.display import clear_output\n",
    "from random import sample\n",
    "import pickle\n",
    "import sys"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "id": "zTXi_bT9cp8x"
   },
   "outputs": [],
   "source": [
    "from keras.layers import Embedding, Dense, Concatenate\n",
    "import keras\n",
    "import keras.layers as L\n",
    "from keras.utils import to_categorical\n",
    "from keras.callbacks import LambdaCallback, ModelCheckpoint\n",
    "from keras.models import Sequential, load_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "id": "OwrqLyd9ZVpG"
   },
   "outputs": [],
   "source": [
    "# for finding the validation accuracy\n",
    "def compute_acc():\n",
    "    # convert validation data\n",
    "    val_x = to_num_matrix(val_data, ch_to_idx)\n",
    "    # validation labels\n",
    "    val_y = np.copy(val_x[:,1:])\n",
    "    \n",
    "    # leave the last character\n",
    "    pred = model.predict(val_x[:,:-1], batch_size=256, verbose=1)\n",
    "    pred_ch = pred.argmax(axis=-1)\n",
    "    \n",
    "    # find acc\n",
    "    return float(np.sum(np.logical_and((pred_ch!=0), (pred_ch == val_y)))) \\\n",
    "                / np.sum(val_y != 0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "id": "1zfsFOeEcMd4"
   },
   "outputs": [],
   "source": [
    "# custom callback\n",
    "def acc_on_epoch_end(epoch, logs):\n",
    "    sys.stdout.flush()\n",
    "    print('\\nValidation Accuracy: ' + str(compute_acc()*100) + ' %')\n",
    "    sys.stdout.flush()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "id": "xDRj3FFPcfDD"
   },
   "outputs": [],
   "source": [
    "# for generating batches\n",
    "def generate_model_batches(names, batch_size=32, pad=0):\n",
    "    # no. of training examples\n",
    "    m = np.arange(len(names))\n",
    "    \n",
    "    while True:\n",
    "        # get a shuffled index list\n",
    "        idx = np.random.permutation(m)\n",
    "        \n",
    "        # start yeilding batches\n",
    "        for start in range(0, len(idx)-1, batch_size):\n",
    "            batch_idx = idx[start:start+batch_size]\n",
    "            batch_words = []\n",
    "            \n",
    "            # take out the words and tags from 'batch_size' no. of training examples\n",
    "            for index in batch_idx:\n",
    "                batch_words.append(names[index])\n",
    "            \n",
    "            # input x\n",
    "            batch_x = to_num_matrix(batch_words, ch_to_idx)\n",
    "            \n",
    "            # output labels \n",
    "            batch_y_ohe = to_categorical(batch_x[:,1:], len(vocab_tokens))\n",
    "            yield batch_x[:,:-1], batch_y_ohe"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "9_I6tQzOW2Oy"
   },
   "source": [
    "## Model\n",
    "We will use a simple GRU model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "id": "xPucu1H9fT88"
   },
   "outputs": [],
   "source": [
    "acc_callback = LambdaCallback(on_epoch_end=acc_on_epoch_end)\n",
    "checkpoint = ModelCheckpoint('drive/Colab Notebooks/best_weights.h5', verbose=1, save_best_only=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 461,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "base_uri": "https://localhost:8080/",
     "height": 375
    },
    "colab_type": "code",
    "executionInfo": {
     "elapsed": 1527,
     "status": "ok",
     "timestamp": 1530254043744,
     "user": {
      "displayName": "Susanta Biswas",
      "photoUrl": "//lh4.googleusercontent.com/-yoEv-8bwEKc/AAAAAAAAAAI/AAAAAAAAAAo/EHyV9CKwDzY/s50-c-k-no/photo.jpg",
      "userId": "101556937833327431006"
     },
     "user_tz": -330
    },
    "id": "fyGnDE78WL9m",
    "outputId": "db15c54e-79b7-4b7e-98b1-0ee64fa4d39a"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_21 (InputLayer)        (None, None)              0         \n",
      "_________________________________________________________________\n",
      "embedding_21 (Embedding)     (None, None, 20)          580       \n",
      "_________________________________________________________________\n",
      "gru_29 (GRU)                 (None, None, 128)         57216     \n",
      "_________________________________________________________________\n",
      "dropout_29 (Dropout)         (None, None, 128)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_29 (Batc (None, None, 128)         512       \n",
      "_________________________________________________________________\n",
