iris(DNN-tensorflow2.0)-Sequential.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import warnings\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "warnings.filterwarnings(\"ignore\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1) 載入資料集"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "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>SepalLengthCm</th>\n",
       "      <th>SepalWidthCm</th>\n",
       "      <th>PetalLengthCm</th>\n",
       "      <th>PetalWidthCm</th>\n",
       "      <th>Species</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>5.1</td>\n",
       "      <td>3.5</td>\n",
       "      <td>1.4</td>\n",
       "      <td>0.2</td>\n",
       "      <td>Iris-setosa</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>4.9</td>\n",
       "      <td>3.0</td>\n",
       "      <td>1.4</td>\n",
       "      <td>0.2</td>\n",
       "      <td>Iris-setosa</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>4.7</td>\n",
       "      <td>3.2</td>\n",
       "      <td>1.3</td>\n",
       "      <td>0.2</td>\n",
       "      <td>Iris-setosa</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4.6</td>\n",
       "      <td>3.1</td>\n",
       "      <td>1.5</td>\n",
       "      <td>0.2</td>\n",
       "      <td>Iris-setosa</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5.0</td>\n",
       "      <td>3.6</td>\n",
       "      <td>1.4</td>\n",
       "      <td>0.2</td>\n",
       "      <td>Iris-setosa</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>145</th>\n",
       "      <td>6.7</td>\n",
       "      <td>3.0</td>\n",
       "      <td>5.2</td>\n",
       "      <td>2.3</td>\n",
       "      <td>Iris-virginica</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>146</th>\n",
       "      <td>6.3</td>\n",
       "      <td>2.5</td>\n",
       "      <td>5.0</td>\n",
       "      <td>1.9</td>\n",
       "      <td>Iris-virginica</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>147</th>\n",
       "      <td>6.5</td>\n",
       "      <td>3.0</td>\n",
       "      <td>5.2</td>\n",
       "      <td>2.0</td>\n",
       "      <td>Iris-virginica</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>148</th>\n",
       "      <td>6.2</td>\n",
       "      <td>3.4</td>\n",
       "      <td>5.4</td>\n",
       "      <td>2.3</td>\n",
       "      <td>Iris-virginica</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>149</th>\n",
       "      <td>5.9</td>\n",
       "      <td>3.0</td>\n",
       "      <td>5.1</td>\n",
       "      <td>1.8</td>\n",
       "      <td>Iris-virginica</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>150 rows × 5 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     SepalLengthCm  SepalWidthCm  PetalLengthCm  PetalWidthCm         Species\n",
       "0              5.1           3.5            1.4           0.2     Iris-setosa\n",
       "1              4.9           3.0            1.4           0.2     Iris-setosa\n",
       "2              4.7           3.2            1.3           0.2     Iris-setosa\n",
       "3              4.6           3.1            1.5           0.2     Iris-setosa\n",
       "4              5.0           3.6            1.4           0.2     Iris-setosa\n",
       "..             ...           ...            ...           ...             ...\n",
       "145            6.7           3.0            5.2           2.3  Iris-virginica\n",
       "146            6.3           2.5            5.0           1.9  Iris-virginica\n",
       "147            6.5           3.0            5.2           2.0  Iris-virginica\n",
       "148            6.2           3.4            5.4           2.3  Iris-virginica\n",
       "149            5.9           3.0            5.1           1.8  Iris-virginica\n",
       "\n",
       "[150 rows x 5 columns]"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_train = pd.read_csv('./data/Iris.csv')\n",
    "df_train = df_train.drop(labels=['Id'],axis=1) # 移除Id\n",
    "df_train"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2) 手動編碼\n",
    "處理名目資料 (Nominal variables) - 資料前處理\n",
    "依據特徵資料的特性，可以選擇手動編碼或自動編碼。\n",
    "\n",
    "### 使用編碼時機?\n",
    "進行深度學習時，神經網路只能處理數值資料。因此我們需要將所有非數字型態的特徵進行轉換。\n",
