ipynb/dama-keras-demo-handwritten-digit-recognition.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# TFUG珠海 TensorFlow-all-around 活动 code-lab\n",
    "[关于 TFUG珠海 . TFUG Livin ZhuHai Life;-)](http://zh.tfug.world/about/)\n",
    "\n",
    "活动: [TensorFlow All\\-Around 珠海 \\| Meetup](https://www.meetup.com/TensorFlow-User-Group-Zhuhai/events/264053619/)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 目标\n",
    "通过最基本入门任务, 来展示基本 ML 处理过程:\n",
    "\n",
    "- 数据整理\n",
    "- 模型定义\n",
    "- 训练模型\n",
    "- 检验模型\n",
    "\n",
    "以便激发大家学习兴趣, 合理理解 AI 实现, 正确结合业务推动项目落地."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 纲要\n",
    "> 毕竟大妈也嗯哼过几年编程, 所以,在此 笔记之外有很多准备, 并没逐一展示, 特此提示\n",
    "\n",
    "- 用 Python 3\n",
    "- 基于 miniconda3 进行运行时版本环境构造\n",
    "- 配置使用 aliyun 的 PyPi 仓库加速模块安装\n",
    "- 先通过脚本原型完成所有代码的探索, 再整理为 ipynb 思考过程记述\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Keras 用 TensorFlow 识别手写数字\n",
    "\n",
    "> 基于标准 MNSIT 数据集\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# fix a random seed for reproducibility\n",
    "# 注意,这里调节 np 的随机核, 必须在加载 keras 系列模块前先导入, 否则....\n",
    "import numpy as np\n",
    "#np.random.seed(123)\n",
    "np.random.seed(9)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "ename": "ModuleNotFoundError",
     "evalue": "No module named 'keras'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mModuleNotFoundError\u001b[0m                       Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-2-86ea1171a113>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0;31m# 加载一系列依赖模块\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m \u001b[0;32mimport\u001b[0m \u001b[0mkeras\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      3\u001b[0m \u001b[0;32mfrom\u001b[0m \u001b[0mkeras\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdatasets\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mmnist\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m \u001b[0;32mfrom\u001b[0m \u001b[0mkeras\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmodels\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mSequential\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      5\u001b[0m \u001b[0;32mfrom\u001b[0m \u001b[0mkeras\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmodels\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0msave_model\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mload_model\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mModuleNotFoundError\u001b[0m: No module named 'keras'"
     ]
    }
   ],
   "source": [
    "# 加载一系列依赖模块\n",
    "import keras\n",
    "from keras.datasets import mnist\n",
    "from keras.models import Sequential\n",
    "from keras.models import save_model,load_model\n",
    "from keras.layers import Conv2D,MaxPool2D,Dense,Flatten, Activation\n",
    "from keras.layers.core import Dense, Activation, Dropout\n",
    "from keras.callbacks import TensorBoard\n",
    "\n",
    "from keras.utils import plot_model\n",
    "from keras.utils import np_utils\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 各种基本参数\n",
    "# 训练次数...为节约时间, 可以调低\n",
    "#nb_epoch = 25\n",
    "nb_epoch = 12\n",
    "\n",
    "num_classes = 10\n",
    "batch_size = 128\n",
    "train_size = 60000\n",
    "test_size = 10000\n",
    "\n",
    "# 横向长度...图片压缩目标\n",
    "v_length = 784\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> 技巧:\n",
    "\n",
    "官方各种代码都是通过 `mnist.load_data()` 试图自动下载:\n",
    "https://s3.amazonaws.com/img-datasets/mnist.npz\n",
    "\n",
    "可惜是的, 这儿距离中国太远, 下载速度非常感人, 只能用其它技术手段先下载到本地,\n",
    "以便 Jupyter 可以从本地加载,\n",
    "\n",
    "大家在运行时, 请:\n",
    "\n",
    "- 先下载:[mnist.npz](http://ipynb.zoomquiet.top/data/mnist.npz)\n",
    "- 放到 jupyter notebook 运行目录中\n",
    "\n",
    "- 和 `tfug-zh-dama-keras-demo-handwritten-digit-recognition.ipynb` 同一层目录\n",
    "\n",