      "time_distributed_21 (TimeDis (None, None, 29)          3741      \n",
      "=================================================================\n",
      "Total params: 62,049\n",
      "Trainable params: 61,793\n",
      "Non-trainable params: 256\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = Sequential()\n",
    "model.add(L.InputLayer([None], dtype='int32'))\n",
    "# embeddings layer\n",
    "model.add(L.Embedding(len(vocab_tokens), 20))\n",
    "\n",
    "# gru layer\n",
    "model.add(L.GRU(128, return_sequences=True, activation='tanh'))\n",
    "model.add(L.Dropout(0.5))\n",
    "model.add(L.BatchNormalization())\n",
    "\n",
    "\n",
    "# apply softmax\n",
    "model.add(L.TimeDistributed(L.Dense(len(vocab_tokens), activation='softmax')))\n",
    "\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 462,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "base_uri": "https://localhost:8080/",
     "height": 9508
    },
    "colab_type": "code",
    "executionInfo": {
     "elapsed": 81775,
     "status": "ok",
     "timestamp": 1530254127879,
     "user": {
      "displayName": "Susanta Biswas",
      "photoUrl": "//lh4.googleusercontent.com/-yoEv-8bwEKc/AAAAAAAAAAI/AAAAAAAAAAo/EHyV9CKwDzY/s50-c-k-no/photo.jpg",
      "userId": "101556937833327431006"
     },
     "user_tz": -330
    },
    "id": "G-J3u8xyX2JO",
    "outputId": "edf98b18-157a-4b0b-d678-7250b4234b5a"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "16/15 [===============================] - 6s 354ms/step - loss: 2.8621 - acc: 0.2947\n",
      "205/205 [==============================] - 2s 9ms/step\n",
      "\n",
      "Validation Accuracy: 8.10580204778157 %\n",
      "Epoch 2/100\n",
      " 3/15 [====>.........................] - ETA: 0s - loss: 2.2915 - acc: 0.4534"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/dist-packages/keras/callbacks.py:435: RuntimeWarning: Can save best model only with val_loss available, skipping.\n",
      "  'skipping.' % (self.monitor), RuntimeWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "16/15 [===============================] - 1s 43ms/step - loss: 2.1473 - acc: 0.4762\n",
      "205/205 [==============================] - 0s 47us/step\n",
      "\n",
      "Validation Accuracy: 16.040955631399317 %\n",
      "Epoch 3/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.7844 - acc: 0.5541\n",
      "205/205 [==============================] - 0s 65us/step\n",
      "\n",
      "Validation Accuracy: 26.27986348122867 %\n",
      "Epoch 4/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.5677 - acc: 0.6022\n",
      "205/205 [==============================] - 0s 50us/step\n",
      "\n",
      "Validation Accuracy: 34.38566552901024 %\n",
      "Epoch 5/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.4266 - acc: 0.6233\n",
      "205/205 [==============================] - 0s 50us/step\n",
      "\n",
      "Validation Accuracy: 32.67918088737201 %\n",
      "Epoch 6/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.3492 - acc: 0.6304\n",
      "205/205 [==============================] - 0s 62us/step\n",
      "\n",
      "Validation Accuracy: 33.36177474402731 %\n",
      "Epoch 7/100\n",
      "16/15 [===============================] - 1s 80ms/step - loss: 1.3219 - acc: 0.6292\n",
      "205/205 [==============================] - 0s 69us/step\n",
      "\n",
      "Validation Accuracy: 35.06825938566553 %\n",
      "Epoch 8/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.2864 - acc: 0.6360\n",
      "205/205 [==============================] - 0s 48us/step\n",
      "\n",
      "Validation Accuracy: 34.470989761092156 %\n",
      "Epoch 9/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.2706 - acc: 0.6342\n",
      "205/205 [==============================] - 0s 56us/step\n",
      "\n",
      "Validation Accuracy: 34.38566552901024 %\n",
      "Epoch 10/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.2470 - acc: 0.6387\n",
      "205/205 [==============================] - 0s 51us/step\n",
      "\n",
      "Validation Accuracy: 35.580204778157 %\n",
      "Epoch 11/100\n",
      " 6/15 [==========>...................] - ETA: 0s - loss: 1.2495 - acc: 0.637316/15 [===============================] - 1s 44ms/step - loss: 1.2381 - acc: 0.6356\n",
      "205/205 [==============================] - 0s 63us/step\n",
      "\n",
      "Validation Accuracy: 34.8122866894198 %\n",
      "Epoch 12/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.2230 - acc: 0.6374\n",