    "\n",
    "ex:\n",
    "\n",
    "|  Iris-setosa |  Iris-versicolor | Iris-virginica  |\n",
    "|:---:|:---:|:---:|\n",
    "|  1 | 2  | 3  |"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_map = {'Iris-setosa': 0, 'Iris-versicolor': 1, 'Iris-virginica': 2}\n",
    "\n",
    "#將編碼後的label map存至df_train['Species']中。\n",
    "df_train['Class'] = df_train['Species'].map(label_map)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "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>SepalLengthCm</th>\n",
       "      <th>SepalWidthCm</th>\n",
       "      <th>PetalLengthCm</th>\n",
       "      <th>PetalWidthCm</th>\n",
       "      <th>Species</th>\n",
       "      <th>Class</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>5.1</td>\n",
       "      <td>3.5</td>\n",
       "      <td>1.4</td>\n",
       "      <td>0.2</td>\n",
       "      <td>Iris-setosa</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>4.9</td>\n",
       "      <td>3.0</td>\n",
       "      <td>1.4</td>\n",
       "      <td>0.2</td>\n",
       "      <td>Iris-setosa</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>4.7</td>\n",
       "      <td>3.2</td>\n",
       "      <td>1.3</td>\n",
       "      <td>0.2</td>\n",
       "      <td>Iris-setosa</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4.6</td>\n",
       "      <td>3.1</td>\n",
       "      <td>1.5</td>\n",
       "      <td>0.2</td>\n",
       "      <td>Iris-setosa</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5.0</td>\n",
       "      <td>3.6</td>\n",
       "      <td>1.4</td>\n",
       "      <td>0.2</td>\n",
       "      <td>Iris-setosa</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>145</th>\n",
       "      <td>6.7</td>\n",
       "      <td>3.0</td>\n",
       "      <td>5.2</td>\n",
       "      <td>2.3</td>\n",
       "      <td>Iris-virginica</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>146</th>\n",
       "      <td>6.3</td>\n",
       "      <td>2.5</td>\n",
       "      <td>5.0</td>\n",
       "      <td>1.9</td>\n",
       "      <td>Iris-virginica</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>147</th>\n",
       "      <td>6.5</td>\n",
       "      <td>3.0</td>\n",
       "      <td>5.2</td>\n",
       "      <td>2.0</td>\n",
       "      <td>Iris-virginica</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>148</th>\n",
       "      <td>6.2</td>\n",
       "      <td>3.4</td>\n",
       "      <td>5.4</td>\n",
       "      <td>2.3</td>\n",
       "      <td>Iris-virginica</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>149</th>\n",
       "      <td>5.9</td>\n",
       "      <td>3.0</td>\n",
       "      <td>5.1</td>\n",
       "      <td>1.8</td>\n",
       "      <td>Iris-virginica</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>150 rows × 6 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     SepalLengthCm  SepalWidthCm  PetalLengthCm  PetalWidthCm         Species  \\\n",
       "0              5.1           3.5            1.4           0.2     Iris-setosa   \n",
       "1              4.9           3.0            1.4           0.2     Iris-setosa   \n",
       "2              4.7           3.2            1.3           0.2     Iris-setosa   \n",
       "3              4.6           3.1            1.5           0.2     Iris-setosa   \n",
       "4              5.0           3.6            1.4           0.2     Iris-setosa   \n",
       "..             ...           ...            ...           ...             ...   \n",
       "145            6.7           3.0            5.2           2.3  Iris-virginica   \n",
       "146            6.3           2.5            5.0           1.9  Iris-virginica   \n",
       "147            6.5           3.0            5.2           2.0  Iris-virginica   \n",
       "148            6.2           3.4            5.4           2.3  Iris-virginica   \n",
       "149            5.9           3.0            5.1           1.8  Iris-virginica   \n",
       "\n",
       "     Class  \n",
       "0        0  \n",
       "1        0  \n",
       "2        0  \n",
       "3        0  \n",
       "4        0  \n",
       "..     ...  \n",
       "145      2  \n",