    "即可;-)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 基于事先下载的 .npz 数据集,改写官方数据加载函式,变成自己的本地加载函式\n",
    "def locload_data(path='mnist.npz'):\n",
    "    f = np.load(path)\n",
    "    x_train, y_train = f['x_train'], f['y_train']\n",
    "    x_test, y_test = f['x_test'], f['y_test']\n",
    "    f.close()\n",
    "    return (x_train, y_train), (x_test, y_test)\n",
    "\n",
    "# 将数据读入\n",
    "(x_train, y_train), (x_test, y_test) = locload_data()\n",
    "# split the mnist data into train and test\n",
    "#(trainData, trainLabels), (testData, testLabels) = mnist.load_data()\n",
    "(trainData, trainLabels), (testData, testLabels) = locload_data()\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 检验加载数据维度"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[INFO] train data shape: (60000, 28, 28)\n",
      "[INFO] test data shape: (10000, 28, 28)\n",
      "[INFO] train samples: 60000\n",
      "[INFO] test samples: 10000\n"
     ]
    }
   ],
   "source": [
    "#print(\"训练数据集:\\n\", x_train.shape,y_train.shape,\"\\n测试数据集:\\n\",x_test.shape,y_test.shape)\n",
    "print(\"[INFO] train data shape: {}\".format(trainData.shape))\n",
    "print(\"[INFO] test data shape: {}\".format(testData.shape))\n",
    "print(\"[INFO] train samples: {}\".format(trainData.shape[0]))\n",
    "print(\"[INFO] test samples: {}\".format(testData.shape[0]))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "6 0 4 1\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# 训练数据集抽查, plot 4 images as gray scale\n",
    "plt.subplot(221)\n",
    "print(y_train[4546],y_train[1],y_train[2],y_train[3])\n",
    "plt.imshow(x_train[4546], cmap=plt.get_cmap('gray'))\n",
    "plt.subplot(222)\n",
    "plt.imshow(x_train[1], cmap=plt.get_cmap('gray'))\n",
    "plt.subplot(223)\n",
    "plt.imshow(x_train[2], cmap=plt.get_cmap('gray'))\n",
    "plt.subplot(224)\n",
    "plt.imshow(x_train[3], cmap=plt.get_cmap('gray'))\n",
    "# show the plot\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##  格式化数据集"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[INFO] train data shape: (60000, 784)\n",
      "[INFO] test data shape: (10000, 784)\n",
      "[INFO] train samples: 60000\n",
      "[INFO] test samples: 10000\n"
     ]
    }
   ],
   "source": [
    "# reshape the dataset\n",
    "trainData = trainData.reshape(train_size, v_length)\n",
    "testData = testData.reshape(test_size, v_length)\n",
    "trainData = trainData.astype(\"float32\")\n",
    "testData = testData.astype(\"float32\")\n",
    "trainData /= 255\n",
    "testData /= 255\n",
    "\n",
    "print(\"[INFO] train data shape: {}\".format(trainData.shape))\n",
    "print(\"[INFO] test data shape: {}\".format(testData.shape))\n",
    "print(\"[INFO] train samples: {}\".format(trainData.shape[0]))\n",
    "print(\"[INFO] test samples: {}\".format(testData.shape[0]))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "# convert class vectors to binary class matrices --> one-hot encoding\n",
    "mTrainLabels = np_utils.to_categorical(trainLabels, num_classes)\n",
    "mTestLabels = np_utils.to_categorical(testLabels, num_classes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 构造模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /Users/zoomq/miniconda3/envs/keras/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py:148: The name tf.placeholder_with_default is deprecated. Please use tf.compat.v1.placeholder_with_default instead.\n",
      "\n",
      "WARNING:tensorflow:From /Users/zoomq/miniconda3/envs/keras/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py:3733: calling dropout (from tensorflow.python.ops.nn_ops) with keep_prob is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use `rate` instead of `keep_prob`. Rate should be set to `rate = 1 - keep_prob`.\n",
      "Model: \"sequential_2\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_2 (Dense)              (None, 512)               401920    \n",
