4. predict_MNIST_with_MLP.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\leesoojin\\Anaconda3\\lib\\site-packages\\h5py\\__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
      "  from ._conv import register_converters as _register_converters\n",
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from keras.models import Sequential\n",
    "from keras.layers import Dense\n",
    "from keras.utils import np_utils\n",
    "from keras.callbacks import ModelCheckpoint, EarlyStopping\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "import keras.backend.tensorflow_backend as K\n",
    "import tensorflow as tf\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib as mpl\n",
    "import seaborn as sns\n",
    "import os\n",
    "\n",
    "config = tf.ConfigProto()\n",
    "config.gpu_options.allow_growth = True\n",
    "session = tf.Session(config=config)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.datasets import mnist"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(60000, 28, 28)\n",
      "(10000, 28, 28)\n",
      "{0, 1, 2, 3, 4, 5, 6, 7, 8, 9}\n",
      "10\n"
     ]
    }
   ],
   "source": [
    "(X_train, y_train), (X_test, y_test) = mnist.load_data()\n",
    "\n",
    "print(X_train.shape)\n",
    "print(X_test.shape)\n",
    "print(set(y_train))\n",
    "print(len(set(y_train)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(60000, 10)\n"
     ]
    }
   ],
   "source": [
    "X_train = X_train.reshape(X_train.shape[0], 28*28).astype('float32') / 255   #255로 나누는 것은 rescale이다.\n",
    "X_test = X_test.reshape(X_test.shape[0], 28*28).astype('float32') / 255   #255로 나누는 것은 rescale이다.\n",
    "\n",
    "y_train = np_utils.to_categorical(y_train, len(set(y_train)))\n",
    "y_test = np_utils.to_categorical(y_test, len(set(y_test)))\n",
    "\n",
    "print(y_train.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 48000 samples, validate on 12000 samples\n",
      "Epoch 1/30\n",
      "48000/48000 [==============================] - 2s 38us/step - loss: 0.3380 - acc: 0.9053 - val_loss: 0.1679 - val_acc: 0.9533\n",
      "\n",
      "Epoch 00001: val_loss improved from inf to 0.16794, saving model to ./model/mnist_mlp.model\n",
      "Epoch 2/30\n",
      "48000/48000 [==============================] - 1s 23us/step - loss: 0.1425 - acc: 0.9596 - val_loss: 0.1223 - val_acc: 0.9658\n",
      "\n",
      "Epoch 00002: val_loss improved from 0.16794 to 0.12235, saving model to ./model/mnist_mlp.model\n",
      "Epoch 3/30\n",
      "48000/48000 [==============================] - 1s 22us/step - loss: 0.0955 - acc: 0.9726 - val_loss: 0.1063 - val_acc: 0.9687\n",
      "\n",
      "Epoch 00003: val_loss improved from 0.12235 to 0.10634, saving model to ./model/mnist_mlp.model\n",
      "Epoch 4/30\n",
      "48000/48000 [==============================] - 1s 22us/step - loss: 0.0711 - acc: 0.9798 - val_loss: 0.0899 - val_acc: 0.9727\n",
      "\n",
      "Epoch 00004: val_loss improved from 0.10634 to 0.08994, saving model to ./model/mnist_mlp.model\n",
      "Epoch 5/30\n",
      "48000/48000 [==============================] - 1s 22us/step - loss: 0.0519 - acc: 0.9852 - val_loss: 0.0818 - val_acc: 0.9750\n",
      "\n",
      "Epoch 00005: val_loss improved from 0.08994 to 0.08180, saving model to ./model/mnist_mlp.model\n",
      "Epoch 6/30\n",
      "48000/48000 [==============================] - 1s 21us/step - loss: 0.0419 - acc: 0.9884 - val_loss: 0.0776 - val_acc: 0.9771\n",
      "\n",
      "Epoch 00006: val_loss improved from 0.08180 to 0.07763, saving model to ./model/mnist_mlp.model\n",
      "Epoch 7/30\n",
      "48000/48000 [==============================] - 1s 22us/step - loss: 0.0307 - acc: 0.9921 - val_loss: 0.0740 - val_acc: 0.9772\n",
      "\n",
      "Epoch 00007: val_loss improved from 0.07763 to 0.07396, saving model to ./model/mnist_mlp.model\n",
      "Epoch 8/30\n",
      "48000/48000 [==============================] - 1s 22us/step - loss: 0.0236 - acc: 0.9939 - val_loss: 0.0734 - val_acc: 0.9768\n",
      "\n",
      "Epoch 00008: val_loss improved from 0.07396 to 0.07339, saving model to ./model/mnist_mlp.model\n",
      "Epoch 9/30\n",
      "48000/48000 [==============================] - 1s 21us/step - loss: 0.0188 - acc: 0.9957 - val_loss: 0.0752 - val_acc: 0.9780\n",
      "\n",
      "Epoch 00009: val_loss did not improve from 0.07339\n",
      "Epoch 10/30\n",
      "48000/48000 [==============================] - 1s 22us/step - loss: 0.0141 - acc: 0.9974 - val_loss: 0.0720 - val_acc: 0.9780\n",
      "\n",
      "Epoch 00010: val_loss improved from 0.07339 to 0.07202, saving model to ./model/mnist_mlp.model\n",