       "146      2  \n",
       "147      2  \n",
       "148      2  \n",
       "149      2  \n",
       "\n",
       "[150 rows x 6 columns]"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_train"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3) 檢查缺失值\n",
    "使用 numpy 所提供的函式來檢查是否有 NA 缺失值，假設有缺失值使用dropna()來移除。使用的時機在於當只有少量的缺失值適用，若遇到有大量缺失值的情況，或是本身的資料量就很少的情況下建議可以透過機器學習的方法補值來預測缺失值。\n",
    "\n",
    "```python\n",
    "# 移除缺失值\n",
    "train=train.dropna()\n",
    "```"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "checked missing data(NAN mount): 0\n"
     ]
    }
   ],
   "source": [
    "df_train = df_train.drop(labels=['Species'],axis=1) # 移除Species\n",
    "# checked missing data\n",
    "print(\"checked missing data(NAN mount):\",len(np.where(np.isnan(df_train))[0]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 4) 將輸出特徵做one-hot encoding\n",
    "如何對欄位class做one-hot encoding？\n",
    "#### class 的值有0,1,2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[1., 0., 0.],\n",
       "       [1., 0., 0.],\n",
       "       [1., 0., 0.],\n",
       "       [1., 0., 0.],\n",
       "       [1., 0., 0.]], dtype=float32)"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from tensorflow.keras.utils import to_categorical\n",
    "y = to_categorical(df_train['Class'])\n",
    "y[0:5]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 4) 切割訓練集與測試集"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "X=df_train.drop(labels=['Class'],axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "X_train , X_test , y_train , y_test = train_test_split(X,y , test_size=.3 , random_state=40)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "訓練資料:  (105, 4)\n",
      "測試資料:  (45, 4)\n"
     ]
    }
   ],
   "source": [
    "print('訓練資料: ',X_train.shape)\n",
    "print('測試資料: ',X_test.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 5) 建立網路模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.2.0\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "from tensorflow.keras.layers import Dense, Activation\n",
    "from tensorflow.keras import Sequential\n",
    "from tensorflow.keras.optimizers import Adam\n",
    "\n",
    "print(tf.__version__)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 此範例使用 Tensorflow2.0 Sequential API 搭建神經網路。\n",
    "model = Sequential()\n",
    "model.add(Dense(8, Activation('relu'), input_dim=X.shape[-1]))\n",
    "model.add(Dense(16, Activation('relu')))\n",
    "model.add(Dense(3, Activation('softmax')))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense (Dense)                (None, 8)                 40        \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 16)                144       \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 3)                 51        \n",
      "=================================================================\n",
      "Total params: 235\n",
      "Trainable params: 235\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()   # Weights = (4+1)*8+(8+1)*16+(16+1)*3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/50\n",
      "94/94 [==============================] - 0s 4ms/step - loss: 1.0565 - acc: 0.2340 - val_loss: 1.0322 - val_acc: 0.5455\n",
      "Epoch 2/50\n",
      "94/94 [==============================] - 0s 934us/step - loss: 0.9374 - acc: 0.5745 - val_loss: 0.9099 - val_acc: 0.3636\n",
      "Epoch 3/50\n",
      "94/94 [==============================] - 0s 940us/step - loss: 0.7849 - acc: 0.5213 - val_loss: 0.8373 - val_acc: 0.3636\n",
      "Epoch 4/50\n",
      "94/94 [==============================] - 0s 936us/step - loss: 0.6991 - acc: 0.6809 - val_loss: 0.7720 - val_acc: 0.3636\n",
      "Epoch 5/50\n",
      "94/94 [==============================] - 0s 931us/step - loss: 0.6222 - acc: 0.7234 - val_loss: 0.7522 - val_acc: 0.3636\n",
      "Epoch 6/50\n",
      "94/94 [==============================] - 0s 938us/step - loss: 0.5766 - acc: 0.6915 - val_loss: 0.6718 - val_acc: 0.4545\n",
      "Epoch 7/50\n",
      "94/94 [==============================] - 0s 961us/step - loss: 0.5267 - acc: 0.7553 - val_loss: 0.6262 - val_acc: 0.6364\n",
      "Epoch 8/50\n",
      "94/94 [==============================] - 0s 925us/step - loss: 0.4888 - acc: 0.8298 - val_loss: 0.5683 - val_acc: 0.6364\n",