      "_________________________________________________________________\n",
      "activation_2 (Activation)    (None, 512)               0         \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 512)               0         \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 256)               131328    \n",
      "_________________________________________________________________\n",
      "activation_3 (Activation)    (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dropout_2 (Dropout)          (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 10)                2570      \n",
      "_________________________________________________________________\n",
      "activation_4 (Activation)    (None, 10)                0         \n",
      "=================================================================\n",
      "Total params: 535,818\n",
      "Trainable params: 535,818\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<IPython.core.display.Image object>"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# create the model\n",
    "model = Sequential()\n",
    "model.add(Dense(512, input_shape=(784,)))\n",
    "model.add(Activation(\"relu\"))\n",
    "model.add(Dropout(0.2))\n",
    "model.add(Dense(256))\n",
    "model.add(Activation(\"relu\"))\n",
    "model.add(Dropout(0.2))\n",
    "model.add(Dense(num_classes))\n",
    "model.add(Activation(\"softmax\"))\n",
    "\n",
    "# summarize the model\n",
    "model.summary()\n",
    "\n",
    "# 保存模型(拓扑图)\n",
    "plot_model(model, to_file='mlp.png')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 训练并检验模式"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /Users/zoomq/miniconda3/envs/keras/lib/python3.7/site-packages/keras/optimizers.py:793: The name tf.train.Optimizer is deprecated. Please use tf.compat.v1.train.Optimizer instead.\n",
      "\n",
      "WARNING:tensorflow:From /Users/zoomq/miniconda3/envs/keras/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py:3576: The name tf.log is deprecated. Please use tf.math.log instead.\n",
      "\n",
      "WARNING:tensorflow:From /Users/zoomq/miniconda3/envs/keras/lib/python3.7/site-packages/tensorflow/python/ops/math_grad.py:1250: add_dispatch_support.<locals>.wrapper (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use tf.where in 2.0, which has the same broadcast rule as np.where\n",
      "Train on 60000 samples, validate on 10000 samples\n",
      "Epoch 1/12\n",
      " - 6s - loss: 0.2688 - acc: 0.9200 - val_loss: 0.1058 - val_acc: 0.9669\n",
      "Epoch 2/12\n",
      " - 5s - loss: 0.1101 - acc: 0.9668 - val_loss: 0.0921 - val_acc: 0.9717\n",
      "Epoch 3/12\n",
      " - 5s - loss: 0.0771 - acc: 0.9760 - val_loss: 0.0697 - val_acc: 0.9789\n",
      "Epoch 4/12\n",
      " - 7s - loss: 0.0604 - acc: 0.9803 - val_loss: 0.0717 - val_acc: 0.9777\n",
      "Epoch 5/12\n",
      " - 6s - loss: 0.0480 - acc: 0.9843 - val_loss: 0.0664 - val_acc: 0.9804\n",
      "Epoch 6/12\n",
      " - 5s - loss: 0.0435 - acc: 0.9858 - val_loss: 0.0703 - val_acc: 0.9808\n",
      "Epoch 7/12\n",
      " - 6s - loss: 0.0361 - acc: 0.9877 - val_loss: 0.0711 - val_acc: 0.9799\n",
      "Epoch 8/12\n",
      " - 6s - loss: 0.0313 - acc: 0.9892 - val_loss: 0.0675 - val_acc: 0.9791\n",
      "Epoch 9/12\n",
      " - 5s - loss: 0.0272 - acc: 0.9909 - val_loss: 0.0558 - val_acc: 0.9852\n",
      "Epoch 10/12\n",
      " - 6s - loss: 0.0266 - acc: 0.9911 - val_loss: 0.0705 - val_acc: 0.9807\n",
      "Epoch 11/12\n",
      " - 5s - loss: 0.0228 - acc: 0.9919 - val_loss: 0.0741 - val_acc: 0.9805\n",
      "Epoch 12/12\n",
      " - 5s - loss: 0.0234 - acc: 0.9926 - val_loss: 0.0674 - val_acc: 0.9822\n"
     ]
    }
   ],
   "source": [
    "# compile the model\n",
    "model.compile(loss=\"categorical_crossentropy\", optimizer=\"adam\", metrics=[\"accuracy\"])\n",
    "# fit the model\n",
