ch06-与学习相关的技巧.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 参数更新方法比较\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch/blob/master/ch06/optimizer_compare_naive.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Figure size 640x480 with 4 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# coding: utf-8\n",
    "import sys, os\n",
    "sys.path.append(os.pardir)  # 为导入父目录中的文件而进行的设置\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from collections import OrderedDict\n",
    "from common.optimizer import *\n",
    "\n",
    "\n",
    "def f(x, y):\n",
    "    return x**2 / 20.0 + y**2\n",
    "\n",
    "\n",
    "def df(x, y):\n",
    "    return x / 10.0, 2.0*y\n",
    "\n",
    "init_pos = (-7.0, 2.0)\n",
    "params = {}\n",
    "params['x'], params['y'] = init_pos[0], init_pos[1]\n",
    "grads = {}\n",
    "grads['x'], grads['y'] = 0, 0\n",
    "\n",
    "\n",
    "optimizers = OrderedDict()\n",
    "optimizers[\"SGD\"] = SGD(lr=0.95)\n",
    "optimizers[\"Momentum\"] = Momentum(lr=0.1)\n",
    "optimizers[\"AdaGrad\"] = AdaGrad(lr=1.5)\n",
    "optimizers[\"Adam\"] = Adam(lr=0.3)\n",
    "\n",
    "idx = 1\n",
    "\n",
    "for key in optimizers:\n",
    "    optimizer = optimizers[key]\n",
    "    x_history = []\n",
    "    y_history = []\n",
    "    params['x'], params['y'] = init_pos[0], init_pos[1]\n",
    "    \n",
    "    for i in range(30):\n",
    "        x_history.append(params['x'])\n",
    "        y_history.append(params['y'])\n",
    "        \n",
    "        grads['x'], grads['y'] = df(params['x'], params['y'])\n",
    "        optimizer.update(params, grads)\n",
    "    \n",
    "\n",
    "    x = np.arange(-10, 10, 0.01)\n",
    "    y = np.arange(-5, 5, 0.01)\n",
    "    \n",
    "    X, Y = np.meshgrid(x, y) \n",
    "    Z = f(X, Y)\n",
    "    \n",
    "    # for simple contour line  \n",
    "    mask = Z > 7\n",
    "    Z[mask] = 0\n",
    "    \n",
    "    # plot \n",
    "    plt.subplot(2, 2, idx)\n",
    "    idx += 1\n",
    "    plt.plot(x_history, y_history, 'o-', color=\"red\")\n",
    "    plt.contour(X, Y, Z)\n",
    "    plt.ylim(-10, 10)\n",
    "    plt.xlim(-10, 10)\n",
    "    plt.plot(0, 0, '+')\n",
    "    #colorbar()\n",
    "    #spring()\n",
    "    plt.title(key)\n",
    "    plt.xlabel(\"x\")\n",
    "    plt.ylabel(\"y\")\n",
    "    \n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# mnist数据集更新比较\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch/blob/master/ch06/optimizer_compare_mnist.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===========iteration:0===========\n",
      "SGD:2.334071385208264\n",
      "Momentum:2.3691778464717386\n",
      "AdaGrad:3.169372953589367\n",
      "Adam:2.2011890132976264\n",
      "===========iteration:100===========\n",
      "SGD:1.4248168695192494\n",
      "Momentum:0.31381358248406654\n",
      "AdaGrad:0.11412993203654723\n",
      "Adam:0.24216233513729085\n",
      "===========iteration:200===========\n",
      "SGD:0.5983810929704139\n",
      "Momentum:0.4193872941690091\n",
      "AdaGrad:0.1819726014208852\n",
      "Adam:0.23314478640111336\n",
      "===========iteration:300===========\n",
      "SGD:0.4424090664903332\n",
      "Momentum:0.19427584746049806\n",
      "AdaGrad:0.0668455189784197\n",
      "Adam:0.10032031222276981\n",
      "===========iteration:400===========\n",
      "SGD:0.4095452120324918\n",
      "Momentum:0.20159765853772652\n",
      "AdaGrad:0.06202351203384572\n",
      "Adam:0.1335064808595999\n",