      "Epoch 9/50\n",
      "94/94 [==============================] - 0s 929us/step - loss: 0.4578 - acc: 0.8404 - val_loss: 0.5223 - val_acc: 0.7273\n",
      "Epoch 10/50\n",
      "94/94 [==============================] - 0s 934us/step - loss: 0.4241 - acc: 0.9043 - val_loss: 0.5787 - val_acc: 0.6364\n",
      "Epoch 11/50\n",
      "94/94 [==============================] - 0s 938us/step - loss: 0.4004 - acc: 0.8936 - val_loss: 0.5700 - val_acc: 0.6364\n",
      "Epoch 12/50\n",
      "94/94 [==============================] - 0s 927us/step - loss: 0.3768 - acc: 0.8723 - val_loss: 0.4468 - val_acc: 0.8182\n",
      "Epoch 13/50\n",
      "94/94 [==============================] - 0s 931us/step - loss: 0.3619 - acc: 0.9149 - val_loss: 0.4284 - val_acc: 0.8182\n",
      "Epoch 14/50\n",
      "94/94 [==============================] - 0s 931us/step - loss: 0.3473 - acc: 0.9149 - val_loss: 0.4880 - val_acc: 0.6364\n",
      "Epoch 15/50\n",
      "94/94 [==============================] - 0s 975us/step - loss: 0.3227 - acc: 0.9362 - val_loss: 0.4811 - val_acc: 0.6364\n",
      "Epoch 16/50\n",
      "94/94 [==============================] - 0s 925us/step - loss: 0.3074 - acc: 0.8830 - val_loss: 0.5395 - val_acc: 0.6364\n",
      "Epoch 17/50\n",
      "94/94 [==============================] - 0s 930us/step - loss: 0.2932 - acc: 0.9468 - val_loss: 0.4627 - val_acc: 0.6364\n",
      "Epoch 18/50\n",
      "94/94 [==============================] - 0s 937us/step - loss: 0.2832 - acc: 0.9468 - val_loss: 0.5657 - val_acc: 0.6364\n",
      "Epoch 19/50\n",
      "94/94 [==============================] - 0s 930us/step - loss: 0.2740 - acc: 0.9362 - val_loss: 0.3420 - val_acc: 0.8182\n",
      "Epoch 20/50\n",
      "94/94 [==============================] - 0s 928us/step - loss: 0.2502 - acc: 0.9574 - val_loss: 0.4433 - val_acc: 0.6364\n",
      "Epoch 21/50\n",
      "94/94 [==============================] - 0s 943us/step - loss: 0.2458 - acc: 0.9681 - val_loss: 0.2549 - val_acc: 0.9091\n",
      "Epoch 22/50\n",
      "94/94 [==============================] - 0s 933us/step - loss: 0.2359 - acc: 0.9468 - val_loss: 0.4017 - val_acc: 0.7273\n",
      "Epoch 23/50\n",
      "94/94 [==============================] - 0s 922us/step - loss: 0.2222 - acc: 0.9574 - val_loss: 0.2641 - val_acc: 0.9091\n",
      "Epoch 24/50\n",
      "94/94 [==============================] - 0s 978us/step - loss: 0.2135 - acc: 0.9681 - val_loss: 0.3584 - val_acc: 0.7273\n",
      "Epoch 25/50\n",
      "94/94 [==============================] - 0s 935us/step - loss: 0.1997 - acc: 0.9574 - val_loss: 0.2959 - val_acc: 0.9091\n",
      "Epoch 26/50\n",
      "94/94 [==============================] - 0s 926us/step - loss: 0.2078 - acc: 0.9362 - val_loss: 0.3559 - val_acc: 0.7273\n",
      "Epoch 27/50\n",
      "94/94 [==============================] - 0s 930us/step - loss: 0.1838 - acc: 0.9574 - val_loss: 0.4747 - val_acc: 0.6364\n",
      "Epoch 28/50\n",
      "94/94 [==============================] - 0s 933us/step - loss: 0.1889 - acc: 0.9787 - val_loss: 0.3109 - val_acc: 0.8182\n",
      "Epoch 29/50\n",
      "94/94 [==============================] - 0s 935us/step - loss: 0.1759 - acc: 0.9894 - val_loss: 0.1814 - val_acc: 0.9091\n",
      "Epoch 30/50\n",
      "94/94 [==============================] - 0s 922us/step - loss: 0.1680 - acc: 0.9894 - val_loss: 0.1950 - val_acc: 0.9091\n",
      "Epoch 31/50\n",
      "94/94 [==============================] - 0s 947us/step - loss: 0.1594 - acc: 0.9681 - val_loss: 0.2751 - val_acc: 0.9091\n",
      "Epoch 32/50\n",
      "94/94 [==============================] - 0s 932us/step - loss: 0.1550 - acc: 0.9681 - val_loss: 0.2892 - val_acc: 0.8182\n",
      "Epoch 33/50\n",
      "94/94 [==============================] - 0s 939us/step - loss: 0.1523 - acc: 0.9681 - val_loss: 0.2801 - val_acc: 0.9091\n",
      "Epoch 34/50\n",
      "94/94 [==============================] - 0s 973us/step - loss: 0.1628 - acc: 0.9574 - val_loss: 0.2565 - val_acc: 0.9091\n",
      "Epoch 35/50\n",
      "94/94 [==============================] - 0s 925us/step - loss: 0.1350 - acc: 0.9787 - val_loss: 0.1339 - val_acc: 1.0000\n",