      "Epoch 11/30\n",
      "48000/48000 [==============================] - 1s 21us/step - loss: 0.0108 - acc: 0.9980 - val_loss: 0.0755 - val_acc: 0.9778\n",
      "\n",
      "Epoch 00011: val_loss did not improve from 0.07202\n",
      "Epoch 12/30\n",
      "48000/48000 [==============================] - 1s 22us/step - loss: 0.0087 - acc: 0.9989 - val_loss: 0.0714 - val_acc: 0.9788\n",
      "\n",
      "Epoch 00012: val_loss improved from 0.07202 to 0.07135, saving model to ./model/mnist_mlp.model\n",
      "Epoch 13/30\n",
      "48000/48000 [==============================] - 1s 22us/step - loss: 0.0066 - acc: 0.9992 - val_loss: 0.0756 - val_acc: 0.9778\n",
      "\n",
      "Epoch 00013: val_loss did not improve from 0.07135\n",
      "Epoch 14/30\n",
      "48000/48000 [==============================] - 1s 22us/step - loss: 0.0060 - acc: 0.9992 - val_loss: 0.0770 - val_acc: 0.9778\n",
      "\n",
      "Epoch 00014: val_loss did not improve from 0.07135\n",
      "Epoch 15/30\n",
      "48000/48000 [==============================] - 1s 23us/step - loss: 0.0041 - acc: 0.9997 - val_loss: 0.0724 - val_acc: 0.9790\n",
      "\n",
      "Epoch 00015: val_loss did not improve from 0.07135\n",
      "Epoch 16/30\n",
      "48000/48000 [==============================] - ETA: 0s - loss: 0.0027 - acc: 0.999 - 1s 23us/step - loss: 0.0027 - acc: 0.9999 - val_loss: 0.0723 - val_acc: 0.9806\n",
      "\n",
      "Epoch 00016: val_loss did not improve from 0.07135\n",
      "Epoch 17/30\n",
      "48000/48000 [==============================] - 1s 23us/step - loss: 0.0021 - acc: 1.0000 - val_loss: 0.0790 - val_acc: 0.9789\n",
      "\n",
      "Epoch 00017: val_loss did not improve from 0.07135\n",
      "Epoch 18/30\n",
      "48000/48000 [==============================] - 1s 22us/step - loss: 0.0023 - acc: 0.9999 - val_loss: 0.0742 - val_acc: 0.9798\n",
      "\n",
      "Epoch 00018: val_loss did not improve from 0.07135\n",
      "Epoch 19/30\n",
      "48000/48000 [==============================] - 1s 22us/step - loss: 0.0016 - acc: 0.9999 - val_loss: 0.0746 - val_acc: 0.9809\n",
      "\n",
      "Epoch 00019: val_loss did not improve from 0.07135\n",
      "Epoch 20/30\n",
      "48000/48000 [==============================] - 1s 23us/step - loss: 0.0013 - acc: 1.0000 - val_loss: 0.0779 - val_acc: 0.9798\n",
      "\n",
      "Epoch 00020: val_loss did not improve from 0.07135\n",
      "Epoch 21/30\n",
      "48000/48000 [==============================] - 1s 22us/step - loss: 0.0010 - acc: 1.0000 - val_loss: 0.0757 - val_acc: 0.9812\n",
      "\n",
      "Epoch 00021: val_loss did not improve from 0.07135\n",
      "Epoch 22/30\n",
      "48000/48000 [==============================] - 1s 22us/step - loss: 7.5664e-04 - acc: 1.0000 - val_loss: 0.0768 - val_acc: 0.9811\n",
      "\n",
      "Epoch 00022: val_loss did not improve from 0.07135\n"
     ]
    }
   ],
   "source": [
    "with K.tf_ops.device('/device:GPU:0'):\n",
    "    model = Sequential()\n",
    "    model.add(Dense(512, input_dim=X_train.shape[1], activation='relu'))\n",
    "    model.add(Dense(10, activation='softmax'))\n",
    "    \n",
    "    model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])\n",
    "    \n",
    "    model_dir = './model/'\n",
    "    \n",
    "    if not os.path.exists(model_dir):\n",
    "        os.mkdir(model_dir)\n",
    "    \n",
    "    modelpath = model_dir+'mnist_mlp.model'\n",
    "    \n",
    "    checkpointer = ModelCheckpoint(filepath = modelpath, monitor='val_loss', verbose=1, save_best_only=True)\n",
    "    \n",
    "    early_stop = EarlyStopping(monitor='val_loss', patience=10)\n",
    "    \n",
    "    history = model.fit(X_train, y_train, validation_split=0.2, epochs=30, batch_size=200, callbacks=[early_stop, checkpointer])\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10000/10000 [==============================] - 0s 30us/step\n",
      "모델 정확도 : 0.9835\n"
     ]
    }
   ],
   "source": [
    "print(\"모델 정확도 : %.4f\" % model.evaluate(X_test, y_test)[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "y_vloss = history.history['val_loss']\n",
    "y_loss = history.history['loss']\n",
    "\n",
    "x_len = np.arange(len(y_loss))\n",
    "\n",
    "plt.plot(x_len, y_vloss, marker='.', c='red', label='test_set_loss')\n",
    "plt.plot(x_len, y_loss, marker='.', c='blue', label='train_set_loss')\n",
    "plt.legend()\n",
    "plt.xlabel(\"epochs\")\n",
    "plt.ylabel(\"loss\")\n",
    "plt.grid()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}