      "205/205 [==============================] - 0s 67us/step\n",
      "\n",
      "Validation Accuracy: 36.177474402730375 %\n",
      "Epoch 13/100\n",
      "16/15 [===============================] - 1s 42ms/step - loss: 1.2130 - acc: 0.6405\n",
      "205/205 [==============================] - 0s 53us/step\n",
      "\n",
      "Validation Accuracy: 35.665529010238906 %\n",
      "Epoch 14/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.2066 - acc: 0.6394\n",
      "205/205 [==============================] - 0s 55us/step\n",
      "\n",
      "Validation Accuracy: 35.92150170648464 %\n",
      "Epoch 15/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1953 - acc: 0.6424\n",
      "205/205 [==============================] - 0s 54us/step\n",
      "\n",
      "Validation Accuracy: 36.092150170648466 %\n",
      "Epoch 16/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.2017 - acc: 0.6392\n",
      "205/205 [==============================] - 0s 62us/step\n",
      "\n",
      "Validation Accuracy: 35.83617747440273 %\n",
      "Epoch 17/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1822 - acc: 0.6454\n",
      "205/205 [==============================] - 0s 49us/step\n",
      "\n",
      "Validation Accuracy: 35.665529010238906 %\n",
      "Epoch 18/100\n",
      "16/15 [===============================] - 1s 42ms/step - loss: 1.1922 - acc: 0.6380\n",
      "205/205 [==============================] - 0s 57us/step\n",
      "\n",
      "Validation Accuracy: 35.23890784982935 %\n",
      "Epoch 19/100\n",
      "16/15 [===============================] - 1s 42ms/step - loss: 1.1874 - acc: 0.6382\n",
      "205/205 [==============================] - 0s 59us/step\n",
      "\n",
      "Validation Accuracy: 35.580204778157 %\n",
      "Epoch 20/100\n",
      " 9/15 [================>.............] - ETA: 0s - loss: 1.1662 - acc: 0.645616/15 [===============================] - 1s 44ms/step - loss: 1.1796 - acc: 0.6402\n",
      "205/205 [==============================] - 0s 60us/step\n",
      "\n",
      "Validation Accuracy: 36.262798634812285 %\n",
      "Epoch 21/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1563 - acc: 0.6484\n",
      "205/205 [==============================] - 0s 49us/step\n",
      "\n",
      "Validation Accuracy: 34.641638225255974 %\n",
      "Epoch 22/100\n",
      "16/15 [===============================] - 1s 45ms/step - loss: 1.1724 - acc: 0.6409\n",
      "205/205 [==============================] - 0s 53us/step\n",
      "\n",
      "Validation Accuracy: 36.177474402730375 %\n",
      "Epoch 23/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1566 - acc: 0.6465\n",
      "205/205 [==============================] - 0s 48us/step\n",
      "\n",
      "Validation Accuracy: 35.49488054607509 %\n",
      "Epoch 24/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1488 - acc: 0.6492\n",
      "205/205 [==============================] - 0s 63us/step\n",
      "\n",
      "Validation Accuracy: 35.153583617747444 %\n",
      "Epoch 25/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1578 - acc: 0.6438\n",
      "205/205 [==============================] - 0s 58us/step\n",
      "\n",
      "Validation Accuracy: 35.750853242320815 %\n",
      "Epoch 26/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1415 - acc: 0.6498\n",
      "205/205 [==============================] - 0s 64us/step\n",
      "\n",
      "Validation Accuracy: 35.49488054607509 %\n",
      "Epoch 27/100\n",
      "16/15 [===============================] - 1s 42ms/step - loss: 1.1558 - acc: 0.6439\n",
      "205/205 [==============================] - 0s 59us/step\n",
      "\n",
      "Validation Accuracy: 36.00682593856656 %\n",
      "Epoch 28/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1395 - acc: 0.6490\n",
      "205/205 [==============================] - 0s 56us/step\n",
      "\n",
      "Validation Accuracy: 33.70307167235495 %\n",
      "Epoch 29/100\n",
      " 9/15 [================>.............] - ETA: 0s - loss: 1.1290 - acc: 0.651416/15 [===============================] - 1s 43ms/step - loss: 1.1430 - acc: 0.6473\n",
      "205/205 [==============================] - 0s 57us/step\n",
      "\n",
      "Validation Accuracy: 34.98293515358362 %\n",
      "Epoch 30/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1433 - acc: 0.6462\n",
      "205/205 [==============================] - 0s 56us/step\n",
      "\n",