      "Epoch 36/50\n",
      "94/94 [==============================] - 0s 929us/step - loss: 0.1452 - acc: 0.9574 - val_loss: 0.2286 - val_acc: 0.9091\n",
      "Epoch 37/50\n",
      "94/94 [==============================] - 0s 931us/step - loss: 0.1358 - acc: 0.9681 - val_loss: 0.2258 - val_acc: 0.9091\n",
      "Epoch 38/50\n",
      "94/94 [==============================] - 0s 930us/step - loss: 0.1346 - acc: 0.9574 - val_loss: 0.1459 - val_acc: 0.9091\n",
      "Epoch 39/50\n",
      "94/94 [==============================] - 0s 929us/step - loss: 0.1225 - acc: 0.9681 - val_loss: 0.2785 - val_acc: 0.9091\n",
      "Epoch 40/50\n",
      "94/94 [==============================] - 0s 927us/step - loss: 0.1272 - acc: 0.9681 - val_loss: 0.4188 - val_acc: 0.7273\n",
      "Epoch 41/50\n",
      "94/94 [==============================] - 0s 932us/step - loss: 0.1247 - acc: 0.9787 - val_loss: 0.2206 - val_acc: 0.9091\n",
      "Epoch 42/50\n",
      "94/94 [==============================] - 0s 927us/step - loss: 0.1217 - acc: 0.9681 - val_loss: 0.1980 - val_acc: 0.9091\n",
      "Epoch 43/50\n",
      "94/94 [==============================] - 0s 935us/step - loss: 0.1192 - acc: 0.9787 - val_loss: 0.2309 - val_acc: 0.9091\n",
      "Epoch 44/50\n",
      "94/94 [==============================] - 0s 968us/step - loss: 0.1182 - acc: 0.9681 - val_loss: 0.2750 - val_acc: 0.9091\n",
      "Epoch 45/50\n",
      "94/94 [==============================] - 0s 927us/step - loss: 0.1197 - acc: 0.9787 - val_loss: 0.3064 - val_acc: 0.8182\n",
      "Epoch 46/50\n",
      "94/94 [==============================] - 0s 926us/step - loss: 0.1250 - acc: 0.9681 - val_loss: 0.3159 - val_acc: 0.8182\n",
      "Epoch 47/50\n",
      "94/94 [==============================] - 0s 929us/step - loss: 0.1269 - acc: 0.9681 - val_loss: 0.1997 - val_acc: 0.9091\n",
      "Epoch 48/50\n",
      "94/94 [==============================] - 0s 937us/step - loss: 0.1186 - acc: 0.9681 - val_loss: 0.3418 - val_acc: 0.8182\n",
      "Epoch 49/50\n",
      "94/94 [==============================] - 0s 930us/step - loss: 0.1288 - acc: 0.9468 - val_loss: 0.1827 - val_acc: 0.9091\n",
      "Epoch 50/50\n",
      "94/94 [==============================] - 0s 923us/step - loss: 0.0993 - acc: 0.9681 - val_loss: 0.3185 - val_acc: 0.8182\n"
     ]
    }
   ],
   "source": [
    "# 編譯模型\n",
    "optim = Adam(lr=0.001)\n",
    "model.compile(loss='categorical_crossentropy',\n",
    "              optimizer=optim,\n",
    "              metrics=['acc'])\n",
    "\n",
    "batch_size=1\n",
    "epochs = 50\n",
    "\n",
    "# 訓練模型\n",
    "history = model.fit(X_train, y_train,\n",
    "                    batch_size=batch_size,\n",
    "                    epochs=epochs,\n",
    "                    verbose=1,\n",
    "                    shuffle=True,\n",
    "                    validation_split=0.1)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 6) 觀察訓練結果"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "history_dict = history.history\n",
    "history_dict.keys()\n",
    "acc = history_dict['acc']\n",
    "val_acc = history_dict['val_acc']\n",
    "loss = history_dict['loss']\n",
    "val_loss = history_dict['val_loss']\n",
    "epochs_ = range(1,len(acc)+1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(epochs_ , loss , label = 'training loss')\n",
    "plt.plot(epochs_ , val_loss , label = 'val los')\n",
    "plt.title('training and val loss')\n",
    "plt.xlabel('epochs')\n",
    "plt.ylabel('loss')\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.clf()\n",
    "plt.plot(epochs_ , acc , label='train accuracy')\n",
    "plt.plot(epochs_ , val_acc , label = 'val accuracy')\n",
    "plt.title('train and val acc')\n",
    "plt.xlabel('epochs')\n",
    "plt.ylabel('acc')\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 測試"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.9666666666666667\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import accuracy_score\n",
    "label=np.argmax(y,axis=1)\n",
    "pred =  np.argmax(model.predict(X), axis=1)\n",
    "print(accuracy_score(label, pred))"
   ]
  },
  {
   "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}