3.2_expression_dnn.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 表情辨識 Image Sentiment Classification\n",
    "## DNN model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda3/lib/python3.6/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": [
    "# coding: utf-8\n",
    "# libraries & packages\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import math\n",
    "from random import randint\n",
    "\n",
    "import sys\n",
    "\n",
    "from keras.models import Sequential\n",
    "from keras.models import load_model\n",
    "from keras.layers import Dense, Dropout, Activation, Flatten\n",
    "from keras.layers import Convolution2D, MaxPooling2D\n",
    "from keras.optimizers import SGD, Adam\n",
    "from keras.utils import np_utils\n",
    "from sklearn.utils import shuffle"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "   label                                            feature\n",
      "0      0  70 80 82 72 58 58 60 63 54 58 60 48 89 115 121...\n",
      "1      0  151 150 147 155 148 133 111 140 170 174 182 15...\n",
      "2      2  231 212 156 164 174 138 161 173 182 200 106 38...\n",
      "3      4  24 32 36 30 32 23 19 20 30 41 21 22 32 34 21 1...\n",
      "4      6  4 0 0 0 0 0 0 0 0 0 0 0 3 15 23 28 48 50 58 84...\n",
      "(28709,) (28709,) (7178,)\n"
     ]
    }
   ],
   "source": [
    "# load data\n",
    "def load_data(train_data_path, test_data_path):\n",
    "    train = pd.read_csv(train_data_path, sep=',', header=0)\n",
    "    test = pd.read_csv(test_data_path, sep=',', header=0)\n",
    "    return (train, test)\n",
    "\n",
    "data_path = '/Users/PK/.kaggle/competitions/ml-2017fall-hw3/'\n",
    "train, test = load_data(data_path + \"train.csv\", data_path + \"test.csv\")\n",
    "X_train_raw =train[\"feature\"]\n",
    "Y_train_raw =train[\"label\"]\n",
    "X_test_raw =test[\"feature\"]\n",
    "print(train.head())\n",
    "print(X_train_raw.shape, Y_train_raw.shape, X_test_raw.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(28709, 2304) (7178, 2304) (28709, 7)\n"
     ]
    }
   ],
   "source": [
    "# Reshape\n",
    "classes = 7\n",
    "def _sep(x):\n",
    "    x = x.split(\" \")\n",
    "    x = np.array(x).astype(\"float32\")\n",
    "    return x\n",
    "\n",
    "def sep_reshape(raw_data):\n",
    "    data = raw_data.apply(_sep)\n",
    "#     data2 = np.stack(data).reshape((len(data),1,48,48))\n",
    "    data2 = np.stack(data)\n",
    "    return data2\n",
    "\n",
    "X_train = sep_reshape(X_train_raw)\n",
    "X_test = sep_reshape(X_test_raw)\n",
    "\n",
    "# normalize inputs from 0-255 to 0.0-1.0X_train /= 255.0\n",
    "X_train /= 255.0\n",
    "X_test /= 255.0\n",
    "\n",
    "# 1-of-N encoding\n",
    "Y_train = np_utils.to_categorical(np.array(Y_train_raw), classes)\n",
    "\n",
    "\n",
    "print(X_train.shape, X_test.shape,Y_train.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0.27450982, 0.3137255 , 0.32156864, ..., 0.41568628, 0.42745098,\n",
       "       0.32156864], dtype=float32)"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_1 (Dense)              (None, 400)               922000    \n",
      "_________________________________________________________________\n",
      "activation_1 (Activation)    (None, 400)               0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 256)               102656    \n",
      "_________________________________________________________________\n",
      "activation_2 (Activation)    (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 256)               65792     \n",
      "_________________________________________________________________\n",
      "activation_3 (Activation)    (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 256)               65792     \n",
      "_________________________________________________________________\n",
      "activation_4 (Activation)    (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense_5 (Dense)              (None, 7)                 1799      \n",
      "_________________________________________________________________\n",
      "activation_5 (Activation)    (None, 7)                 0         \n",
      "=================================================================\n",
      "Total params: 1,158,039\n",
      "Trainable params: 1,158,039\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Train on 28220 samples, validate on 489 samples\n",
      "Epoch 1/20\n",
      "28220/28220 [==============================] - 10s - loss: 1.7950 - acc: 0.2587 - val_loss: 1.7751 - val_acc: 0.2658\n",
      "Epoch 2/20\n",