    "history = model.fit(trainData, mTrainLabels\n",
    "                    , validation_data = (testData, mTestLabels)\n",
    "                    , batch_size = batch_size\n",
    "                    #, nb_epoch=nb_epoch\n",
    "                    , epochs = nb_epoch\n",
    "                    , verbose = 2)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> Epoch 3/12\n",
    " - 5s - loss: 0.0771 - acc: 0.9760 - val_loss: 0.0697 - val_acc: 0.9789\n",
    " \n",
    " 意思是:\n",
    " \n",
    " - 第三轮 Epoch\n",
    " - 用 5秒完成\n",
    " - 获得精确度 97.6%\n",
    " - 错误率: 0.2% 左右"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dict_keys(['val_loss', 'val_acc', 'loss', 'acc'])\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[INFO] test score - 0.06735334231600937\n",
      "[INFO] test accuracy - 0.9822\n"
     ]
    }
   ],
   "source": [
    "# print the history keys\n",
    "print(history.history.keys())\n",
    "\n",
    "# evaluate the model\n",
    "scores = model.evaluate(testData, mTestLabels, verbose=0)\n",
    "\n",
    "# history plot for accuracy\n",
    "plt.plot(history.history[\"acc\"])\n",
    "plt.plot(history.history[\"val_acc\"])\n",
    "plt.title(\"Model Accuracy\")\n",
    "plt.xlabel(\"Epoch\")\n",
    "plt.ylabel(\"Accuracy\")\n",
    "plt.legend([\"train\", \"test\"], loc=\"upper left\")\n",
    "plt.show()\n",
    "\n",
    "# history plot for accuracy\n",
    "plt.plot(history.history[\"loss\"])\n",
    "plt.plot(history.history[\"val_loss\"])\n",
    "plt.title(\"Model Loss\")\n",
    "plt.xlabel(\"Epoch\")\n",
    "plt.ylabel(\"Loss\")\n",
    "plt.legend([\"train\", \"test\"], loc=\"upper left\")\n",
    "plt.show()\n",
    "\n",
    "# print the results\n",
    "print(\"[INFO] test score - {}\".format(scores[0]))\n",
    "print(\"[INFO] test accuracy - {}\".format(scores[1]))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> 模型最关键效能指数\n",
    "\n",
    "- Accuracy ~ 精准度, 仅仅通过12次训练, 就获得 98.35% 的精确度\n",
    "- Loss ~ 错误率, 而对测试集的错误,也稳定在 0.6% 左右"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 检验模型\n",
    "> 尝试对训练出来的成果进行可视化检验"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(10000, 784)\n"
     ]
    }
   ],
   "source": [
    "# 测试目标数据集体量\n",
    "print(testData.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "随机位置: [1996, 2052, 2916, 9987, 6429, 5431, 3588, 7588, 9953, 6433]\n"
     ]
    }
   ],
   "source": [
    "# 尝试用随机模块来生成随机数, 从目标数据集中指定图片来检验\n",
    "\n",
    "import random\n",
    "selected_index = random.sample(range(len(testData)), k=10)\n",
    "print(\"随机位置:\",selected_index)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import random\n",
    "selected_index = random.sample(range(len(testData)), k=10)\n",
    "# grab some test images from the test data\n",
    "test_images = testData[selected_index[0]:selected_index[0]+4]\n",
    "\n",
    "# reshape the test images to standard 28x28 format\n",
    "test_images = test_images.reshape(test_images.shape[0], 28, 28)\n",
    "#print(\"[INFO] test images shape - {}\".format(test_images.shape))\n",
    "# loop over each of the test images\n",
    "for i, test_image in enumerate(test_images, start=1):\n",
    "    # grab a copy of test image for viewing\n",
    "    org_image = test_image\n",
    "\n",
    "    # reshape the test image to [1x784] format so that our model understands\n",
    "    test_image = test_image.reshape(1,784)\n",
    "    \n",
    "    # make prediction on test image using our trained model\n",
    "    prediction = model.predict_classes(test_image, verbose=0)\n",
    "\n",
    "    # display the prediction and image\n",
    "    #print(\"[INFO] I think the digit is - {}\".format(prediction[0]))\n",
    "    plt.subplot(220+i)\n",
    "    plt.axis('off')\n",
    "    plt.title(\"guess~> %s\"%prediction[0], color='blue')\n",
    "    plt.imshow(org_image, cmap=plt.get_cmap('gray'))\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> 结果出来了:\n",