      "===========iteration:500===========\n",
      "SGD:0.3751587932274521\n",
      "Momentum:0.14034947157815764\n",
      "AdaGrad:0.07750560336402089\n",
      "Adam:0.08128458684738674\n",
      "===========iteration:600===========\n",
      "SGD:0.28754545224331907\n",
      "Momentum:0.10269764729117203\n",
      "AdaGrad:0.05968246737943076\n",
      "Adam:0.11412910828318369\n",
      "===========iteration:700===========\n",
      "SGD:0.33214031821972606\n",
      "Momentum:0.13491529516697412\n",
      "AdaGrad:0.0588797208625245\n",
      "Adam:0.05126142892464845\n",
      "===========iteration:800===========\n",
      "SGD:0.32319405248015054\n",
      "Momentum:0.13806041191120014\n",
      "AdaGrad:0.08840350671980853\n",
      "Adam:0.13080072930087006\n",
      "===========iteration:900===========\n",
      "SGD:0.40107043562552896\n",
      "Momentum:0.14138556433131536\n",
      "AdaGrad:0.09492224633186216\n",
      "Adam:0.06764230778173633\n",
      "===========iteration:1000===========\n",
      "SGD:0.18580767074122262\n",
      "Momentum:0.03853898309115796\n",
      "AdaGrad:0.016024694197810854\n",
      "Adam:0.05224139591061043\n",
      "===========iteration:1100===========\n",
      "SGD:0.3326492643316653\n",
      "Momentum:0.11638410708276503\n",
      "AdaGrad:0.08110800602065249\n",
      "Adam:0.1160034329549963\n",
      "===========iteration:1200===========\n",
      "SGD:0.20616036954937111\n",
      "Momentum:0.1349364085703713\n",
      "AdaGrad:0.07067539519109674\n",
      "Adam:0.10879903294905693\n",
      "===========iteration:1300===========\n",
      "SGD:0.17700083570103864\n",
      "Momentum:0.0529521283267473\n",
      "AdaGrad:0.02182168659676837\n",
      "Adam:0.04738986677862477\n",
      "===========iteration:1400===========\n",
      "SGD:0.18922356751668945\n",
      "Momentum:0.05927696055808592\n",
      "AdaGrad:0.022464614330266943\n",
      "Adam:0.025339742735612722\n",
      "===========iteration:1500===========\n",
      "SGD:0.20645752537383533\n",
      "Momentum:0.03779214320425141\n",
      "AdaGrad:0.02792198455074299\n",
      "Adam:0.03737557342824377\n",
      "===========iteration:1600===========\n",
      "SGD:0.22037265787674276\n",
      "Momentum:0.09495515045411379\n",
      "AdaGrad:0.0534467659802194\n",
      "Adam:0.031553392783269324\n",
      "===========iteration:1700===========\n",
      "SGD:0.2448550850833363\n",
      "Momentum:0.12829015894029996\n",
      "AdaGrad:0.0601422767787269\n",
      "Adam:0.06701559245476615\n",
      "===========iteration:1800===========\n",
      "SGD:0.23775723421343573\n",
      "Momentum:0.0710915454710277\n",
      "AdaGrad:0.03346350129021551\n",
      "Adam:0.03082938643226961\n",
      "===========iteration:1900===========\n",
      "SGD:0.28425196998226004\n",
      "Momentum:0.051791518354410634\n",
      "AdaGrad:0.029313840844783036\n",
      "Adam:0.02073943784355358\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"
    }
   ],
   "source": [
    "# coding: utf-8\n",
    "import os\n",
    "import sys\n",
    "sys.path.append(os.pardir)  # 为导入父目录中的文件而进行的设置\n",
    "import matplotlib.pyplot as plt\n",
    "from dataset.mnist import load_mnist\n",
    "from common.util import smooth_curve\n",
    "from common.multi_layer_net import MultiLayerNet\n",
    "from common.optimizer import *\n",
    "\n",
    "\n",
    "# 0:读入MNIST数据==========\n",
    "(x_train, t_train), (x_test, t_test) = load_mnist(normalize=True)\n",
    "\n",
    "train_size = x_train.shape[0]\n",