      "28220/28220 [==============================] - 9s - loss: 1.6933 - acc: 0.3259 - val_loss: 1.7007 - val_acc: 0.3313\n",
      "Epoch 3/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.6615 - acc: 0.3439 - val_loss: 1.7061 - val_acc: 0.3006\n",
      "Epoch 4/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.6415 - acc: 0.3556 - val_loss: 1.6658 - val_acc: 0.3333\n",
      "Epoch 5/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.6210 - acc: 0.3634 - val_loss: 1.7252 - val_acc: 0.3272\n",
      "Epoch 6/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.6030 - acc: 0.3723 - val_loss: 1.6230 - val_acc: 0.3476\n",
      "Epoch 7/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.5871 - acc: 0.3764 - val_loss: 1.6182 - val_acc: 0.3640\n",
      "Epoch 8/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.5671 - acc: 0.3880 - val_loss: 1.6215 - val_acc: 0.3824\n",
      "Epoch 9/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.5604 - acc: 0.3866 - val_loss: 1.6325 - val_acc: 0.3579\n",
      "Epoch 10/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.5427 - acc: 0.3962 - val_loss: 1.5857 - val_acc: 0.3640\n",
      "Epoch 11/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.5307 - acc: 0.3991 - val_loss: 1.5897 - val_acc: 0.3885\n",
      "Epoch 12/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.5186 - acc: 0.4090 - val_loss: 1.6556 - val_acc: 0.3456\n",
      "Epoch 13/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.5058 - acc: 0.4136 - val_loss: 1.5972 - val_acc: 0.3620\n",
      "Epoch 14/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.4793 - acc: 0.4258 - val_loss: 1.6060 - val_acc: 0.3620\n",
      "Epoch 15/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.4672 - acc: 0.4281 - val_loss: 1.5977 - val_acc: 0.4008\n",
      "Epoch 16/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.4503 - acc: 0.4365 - val_loss: 1.5993 - val_acc: 0.3456\n",
      "Epoch 17/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.4417 - acc: 0.4398 - val_loss: 1.5792 - val_acc: 0.3865\n",
      "Epoch 18/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.4212 - acc: 0.4465 - val_loss: 1.5726 - val_acc: 0.3865\n",
      "Epoch 19/20\n",
      "28220/28220 [==============================] - 8s - loss: 1.4099 - acc: 0.4527 - val_loss: 1.5440 - val_acc: 0.3906\n",
      "Epoch 20/20\n",
      "28220/28220 [==============================] - 7s - loss: 1.3843 - acc: 0.4640 - val_loss: 1.5971 - val_acc: 0.4049\n"
     ]
    }
   ],
   "source": [
    "'''Build a dnn model'''\n",
    "# CNN use Total params: 1,031,153\n",
    "model_adam = Sequential()\n",
    "model_adam.add(Dense(400, input_dim=48*48))\n",
    "model_adam.add(Activation('softplus'))\n",
    "model_adam.add(Dense(256))\n",
    "model_adam.add(Activation('softplus'))\n",
    "model_adam.add(Dense(256))\n",
    "model_adam.add(Activation('softplus'))\n",
    "model_adam.add(Dense(256))\n",
    "model_adam.add(Activation('softplus'))\n",
    "model_adam.add(Dense(7))\n",
    "model_adam.add(Activation('softmax'))\n",
    "\n",
    "''' Setting optimizer as Adam '''\n",
    "from keras.optimizers import Adam\n",
    "model_adam.compile(loss= 'categorical_crossentropy',\n",
    "                  optimizer='Adam',\n",
    "                  metrics=['accuracy'])\n",
    "# # check parameters of every layers\n",
    "model_adam.summary()\n",
    "\n",
    "''' training'''\n",
    "batch_size = 100\n",
    "epoch = 20\n",
    "v_s =0.017\n",
    "# validation data comes from training data\n",
    "fit = model_adam.fit(X_train, Y_train, batch_size=batch_size,\n",
    "      epochs=epoch, validation_split=v_s, shuffle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1a37b71550>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 驗證\n",
    "def plot_acc_loss(history):\n",
    "#     plt.clf()\n",
    "    fig = plt.figure()\n",
    "    fig.set_figwidth(10)\n",
    "    plt.subplot(1,2,1)\n",
    "    plt.xlabel('Epoch')\n",
    "    plt.ylabel('loss')\n",
    "    plt.title('Training data') \n",
    "    plt.plot(history.history['loss']) \n",
    "    plt.plot(history.history['val_loss']) \n",
    "    plt.legend(['training','validation'],loc='lower right') \n",
    "    \n",
    "    plt.subplot(1,2,2)\n",
    "    plt.xlabel('Epoch')\n",
    "    plt.ylabel('Accuracy')\n",
    "    plt.title('Training data') \n",
    "    plt.plot(history.history['acc']) \n",
    "    plt.plot(history.history['val_acc']) \n",
    "    plt.legend(['training','validation'],loc='lower right') \n",
    "    plt.show()\n",
    "plot_acc_loss(fit)"
   ]
  },
  {
   "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}