    "\n",
    "随机选择的4张照片, 由模型猜测手写数字是几?\n",
    "基本都对 ;-)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# refer.\n",
    "\n",
    "- [面向机器学习初学者的 MNIST 初级教程](http://tensorfly.cn/tfdoc/tutorials/mnist_beginners.html)\n",
    "- [Handwritten Digit Recognition using Deep Learning, Keras and Python – Gogul Ilango](https://gogul09.github.io/software/digits-recognition-mlp)\n",
    "    + [Dwijraj/Hand-Written-Digit-Recognition: This is a repository with various classifier written in jupyter notebook that works on hand written digits images.](https://github.com/Dwijraj/Hand-Written-Digit-Recognition)\n",
    "    + [keras/mnist_cnn.py at master · keras-team/keras](https://github.com/keras-team/keras/blob/master/examples/mnist_cnn.py)\n",
    "    + [kerasのmnistのサンプルを読んでみる - Qiita](https://qiita.com/ash8h/items/29e24fc617b832fba136)\n",
    "- [基于Keras+CNN的MNIST数据集手写数字分类 - 简书](https://www.jianshu.com/p/3a8b310227e6)\n",
    "    + [Keras_mnist学习 - 简书](https://www.jianshu.com/p/769fb8d6cab0)\n",
    "    + [keras实现mnist数据集手写数字识别 - 胡卫雄 - 博客园](https://www.cnblogs.com/ncuhwxiong/p/9774515.html)\n",
    "    + [深度学习：基于keras的mnist手写数字识别 | AI柠檬](https://blog.ailemon.me/2018/05/11/deep-learning-mnist-handwritten-digit-recognition-by-keras/)\n",
    "- ...\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## env.\n",
    "> 嗯哼一下, 大妈演示时用的环境, 大家可以根据自己的情况配置合适的...\n",
    "\n",
    "```\n",
    "\n",
    "༄  screenfetch\n",
    "\n",
    "                 -/+:.          zoomq@ZQ160626rMBP\n",
    "                :++++.          OS: 64bit Mac OS X 10.12.6 16G2128\n",
    "               /+++/.           Kernel: x86_64 Darwin 16.7.0\n",
    "       .:-::- .+/:-``.::-       Uptime: 1d 14h 51m\n",
    "    .:/++++++/::::/++++++/:`    Packages: 235\n",
    "  .:///////////////////////:`   Shell: bash\n",
    "  ////////////////////////`     Resolution: 2560x1600\n",
    " -+++++++++++++++++++++++`      DE: Aqua\n",
    " /++++++++++++++++++++++/       WM: Quartz Compositor\n",
    " /sssssssssssssssssssssss.      WM Theme: Blue\n",
    " :ssssssssssssssssssssssss-     CPU: Intel Core i7-5557U @ 3.10GHz\n",
    "  osssssssssssssssssssssssso/`  GPU: Intel Iris Graphics 6100\n",
    "  `syyyyyyyyyyyyyyyyyyyyyyyy+`  RAM: 8996MiB / 16384MiB\n",
    "   `ossssssssssssssssssssss/\n",
    "     :ooooooooooooooooooo+.\n",
    "      `:+oo+/:-..-:/+o+/-\n",
    "\n",
    "\n",
    "\n",
    "༄  conda info\n",
    "\n",
    "     active environment : keras\n",
    "    active env location : /Users/zoomq/miniconda3/envs/keras\n",
    "            shell level : 1\n",
    "       user config file : /Users/zoomq/.condarc\n",
    " populated config files : /Users/zoomq/.condarc\n",
    "          conda version : 4.6.14\n",
    "    conda-build version : not installed\n",
    "         python version : 3.7.3.final.0\n",
    "       base environment : /Users/zoomq/miniconda3  (writable)\n",
    "           channel URLs : https://repo.anaconda.com/pkgs/main/osx-64\n",
    "                          https://repo.anaconda.com/pkgs/main/noarch\n",
    "                          https://repo.anaconda.com/pkgs/free/osx-64\n",
    "                          https://repo.anaconda.com/pkgs/free/noarch\n",
    "                          https://repo.anaconda.com/pkgs/r/osx-64\n",
    "                          https://repo.anaconda.com/pkgs/r/noarch\n",
    "          package cache : /Users/zoomq/miniconda3/pkgs\n",
    "                          /Users/zoomq/.conda/pkgs\n",
    "       envs directories : /Users/zoomq/miniconda3/envs\n",
    "                          /Users/zoomq/.conda/envs\n",
    "               platform : osx-64\n",
    "             user-agent : conda/4.6.14 requests/2.21.0 CPython/3.7.3 Darwin/16.7.0 OSX/10.12.6\n",
    "                UID:GID : 501:20\n",
    "             netrc file : /Users/zoomq/.netrc\n",