      "Validation Accuracy: 34.8122866894198 %\n",
      "Epoch 31/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1254 - acc: 0.6513\n",
      "205/205 [==============================] - 0s 65us/step\n",
      "\n",
      "Validation Accuracy: 32.508532423208194 %\n",
      "Epoch 32/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1255 - acc: 0.6501\n",
      "205/205 [==============================] - 0s 51us/step\n",
      "\n",
      "Validation Accuracy: 34.55631399317406 %\n",
      "Epoch 33/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1404 - acc: 0.6467\n",
      "205/205 [==============================] - 0s 59us/step\n",
      "\n",
      "Validation Accuracy: 34.8976109215017 %\n",
      "Epoch 34/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1111 - acc: 0.6555\n",
      "205/205 [==============================] - 0s 48us/step\n",
      "\n",
      "Validation Accuracy: 33.020477815699664 %\n",
      "Epoch 35/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1379 - acc: 0.6456\n",
      "205/205 [==============================] - 0s 47us/step\n",
      "\n",
      "Validation Accuracy: 34.129692832764505 %\n",
      "Epoch 36/100\n",
      "16/15 [===============================] - 1s 42ms/step - loss: 1.1300 - acc: 0.6486\n",
      "205/205 [==============================] - 0s 55us/step\n",
      "\n",
      "Validation Accuracy: 33.617747440273035 %\n",
      "Epoch 37/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1250 - acc: 0.6505\n",
      "205/205 [==============================] - 0s 60us/step\n",
      "\n",
      "Validation Accuracy: 33.95904436860068 %\n",
      "Epoch 38/100\n",
      " 6/15 [==========>...................] - ETA: 0s - loss: 1.1269 - acc: 0.650016/15 [===============================] - 1s 44ms/step - loss: 1.1308 - acc: 0.6485\n",
      "205/205 [==============================] - 0s 54us/step\n",
      "\n",
      "Validation Accuracy: 32.84982935153584 %\n",
      "Epoch 39/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1182 - acc: 0.6517\n",
      "205/205 [==============================] - 0s 52us/step\n",
      "\n",
      "Validation Accuracy: 34.129692832764505 %\n",
      "Epoch 40/100\n",
      "16/15 [===============================] - 1s 45ms/step - loss: 1.1367 - acc: 0.6482\n",
      "205/205 [==============================] - 0s 58us/step\n",
      "\n",
      "Validation Accuracy: 32.93515358361775 %\n",
      "Epoch 41/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1356 - acc: 0.6452\n",
      "205/205 [==============================] - 0s 58us/step\n",
      "\n",
      "Validation Accuracy: 35.665529010238906 %\n",
      "Epoch 42/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1257 - acc: 0.6502\n",
      "205/205 [==============================] - 0s 58us/step\n",
      "\n",
      "Validation Accuracy: 33.70307167235495 %\n",
      "Epoch 43/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1253 - acc: 0.6484\n",
      "205/205 [==============================] - 0s 56us/step\n",
      "\n",
      "Validation Accuracy: 31.48464163822526 %\n",
      "Epoch 44/100\n",
      "16/15 [===============================] - 1s 45ms/step - loss: 1.1283 - acc: 0.6493\n",
      "205/205 [==============================] - 0s 69us/step\n",
      "\n",
      "Validation Accuracy: 33.36177474402731 %\n",
      "Epoch 45/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1072 - acc: 0.6542\n",
      "205/205 [==============================] - 0s 63us/step\n",
      "\n",
      "Validation Accuracy: 35.23890784982935 %\n",
      "Epoch 46/100\n",
      "16/15 [===============================] - 1s 45ms/step - loss: 1.1288 - acc: 0.6481\n",
      "205/205 [==============================] - 0s 62us/step\n",
      "\n",
      "Validation Accuracy: 35.40955631399317 %\n",
      "Epoch 47/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " 1/15 [>.............................] - ETA: 0s - loss: 1.0941 - acc: 0.664116/15 [===============================] - 1s 43ms/step - loss: 1.1010 - acc: 0.6571\n",
      "205/205 [==============================] - 0s 53us/step\n",
      "\n",
      "Validation Accuracy: 34.470989761092156 %\n",
      "Epoch 48/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1144 - acc: 0.6514\n",
      "205/205 [==============================] - 0s 60us/step\n",
      "\n",
      "Validation Accuracy: 33.87372013651877 %\n",
      "Epoch 49/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1153 - acc: 0.6516\n",