    "batch_size = 128\n",
    "max_iterations = 2000\n",
    "\n",
    "\n",
    "# 1:设置实验信息==========\n",
    "optimizers = {}\n",
    "optimizers['SGD'] = SGD()\n",
    "optimizers['Momentum'] = Momentum()\n",
    "optimizers['AdaGrad'] = AdaGrad()\n",
    "optimizers['Adam'] = Adam()\n",
    "#optimizers['RMSprop'] = RMSprop()\n",
    "\n",
    "networks = {}\n",
    "train_loss = {}\n",
    "for key in optimizers.keys():\n",
    "    networks[key] = MultiLayerNet(\n",
    "        input_size=784, hidden_size_list=[100, 100, 100, 100],\n",
    "        output_size=10)\n",
    "    train_loss[key] = []    \n",
    "\n",
    "\n",
    "# 2:开始训练==========\n",
    "for i in range(max_iterations):\n",
    "    batch_mask = np.random.choice(train_size, batch_size)\n",
    "    x_batch = x_train[batch_mask]\n",
    "    t_batch = t_train[batch_mask]\n",
    "    \n",
    "    for key in optimizers.keys():\n",
    "        grads = networks[key].gradient(x_batch, t_batch)\n",
    "        optimizers[key].update(networks[key].params, grads)\n",
    "    \n",
    "        loss = networks[key].loss(x_batch, t_batch)\n",
    "        train_loss[key].append(loss)\n",
    "    \n",
    "    if i % 100 == 0:\n",
    "        print( \"===========\" + \"iteration:\" + str(i) + \"===========\")\n",
    "        for key in optimizers.keys():\n",
    "            loss = networks[key].loss(x_batch, t_batch)\n",
    "            print(key + \":\" + str(loss))\n",
    "\n",
    "\n",
    "# 3.绘制图形==========\n",
    "markers = {\"SGD\": \"o\", \"Momentum\": \"x\", \"AdaGrad\": \"s\", \"Adam\": \"D\"}\n",
    "x = np.arange(max_iterations)\n",
    "for key in optimizers.keys():\n",
    "    plt.plot(x, smooth_curve(train_loss[key]), marker=markers[key], markevery=100, label=key)\n",
    "plt.xlabel(\"iterations\")\n",
    "plt.ylabel(\"loss\")\n",
    "plt.ylim(0, 1)\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 激活值的分布\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch/blob/master/ch06/weight_init_activation_histogram.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 5 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# coding: utf-8\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "\n",
    "def sigmoid(x):\n",
    "    return 1 / (1 + np.exp(-x))\n",
    "\n",
    "\n",
    "def ReLU(x):\n",
    "    return np.maximum(0, x)\n",
    "\n",
    "\n",
    "def tanh(x):\n",
    "    return np.tanh(x)\n",
    "    \n",
    "input_data = np.random.randn(1000, 100)  # 1000个数据\n",
    "node_num = 100  # 各隐藏层节点（神经元）的数量\n",
    "hidden_layer_size = 5  # 隐藏层有5层\n",
    "activations = {}  # 在这里保存激活值的结果\n",
    "\n",
    "x = input_data\n",
    "\n",
    "for i in range(hidden_layer_size):\n",
    "    if i != 0:\n",
    "        x = activations[i-1]\n",
    "\n",
    "    # 尝试用不同的初始值进行实验！\n",
    "    # w = np.random.randn(node_num, node_num) * 1\n",
    "    # w = np.random.randn(node_num, node_num) * 0.01\n",
    "    w = np.random.randn(node_num, node_num) * np.sqrt(1.0 / node_num)\n",
    "    # w = np.random.randn(node_num, node_num) * np.sqrt(2.0 / node_num)\n",
    "\n",
    "\n",
    "    a = np.dot(x, w)\n",
    "\n",
    "\n",
    "    # 也试着换一下激活函数的种类来进行实验！\n",
    "    z = sigmoid(a)\n",
    "    # z = ReLU(a)\n",
    "    # z = tanh(a)\n",
    "\n",
    "    activations[i] = z\n",
    "\n",
    "# 绘制直方图\n",
    "for i, a in activations.items():\n",
    "    plt.subplot(1, len(activations), i+1)\n",
    "    plt.title(str(i+1) + \"-layer\")\n",