    "           offline mode : False\n",
    "           \n",
    "\n",
    "\n",
    "༄  conda list\n",
    "# packages in environment at /Users/zoomq/miniconda3/envs/keras:\n",
    "#\n",
    "# Name                    Version                   Build  Channel\n",
    "absl-py                   0.8.0                    pypi_0    pypi\n",
    "appnope                   0.1.0                    pypi_0    pypi\n",
    "asn1crypto                0.24.0                   py37_0\n",
    "astor                     0.8.0                    pypi_0    pypi\n",
    "attrs                     19.1.0                   pypi_0    pypi\n",
    "backcall                  0.1.0                    pypi_0    pypi\n",
    "bleach                    3.1.0                    pypi_0    pypi\n",
    "ca-certificates           2019.5.15                     1\n",
    "certifi                   2019.6.16                py37_1\n",
    "cffi                      1.12.3           py37hb5b8e2f_0\n",
    "chardet                   3.0.4                 py37_1003\n",
    "cryptography              2.7              py37ha12b0ac_0\n",
    "cycler                    0.10.0                   pypi_0    pypi\n",
    "decorator                 4.4.0                    pypi_0    pypi\n",
    "defusedxml                0.6.0                    pypi_0    pypi\n",
    "entrypoints               0.3                      pypi_0    pypi\n",
    "gast                      0.2.2                    pypi_0    pypi\n",
    "google-pasta              0.1.7                    pypi_0    pypi\n",
    "grpcio                    1.23.0                   pypi_0    pypi\n",
    "h5py                      2.9.0                    pypi_0    pypi\n",
    "idna                      2.8                      py37_0\n",
    "ipykernel                 5.1.2                    pypi_0    pypi\n",
    "ipython                   7.8.0                    pypi_0    pypi\n",
    "ipython-genutils          0.2.0                    pypi_0    pypi\n",
    "ipywidgets                7.5.1                    pypi_0    pypi\n",
    "jedi                      0.15.1                   pypi_0    pypi\n",
    "jinja2                    2.10.1                   pypi_0    pypi\n",
    "jsonschema                3.0.2                    pypi_0    pypi\n",
    "jupyter                   1.0.0                    pypi_0    pypi\n",
    "jupyter-client            5.3.1                    pypi_0    pypi\n",
    "jupyter-console           6.0.0                    pypi_0    pypi\n",
    "jupyter-core              4.5.0                    pypi_0    pypi\n",
    "keras                     2.2.5                    pypi_0    pypi\n",
    "keras-applications        1.0.8                    pypi_0    pypi\n",
    "keras-preprocessing       1.1.0                    pypi_0    pypi\n",
    "kiwisolver                1.1.0                    pypi_0    pypi\n",
    "libcxx                    4.0.1                hcfea43d_1\n",
    "libcxxabi                 4.0.1                hcfea43d_1\n",
    "libedit                   3.1.20181209         hb402a30_0\n",
    "libffi                    3.2.1                h475c297_4\n",
    "markdown                  3.1.1                    pypi_0    pypi\n",
    "markupsafe                1.1.1                    pypi_0    pypi\n",
    "matplotlib                3.1.1                    pypi_0    pypi\n",
    "mistune                   0.8.4                    pypi_0    pypi\n",
    "nbconvert                 5.6.0                    pypi_0    pypi\n",
    "nbformat                  4.4.0                    pypi_0    pypi\n",
    "ncurses                   6.1                  h0a44026_1\n",
    "notebook                  6.0.1                    pypi_0    pypi\n",
    "numpy                     1.17.1                   pypi_0    pypi\n",
    "openssl                   1.1.1c               h1de35cc_1\n",