      "205/205 [==============================] - 0s 57us/step\n",
      "\n",
      "Validation Accuracy: 33.44709897610921 %\n",
      "Epoch 50/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1143 - acc: 0.6529\n",
      "205/205 [==============================] - 0s 63us/step\n",
      "\n",
      "Validation Accuracy: 32.76450511945392 %\n",
      "Epoch 51/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1092 - acc: 0.6521\n",
      "205/205 [==============================] - 0s 55us/step\n",
      "\n",
      "Validation Accuracy: 35.32423208191126 %\n",
      "Epoch 52/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1080 - acc: 0.6532\n",
      "205/205 [==============================] - 0s 63us/step\n",
      "\n",
      "Validation Accuracy: 34.8122866894198 %\n",
      "Epoch 53/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1020 - acc: 0.6558\n",
      "205/205 [==============================] - 0s 61us/step\n",
      "\n",
      "Validation Accuracy: 35.40955631399317 %\n",
      "Epoch 54/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1192 - acc: 0.6515\n",
      "205/205 [==============================] - 0s 59us/step\n",
      "\n",
      "Validation Accuracy: 35.153583617747444 %\n",
      "Epoch 55/100\n",
      "16/15 [===============================] - 1s 42ms/step - loss: 1.1168 - acc: 0.6531\n",
      "205/205 [==============================] - 0s 60us/step\n",
      "\n",
      "Validation Accuracy: 34.8122866894198 %\n",
      "Epoch 56/100\n",
      " 6/15 [==========>...................] - ETA: 0s - loss: 1.0868 - acc: 0.659016/15 [===============================] - 1s 43ms/step - loss: 1.1081 - acc: 0.6529\n",
      "205/205 [==============================] - 0s 59us/step\n",
      "\n",
      "Validation Accuracy: 33.44709897610921 %\n",
      "Epoch 57/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1011 - acc: 0.6559\n",
      "205/205 [==============================] - 0s 57us/step\n",
      "\n",
      "Validation Accuracy: 34.641638225255974 %\n",
      "Epoch 58/100\n",
      "16/15 [===============================] - 1s 45ms/step - loss: 1.0973 - acc: 0.6569\n",
      "205/205 [==============================] - 0s 59us/step\n",
      "\n",
      "Validation Accuracy: 33.87372013651877 %\n",
      "Epoch 59/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1052 - acc: 0.6552\n",
      "205/205 [==============================] - 0s 59us/step\n",
      "\n",
      "Validation Accuracy: 35.06825938566553 %\n",
      "Epoch 60/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1014 - acc: 0.6566\n",
      "205/205 [==============================] - 0s 51us/step\n",
      "\n",
      "Validation Accuracy: 35.580204778157 %\n",
      "Epoch 61/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1299 - acc: 0.6460\n",
      "205/205 [==============================] - 0s 57us/step\n",
      "\n",
      "Validation Accuracy: 36.262798634812285 %\n",
      "Epoch 62/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1218 - acc: 0.6487\n",
      "205/205 [==============================] - 0s 51us/step\n",
      "\n",
      "Validation Accuracy: 37.116040955631405 %\n",
      "Epoch 63/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1033 - acc: 0.6543\n",
      "205/205 [==============================] - 0s 49us/step\n",
      "\n",
      "Validation Accuracy: 35.665529010238906 %\n",
      "Epoch 64/100\n",
      "16/15 [===============================] - 1s 45ms/step - loss: 1.0976 - acc: 0.6563\n",
      "205/205 [==============================] - 0s 49us/step\n",
      "\n",
      "Validation Accuracy: 32.84982935153584 %\n",
      "Epoch 65/100\n",
      " 5/15 [========>.....................] - ETA: 0s - loss: 1.0633 - acc: 0.669016/15 [===============================] - 1s 44ms/step - loss: 1.0738 - acc: 0.6628\n",
      "205/205 [==============================] - 0s 59us/step\n",
      "\n",
      "Validation Accuracy: 32.25255972696246 %\n",
      "Epoch 66/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1056 - acc: 0.6524\n",
      "205/205 [==============================] - 0s 50us/step\n",
      "\n",
      "Validation Accuracy: 34.38566552901024 %\n",
      "Epoch 67/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.0910 - acc: 0.6598\n",
      "205/205 [==============================] - 0s 49us/step\n",
      "\n",
      "Validation Accuracy: 34.470989761092156 %\n",
      "Epoch 68/100\n",