    "    if i != 0: plt.yticks([], [])\n",
    "    # plt.xlim(0.1, 1)\n",
    "    # plt.ylim(0, 7000)\n",
    "    plt.hist(a.flatten(), 30, range=(0,1))\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 权重初始值比较\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch/blob/master/ch06/weight_init_compare.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===========iteration:0===========\n",
      "std=0.01:2.302515471747407\n",
      "Xavier:2.3210003557046237\n",
      "He:2.31333297008918\n",
      "===========iteration:100===========\n",
      "std=0.01:2.301834959898998\n",
      "Xavier:2.275458209893899\n",
      "He:1.5594900644256335\n",
      "===========iteration:200===========\n",
      "std=0.01:2.2994267994287982\n",
      "Xavier:2.2159686513659267\n",
      "He:0.714738618948348\n",
      "===========iteration:300===========\n",
      "std=0.01:2.300657306509801\n",
      "Xavier:2.0317902208198975\n",
      "He:0.5849956916013709\n",
      "===========iteration:400===========\n",
      "std=0.01:2.3022105726734345\n",
      "Xavier:1.6909635184185434\n",
      "He:0.483001107020099\n",
      "===========iteration:500===========\n",
      "std=0.01:2.302351213583918\n",
      "Xavier:1.2662492346690404\n",
      "He:0.34994905591474745\n",
      "===========iteration:600===========\n",
      "std=0.01:2.301572125596412\n",
      "Xavier:0.844339446861255\n",
      "He:0.30199587729025995\n",
      "===========iteration:700===========\n",
      "std=0.01:2.3007649916057105\n",
      "Xavier:0.7537531841031204\n",
      "He:0.4008446787556119\n",
      "===========iteration:800===========\n",
      "std=0.01:2.299754334461919\n",
      "Xavier:0.5227419586481694\n",
      "He:0.2675401624045456\n",
      "===========iteration:900===========\n",
      "std=0.01:2.2984636451733156\n",
      "Xavier:0.4812860690786466\n",
      "He:0.30958463969030986\n",
      "===========iteration:1000===========\n",
      "std=0.01:2.305574599087242\n",
      "Xavier:0.5521750377540211\n",
      "He:0.432061645052554\n",
      "===========iteration:1100===========\n",
      "std=0.01:2.304846865418721\n",
      "Xavier:0.5670664874709379\n",
      "He:0.37371247902046795\n",
      "===========iteration:1200===========\n",
      "std=0.01:2.301308794105871\n",
      "Xavier:0.4509517137838819\n",
      "He:0.303562904047647\n",
      "===========iteration:1300===========\n",
      "std=0.01:2.3060297839277943\n",
      "Xavier:0.3660900618261912\n",
      "He:0.23277821237259747\n",
      "===========iteration:1400===========\n",
      "std=0.01:2.2954258923030597\n",
      "Xavier:0.3862744532218011\n",
      "He:0.2752295701921748\n",
      "===========iteration:1500===========\n",
      "std=0.01:2.2964897703410703\n",
      "Xavier:0.2781369191690579\n",
      "He:0.19293132414550354\n",
      "===========iteration:1600===========\n",
      "std=0.01:2.2976145541493924\n",
      "Xavier:0.367356873977376\n",
      "He:0.23912208745992755\n",
      "===========iteration:1700===========\n",
      "std=0.01:2.303923429739336\n",
      "Xavier:0.221372526538493\n",
      "He:0.17173390811001915\n",
      "===========iteration:1800===========\n",
      "std=0.01:2.3046923627864464\n",
      "Xavier:0.36054875676055675\n",
      "He:0.23287920917454058\n",
      "===========iteration:1900===========\n",
      "std=0.01:2.295429773070275\n",
      "Xavier:0.21223665981672446\n",
      "He:0.13955425671673818\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"
    }
   ],
   "source": [