    "pandocfilters             1.4.2                    pypi_0    pypi\n",
    "parso                     0.5.1                    pypi_0    pypi\n",
    "pexpect                   4.7.0                    pypi_0    pypi\n",
    "pickleshare               0.7.5                    pypi_0    pypi\n",
    "pip                       19.2.2                   py37_0\n",
    "prometheus-client         0.7.1                    pypi_0    pypi\n",
    "prompt-toolkit            2.0.9                    pypi_0    pypi\n",
    "protobuf                  3.9.1                    pypi_0    pypi\n",
    "ptyprocess                0.6.0                    pypi_0    pypi\n",
    "pycosat                   0.6.3            py37h1de35cc_0\n",
    "pycparser                 2.19                     py37_0\n",
    "pydot                     1.4.1                    pypi_0    pypi\n",
    "pygments                  2.4.2                    pypi_0    pypi\n",
    "pyopenssl                 19.0.0                   py37_0\n",
    "pyparsing                 2.4.2                    pypi_0    pypi\n",
    "pyrsistent                0.15.4                   pypi_0    pypi\n",
    "pysocks                   1.7.0                    py37_0\n",
    "python                    3.7.4                h359304d_1\n",
    "python-dateutil           2.8.0                    pypi_0    pypi\n",
    "python.app                2                        py37_9\n",
    "pyyaml                    5.1.2                    pypi_0    pypi\n",
    "pyzmq                     18.1.0                   pypi_0    pypi\n",
    "qtconsole                 4.5.5                    pypi_0    pypi\n",
    "readline                  7.0                  h1de35cc_5\n",
    "requests                  2.22.0                   py37_0\n",
    "ruamel_yaml               0.15.46          py37h1de35cc_0\n",
    "scipy                     1.3.1                    pypi_0    pypi\n",
    "send2trash                1.5.0                    pypi_0    pypi\n",
    "setuptools                41.0.1                   py37_0\n",
    "six                       1.12.0                   py37_0\n",
    "sqlite                    3.29.0               ha441bb4_0\n",
    "tensorboard               1.14.0                   pypi_0    pypi\n",
    "tensorflow                1.14.0                   pypi_0    pypi\n",
    "tensorflow-estimator      1.14.0                   pypi_0    pypi\n",
    "termcolor                 1.1.0                    pypi_0    pypi\n",
    "terminado                 0.8.2                    pypi_0    pypi\n",
    "testpath                  0.4.2                    pypi_0    pypi\n",
    "tk                        8.6.8                ha441bb4_0\n",
    "tornado                   6.0.3                    pypi_0    pypi\n",
    "traitlets                 4.3.2                    pypi_0    pypi\n",
    "urllib3                   1.24.2                   py37_0\n",
    "wcwidth                   0.1.7                    pypi_0    pypi\n",
    "webencodings              0.5.1                    pypi_0    pypi\n",
    "werkzeug                  0.15.6                   pypi_0    pypi\n",
    "wheel                     0.33.4                   py37_0\n",
    "widgetsnbextension        3.5.1                    pypi_0    pypi\n",
    "wrapt                     1.11.2                   pypi_0    pypi\n",
    "xz                        5.2.4                h1de35cc_4\n",
    "yaml                      0.1.7                hc338f04_2\n",
    "zlib                      1.2.11               h1de35cc_3\n",
    "\n",
    "```\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "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": "3.6.3"
  },
  "toc": {
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": "block",
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}