      "16/15 [===============================] - 1s 45ms/step - loss: 1.1125 - acc: 0.6541\n",
      "205/205 [==============================] - 0s 51us/step\n",
      "\n",
      "Validation Accuracy: 34.8976109215017 %\n",
      "Epoch 69/100\n",
      "16/15 [===============================] - 1s 45ms/step - loss: 1.0953 - acc: 0.6577\n",
      "205/205 [==============================] - 0s 59us/step\n",
      "\n",
      "Validation Accuracy: 35.23890784982935 %\n",
      "Epoch 70/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.0937 - acc: 0.6561\n",
      "205/205 [==============================] - 0s 61us/step\n",
      "\n",
      "Validation Accuracy: 34.38566552901024 %\n",
      "Epoch 71/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.0978 - acc: 0.6560\n",
      "205/205 [==============================] - 0s 59us/step\n",
      "\n",
      "Validation Accuracy: 33.532423208191126 %\n",
      "Epoch 72/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.0876 - acc: 0.6595\n",
      "205/205 [==============================] - 0s 59us/step\n",
      "\n",
      "Validation Accuracy: 33.532423208191126 %\n",
      "Epoch 73/100\n",
      "16/15 [===============================] - 1s 46ms/step - loss: 1.0914 - acc: 0.6586\n",
      "205/205 [==============================] - 0s 63us/step\n",
      "\n",
      "Validation Accuracy: 33.87372013651877 %\n",
      "Epoch 74/100\n",
      " 6/15 [==========>...................] - ETA: 0s - loss: 1.0556 - acc: 0.667616/15 [===============================] - 1s 44ms/step - loss: 1.0818 - acc: 0.6617\n",
      "205/205 [==============================] - 0s 52us/step\n",
      "\n",
      "Validation Accuracy: 34.55631399317406 %\n",
      "Epoch 75/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.0957 - acc: 0.6578\n",
      "205/205 [==============================] - 0s 59us/step\n",
      "\n",
      "Validation Accuracy: 34.8976109215017 %\n",
      "Epoch 76/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.1071 - acc: 0.6524\n",
      "205/205 [==============================] - 0s 70us/step\n",
      "\n",
      "Validation Accuracy: 34.98293515358362 %\n",
      "Epoch 77/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.0855 - acc: 0.6617\n",
      "205/205 [==============================] - 0s 61us/step\n",
      "\n",
      "Validation Accuracy: 34.98293515358362 %\n",
      "Epoch 78/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.1152 - acc: 0.6494\n",
      "205/205 [==============================] - 0s 60us/step\n",
      "\n",
      "Validation Accuracy: 34.55631399317406 %\n",
      "Epoch 79/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.0942 - acc: 0.6567\n",
      "205/205 [==============================] - 0s 50us/step\n",
      "\n",
      "Validation Accuracy: 33.44709897610921 %\n",
      "Epoch 80/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.0762 - acc: 0.6635\n",
      "205/205 [==============================] - 0s 64us/step\n",
      "\n",
      "Validation Accuracy: 31.569965870307165 %\n",
      "Epoch 81/100\n",
      "16/15 [===============================] - 1s 45ms/step - loss: 1.0998 - acc: 0.6544\n",
      "205/205 [==============================] - 0s 56us/step\n",
      "\n",
      "Validation Accuracy: 32.67918088737201 %\n",
      "Epoch 82/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.0969 - acc: 0.6545\n",
      "205/205 [==============================] - 0s 69us/step\n",
      "\n",
      "Validation Accuracy: 33.95904436860068 %\n",
      "Epoch 83/100\n",
      " 3/15 [====>.........................] - ETA: 0s - loss: 1.0946 - acc: 0.658216/15 [===============================] - 1s 44ms/step - loss: 1.0940 - acc: 0.6572\n",
      "205/205 [==============================] - 0s 51us/step\n",
      "\n",
      "Validation Accuracy: 34.470989761092156 %\n",
      "Epoch 84/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.0720 - acc: 0.6632\n",
      "205/205 [==============================] - 0s 51us/step\n",
      "\n",
      "Validation Accuracy: 33.44709897610921 %\n",
      "Epoch 85/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.0803 - acc: 0.6592\n",
      "205/205 [==============================] - 0s 60us/step\n",
      "\n",
      "Validation Accuracy: 32.081911262798634 %\n",
      "Epoch 86/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.0686 - acc: 0.6646\n",
      "205/205 [==============================] - 0s 61us/step\n",