    "# coding: utf-8\n",
    "import os\n",
    "import sys\n",
    "\n",
    "sys.path.append(os.pardir)  # 为导入父目录中的文件而进行的设置\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from dataset.mnist import load_mnist\n",
    "from common.util import smooth_curve\n",
    "from common.multi_layer_net import MultiLayerNet\n",
    "from common.optimizer import SGD\n",
    "\n",
    "\n",
    "# 0:读入MNIST数据==========\n",
    "(x_train, t_train), (x_test, t_test) = load_mnist(normalize=True)\n",
    "\n",
    "train_size = x_train.shape[0]\n",
    "batch_size = 128\n",
    "max_iterations = 2000\n",
    "\n",
    "\n",
    "# 1:设置实验信息==========\n",
    "weight_init_types = {'std=0.01': 0.01, 'Xavier': 'sigmoid', 'He': 'relu'}\n",
    "optimizer = SGD(lr=0.01)\n",
    "\n",
    "networks = {}\n",
    "train_loss = {}\n",
    "for key, weight_type in weight_init_types.items():\n",
    "    networks[key] = MultiLayerNet(input_size=784, hidden_size_list=[100, 100, 100, 100],\n",
    "                                  output_size=10, weight_init_std=weight_type)\n",
    "    train_loss[key] = []\n",
    "\n",
    "\n",
    "# 2:开始训练==========\n",
    "for i in range(max_iterations):\n",
    "    batch_mask = np.random.choice(train_size, batch_size)\n",
    "    x_batch = x_train[batch_mask]\n",
    "    t_batch = t_train[batch_mask]\n",
    "    \n",
    "    for key in weight_init_types.keys():\n",
    "        grads = networks[key].gradient(x_batch, t_batch)\n",
    "        optimizer.update(networks[key].params, grads)\n",
    "    \n",
    "        loss = networks[key].loss(x_batch, t_batch)\n",
    "        train_loss[key].append(loss)\n",
    "    \n",
    "    if i % 100 == 0:\n",
    "        print(\"===========\" + \"iteration:\" + str(i) + \"===========\")\n",
    "        for key in weight_init_types.keys():\n",
    "            loss = networks[key].loss(x_batch, t_batch)\n",
    "            print(key + \":\" + str(loss))\n",
    "\n",
    "\n",
    "# 3.绘制图形==========\n",
    "markers = {'std=0.01': 'o', 'Xavier': 's', 'He': 'D'}\n",
    "x = np.arange(max_iterations)\n",
    "for key in weight_init_types.keys():\n",
    "    plt.plot(x, smooth_curve(train_loss[key]), marker=markers[key], markevery=100, label=key)\n",
    "plt.xlabel(\"iterations\")\n",
    "plt.ylabel(\"loss\")\n",
    "plt.ylim(0, 2.5)\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 使用 Batch Norm后的效果\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch/blob/master/ch06/batch_norm_test.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "============== 1/16 ==============\n",
      "epoch:0 | 0.117 - 0.098\n",
      "epoch:1 | 0.097 - 0.087\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\teimeichi\\deeplearning_from_scratch\\common\\layers.py:12: RuntimeWarning: invalid value encountered in less_equal\n",
      "  self.mask = (x <= 0)\n",
      "C:\\Users\\teimeichi\\deeplearning_from_scratch\\common\\multi_layer_net_extend.py:101: RuntimeWarning: overflow encountered in square\n",
      "  weight_decay += 0.5 * self.weight_decay_lambda * np.sum(W**2)\n",
      "C:\\Users\\teimeichi\\deeplearning_from_scratch\\common\\multi_layer_net_extend.py:101: RuntimeWarning: invalid value encountered in double_scalars\n",
      "  weight_decay += 0.5 * self.weight_decay_lambda * np.sum(W**2)\n",
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\numpy\\core\\fromnumeric.py:86: RuntimeWarning: overflow encountered in reduce\n",
      "  return ufunc.reduce(obj, axis, dtype, out, **passkwargs)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch:2 | 0.097 - 0.123\n",