      "\n",
      "Validation Accuracy: 32.76450511945392 %\n",
      "Epoch 87/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.0950 - acc: 0.6563\n",
      "205/205 [==============================] - 0s 64us/step\n",
      "\n",
      "Validation Accuracy: 34.8122866894198 %\n",
      "Epoch 88/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.0841 - acc: 0.6616\n",
      "205/205 [==============================] - 0s 61us/step\n",
      "\n",
      "Validation Accuracy: 35.92150170648464 %\n",
      "Epoch 89/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.0829 - acc: 0.6605\n",
      "205/205 [==============================] - 0s 63us/step\n",
      "\n",
      "Validation Accuracy: 33.19112627986348 %\n",
      "Epoch 90/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.0578 - acc: 0.6681\n",
      "205/205 [==============================] - 0s 52us/step\n",
      "\n",
      "Validation Accuracy: 34.215017064846414 %\n",
      "Epoch 91/100\n",
      "16/15 [===============================] - 1s 43ms/step - loss: 1.0912 - acc: 0.6561\n",
      "205/205 [==============================] - 0s 55us/step\n",
      "\n",
      "Validation Accuracy: 35.23890784982935 %\n",
      "Epoch 92/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " 4/15 [======>.......................] - ETA: 0s - loss: 1.0655 - acc: 0.664416/15 [===============================] - 1s 44ms/step - loss: 1.0668 - acc: 0.6655\n",
      "205/205 [==============================] - 0s 56us/step\n",
      "\n",
      "Validation Accuracy: 34.8976109215017 %\n",
      "Epoch 93/100\n",
      "16/15 [===============================] - 1s 42ms/step - loss: 1.0957 - acc: 0.6565\n",
      "205/205 [==============================] - 0s 61us/step\n",
      "\n",
      "Validation Accuracy: 35.83617747440273 %\n",
      "Epoch 94/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.0966 - acc: 0.6551\n",
      "205/205 [==============================] - 0s 60us/step\n",
      "\n",
      "Validation Accuracy: 35.665529010238906 %\n",
      "Epoch 95/100\n",
      "16/15 [===============================] - 1s 45ms/step - loss: 1.0681 - acc: 0.6641\n",
      "205/205 [==============================] - 0s 64us/step\n",
      "\n",
      "Validation Accuracy: 33.87372013651877 %\n",
      "Epoch 96/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.0560 - acc: 0.6659\n",
      "205/205 [==============================] - 0s 52us/step\n",
      "\n",
      "Validation Accuracy: 34.129692832764505 %\n",
      "Epoch 97/100\n",
      "16/15 [===============================] - 1s 45ms/step - loss: 1.0822 - acc: 0.6591\n",
      "205/205 [==============================] - 0s 60us/step\n",
      "\n",
      "Validation Accuracy: 34.8976109215017 %\n",
      "Epoch 98/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.0805 - acc: 0.6598\n",
      "205/205 [==============================] - 0s 64us/step\n",
      "\n",
      "Validation Accuracy: 35.665529010238906 %\n",
      "Epoch 99/100\n",
      "16/15 [===============================] - 1s 44ms/step - loss: 1.0732 - acc: 0.6622\n",
      "205/205 [==============================] - 0s 51us/step\n",
      "\n",
      "Validation Accuracy: 35.153583617747444 %\n",
      "Epoch 100/100\n",
      "16/15 [===============================] - 1s 45ms/step - loss: 1.0730 - acc: 0.6610\n",
      "205/205 [==============================] - 0s 62us/step\n",
      "\n",
      "Validation Accuracy: 36.43344709897611 %\n"
     ]
    }
   ],
   "source": [
    "BATCH_SIZE = 256\n",
    "\n",
    "# select optimizer\n",
    "optimizer = keras.optimizers.Adam(clipvalue=1.2)\n",
    "model.compile(optimizer, loss='categorical_crossentropy', metrics=['accuracy'])\n",
    "\n",
    "# train model\n",
    "hist = model.fit_generator(generate_model_batches(train_data, batch_size=BATCH_SIZE), steps_per_epoch=len(train_data)/BATCH_SIZE,\n",
    "                          callbacks=[acc_callback, checkpoint], epochs=100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "id": "PR7vTNjek_76"
   },
   "outputs": [],
   "source": [
    "# save model \n",
    "model.save('drive/Colab Notebooks/model.h5')\n",
    "model.save_weights('drive/Colab Notebooks/weights.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 464,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "base_uri": "https://localhost:8080/",
     "height": 695
    },
    "colab_type": "code",
    "executionInfo": {
     "elapsed": 1849,
     "status": "ok",
     "timestamp": 1530254137948,
     "user": {
      "displayName": "Susanta Biswas",
      "photoUrl": "//lh4.googleusercontent.com/-yoEv-8bwEKc/AAAAAAAAAAI/AAAAAAAAAAo/EHyV9CKwDzY/s50-c-k-no/photo.jpg",
      "userId": "101556937833327431006"
     },
     "user_tz": -330
    },
    "id": "U2QLbwF589Vi",
    "outputId": "25542328-f093-463d-9ab2-3fae4d02e0db"
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA7gAAAFTCAYAAAATJaAzAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n2z0SdbJ3eAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f2ea5ca8ef0>"
      ]
     },
     "metadata": {
      "tags": []
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA7gAAAFTCAYAAAATJaAzAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n5Arb3QAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f2ea83d59b0>"
      ]
     },
     "metadata": {
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# loss plot\n",
    "plt.figure(figsize=(16,5))\n",
    "plt.subplot(1,1,1)\n",
    "plt.suptitle('Optimizer: Adam', fontsize=10)\n",
    "plt.ylabel('Loss', fontsize=16)\n",
    "plt.plot(hist.history['loss'], color='b', label='Training Loss')\n",
    "plt.legend(loc='upper right')\n",
    "plt.savefig('drive/Colab Notebooks/loss.png')\n",
    "\n",
    "# acc plot\n",
    "plt.figure(figsize=(16,5))\n",
    "plt.subplot(1,1,1)\n",
    "plt.suptitle('Optimizer: Adam', fontsize=10)\n",
    "plt.ylabel('Accuracy', fontsize=16)\n",
    "plt.plot(hist.history['acc'], color='b', label='Training Loss')\n",
    "plt.legend(loc='lower right')\n",
    "\n",
    "plt.savefig('drive/Colab Notebooks/acc.png')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "WNf_RoyhAXi_"
   },
   "source": [
    "## Sample Text\n",
    "Generate new names by feeding a seed input."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "id": "IuXgLuN4Aexw"
   },
   "outputs": [],
   "source": [
    "# samples text from language model\n",
    "def sample_text(seed_phrase='a', max_length=max_name_len):\n",
    "  \n",
    "            \n",
    "    # input x to numerical index\n",
    "    text_x = to_num_matrix(list(list(seed_phrase)), ch_to_idx)\n",
    "    # index of current character\n",
    "    ch = ch_to_idx[seed_phrase]\n",
    "    \n",
    "    # list of predicted character indices\n",
    "    indices = []\n",
    "    # counter\n",
    "    i = 0\n",
    "    \n",
    "    #start generating\n",
    "    while ch != 0 and i<max_length:\n",
    "        # predict using the input character\n",
    "        pred = model.predict(text_x)[0]\n",
    "        \n",
    "        # get the char with highest prob.\n",
    "        ch = np.random.choice(np.arange(len(vocab_tokens)) , p = pred.ravel())\n",
    "        \n",
    "        if ch!=0:\n",
    "            indices.append(ch)\n",
    "        \n",
    "        # feed the current char as the next input\n",
    "        text_x = to_num_matrix(list(list(idx_to_ch[ch])), ch_to_idx)\n",
    "        \n",
    "        i = i+ 1\n",
    "    \n",
    "    \n",
    "    return ''.join([idx_to_ch[idx] for idx in indices])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 472,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "base_uri": "https://localhost:8080/",
     "height": 206
    },
    "colab_type": "code",
    "executionInfo": {
     "elapsed": 909,
     "status": "ok",
     "timestamp": 1530254173668,
     "user": {
      "displayName": "Susanta Biswas",
      "photoUrl": "//lh4.googleusercontent.com/-yoEv-8bwEKc/AAAAAAAAAAI/AAAAAAAAAAo/EHyV9CKwDzY/s50-c-k-no/photo.jpg",
      "userId": "101556937833327431006"
     },
     "user_tz": -330
    },
    "id": "L24vXkEIP04D",
    "outputId": "f5cb4788-3828-4404-c965-e02f611e1e44"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "rikata\n",
      "denesh\n",
      "nirami\n",
      "pobhes\n",
      "reeera\n",
      "prabak\n",
      "moodar\n",
      "janika\n",
      "fusuvm\n",
      "rukami\n"
     ]
    }
   ],
   "source": [
    "i = 0\n",
    "while i < 10:\n",
    "    name = sample_text(max_length=6)\n",
    "    if name != '':\n",
    "        print(name)\n",
    "        i += 1"
   ]
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "collapsed_sections": [],
   "default_view": {},
   "name": "Name Generator.ipynb",
   "provenance": [],
   "version": "0.3.2",
   "views": {}
  },
  "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": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}