      "epoch:3 | 0.097 - 0.14\n",
      "epoch:4 | 0.097 - 0.156\n",
      "epoch:5 | 0.097 - 0.187\n",
      "epoch:6 | 0.097 - 0.216\n",
      "epoch:7 | 0.097 - 0.229\n",
      "epoch:8 | 0.097 - 0.255\n",
      "epoch:9 | 0.097 - 0.283\n",
      "epoch:10 | 0.097 - 0.296\n",
      "epoch:11 | 0.097 - 0.316\n",
      "epoch:12 | 0.097 - 0.33\n",
      "epoch:13 | 0.097 - 0.348\n",
      "epoch:14 | 0.097 - 0.371\n",
      "epoch:15 | 0.097 - 0.381\n",
      "epoch:16 | 0.097 - 0.399\n",
      "epoch:17 | 0.097 - 0.403\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch:18 | 0.097 - 0.419\n",
      "epoch:19 | 0.097 - 0.43\n",
      "============== 2/16 ==============\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\teimeichi\\deeplearning_from_scratch\\common\\functions.py:34: RuntimeWarning: invalid value encountered in subtract\n",
      "  x = x - np.max(x, axis=0)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch:0 | 0.092 - 0.091\n",
      "epoch:1 | 0.097 - 0.078\n",
      "epoch:2 | 0.097 - 0.127\n",
      "epoch:3 | 0.097 - 0.14\n",
      "epoch:4 | 0.097 - 0.164\n",
      "epoch:5 | 0.097 - 0.186\n",
      "epoch:6 | 0.097 - 0.202\n",
      "epoch:7 | 0.097 - 0.233\n",
      "epoch:8 | 0.097 - 0.263\n",
      "epoch:9 | 0.097 - 0.287\n",
      "epoch:10 | 0.097 - 0.312\n",
      "epoch:11 | 0.097 - 0.333\n",
      "epoch:12 | 0.097 - 0.363\n",
      "epoch:13 | 0.097 - 0.386\n",
      "epoch:14 | 0.097 - 0.402\n",
      "epoch:15 | 0.097 - 0.419\n",
      "epoch:16 | 0.097 - 0.447\n",
      "epoch:17 | 0.097 - 0.459\n",
      "epoch:18 | 0.097 - 0.479\n",
      "epoch:19 | 0.097 - 0.492\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "============== 3/16 ==============\n",
      "epoch:0 | 0.097 - 0.103\n",
      "epoch:1 | 0.325 - 0.117\n",
      "epoch:2 | 0.49 - 0.15\n",
      "epoch:3 | 0.587 - 0.186\n",
      "epoch:4 | 0.651 - 0.216\n",
      "epoch:5 | 0.729 - 0.25\n",
      "epoch:6 | 0.783 - 0.297\n",
      "epoch:7 | 0.799 - 0.322\n",
      "epoch:8 | 0.847 - 0.357\n",
      "epoch:9 | 0.885 - 0.388\n",
      "epoch:10 | 0.909 - 0.417\n",
      "epoch:11 | 0.919 - 0.449\n",
      "epoch:12 | 0.935 - 0.473\n",
      "epoch:13 | 0.948 - 0.503\n",
      "epoch:14 | 0.956 - 0.532\n",
      "epoch:15 | 0.961 - 0.55\n",
      "epoch:16 | 0.974 - 0.577\n",
      "epoch:17 | 0.981 - 0.596\n",
      "epoch:18 | 0.983 - 0.604\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch:19 | 0.987 - 0.626\n",
      "============== 4/16 ==============\n",
      "epoch:0 | 0.082 - 0.073\n",
      "epoch:1 | 0.246 - 0.09\n",
      "epoch:2 | 0.36 - 0.187\n",
      "epoch:3 | 0.432 - 0.261\n",
      "epoch:4 | 0.499 - 0.324\n",
      "epoch:5 | 0.564 - 0.391\n",
      "epoch:6 | 0.616 - 0.454\n",
      "epoch:7 | 0.66 - 0.504\n",
      "epoch:8 | 0.68 - 0.543\n",
      "epoch:9 | 0.706 - 0.58\n",
      "epoch:10 | 0.73 - 0.61\n",
      "epoch:11 | 0.751 - 0.635\n",
      "epoch:12 | 0.759 - 0.663\n",
      "epoch:13 | 0.774 - 0.678\n",
      "epoch:14 | 0.782 - 0.701\n",
      "epoch:15 | 0.795 - 0.719\n",
      "epoch:16 | 0.811 - 0.739\n",
      "epoch:17 | 0.824 - 0.755\n",
      "epoch:18 | 0.824 - 0.769\n",
      "epoch:19 | 0.831 - 0.782\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "============== 5/16 ==============\n",
      "epoch:0 | 0.096 - 0.098\n",
      "epoch:1 | 0.106 - 0.119\n",
      "epoch:2 | 0.114 - 0.273\n",
      "epoch:3 | 0.116 - 0.408\n",
      "epoch:4 | 0.121 - 0.489\n",
      "epoch:5 | 0.129 - 0.543\n",
      "epoch:6 | 0.136 - 0.592\n",
      "epoch:7 | 0.15 - 0.64\n",
      "epoch:8 | 0.155 - 0.687\n",
      "epoch:9 | 0.166 - 0.703\n",
      "epoch:10 | 0.182 - 0.736\n",
      "epoch:11 | 0.202 - 0.755\n",
      "epoch:12 | 0.217 - 0.784\n",
      "epoch:13 | 0.216 - 0.808\n",
      "epoch:14 | 0.228 - 0.819\n",
      "epoch:15 | 0.234 - 0.837\n",
      "epoch:16 | 0.253 - 0.845\n",
      "epoch:17 | 0.249 - 0.851\n",
      "epoch:18 | 0.26 - 0.856\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch:19 | 0.281 - 0.864\n",
      "============== 6/16 ==============\n",
      "epoch:0 | 0.101 - 0.113\n",
      "epoch:1 | 0.149 - 0.176\n",
      "epoch:2 | 0.128 - 0.378\n",
      "epoch:3 | 0.115 - 0.574\n",
      "epoch:4 | 0.118 - 0.668\n",
      "epoch:5 | 0.116 - 0.717\n",
      "epoch:6 | 0.114 - 0.763\n",
      "epoch:7 | 0.116 - 0.798\n",
      "epoch:8 | 0.114 - 0.818\n",
      "epoch:9 | 0.116 - 0.834\n",
      "epoch:10 | 0.115 - 0.85\n",
      "epoch:11 | 0.114 - 0.864\n",
      "epoch:12 | 0.116 - 0.877\n",
      "epoch:13 | 0.116 - 0.886\n",
      "epoch:14 | 0.116 - 0.897\n",
      "epoch:15 | 0.116 - 0.909\n",
      "epoch:16 | 0.115 - 0.922\n",
      "epoch:17 | 0.116 - 0.933\n",
      "epoch:18 | 0.116 - 0.935\n",
      "epoch:19 | 0.116 - 0.94\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "============== 7/16 ==============\n",
      "epoch:0 | 0.116 - 0.078\n",
      "epoch:1 | 0.116 - 0.356\n",
      "epoch:2 | 0.116 - 0.669\n",
      "epoch:3 | 0.116 - 0.762\n",
      "epoch:4 | 0.116 - 0.782\n",
      "epoch:5 | 0.116 - 0.814\n",
      "epoch:6 | 0.116 - 0.848\n",
      "epoch:7 | 0.116 - 0.866\n",
      "epoch:8 | 0.116 - 0.89\n",
      "epoch:9 | 0.116 - 0.908\n",
      "epoch:10 | 0.116 - 0.927\n",
      "epoch:11 | 0.116 - 0.936\n",
      "epoch:12 | 0.116 - 0.955\n",
      "epoch:13 | 0.116 - 0.97\n",
      "epoch:14 | 0.116 - 0.978\n",
      "epoch:15 | 0.116 - 0.978\n",
      "epoch:16 | 0.116 - 0.985\n",
      "epoch:17 | 0.116 - 0.988\n",
      "epoch:18 | 0.116 - 0.989\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch:19 | 0.116 - 0.991\n",
      "============== 8/16 ==============\n",
      "epoch:0 | 0.1 - 0.097\n",
      "epoch:1 | 0.116 - 0.361\n",
      "epoch:2 | 0.116 - 0.625\n",
      "epoch:3 | 0.116 - 0.706\n",
      "epoch:4 | 0.116 - 0.755\n",
      "epoch:5 | 0.116 - 0.815\n",
      "epoch:6 | 0.116 - 0.873\n",
      "epoch:7 | 0.116 - 0.917\n",
      "epoch:8 | 0.116 - 0.943\n",
      "epoch:9 | 0.116 - 0.961\n",
      "epoch:10 | 0.116 - 0.978\n",
      "epoch:11 | 0.116 - 0.984\n",
      "epoch:12 | 0.116 - 0.987\n",
      "epoch:13 | 0.116 - 0.989\n",
      "epoch:14 | 0.116 - 0.996\n",
      "epoch:15 | 0.116 - 0.998\n",
      "epoch:16 | 0.116 - 0.998\n",
      "epoch:17 | 0.116 - 0.999\n",
      "epoch:18 | 0.116 - 0.999\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch:19 | 0.116 - 0.999\n",
      "============== 9/16 ==============\n",
      "epoch:0 | 0.117 - 0.101\n",
      "epoch:1 | 0.117 - 0.456\n",
      "epoch:2 | 0.117 - 0.794\n",
      "epoch:3 | 0.117 - 0.864\n",
      "epoch:4 | 0.117 - 0.913\n",
      "epoch:5 | 0.117 - 0.945\n",
      "epoch:6 | 0.117 - 0.973\n",
      "epoch:7 | 0.117 - 0.982\n",
      "epoch:8 | 0.117 - 0.988\n",
      "epoch:9 | 0.117 - 0.989\n",
      "epoch:10 | 0.117 - 0.98\n",
      "epoch:11 | 0.117 - 0.993\n",
      "epoch:12 | 0.117 - 0.996\n",
      "epoch:13 | 0.117 - 0.998\n",
      "epoch:14 | 0.117 - 0.998\n",
      "epoch:15 | 0.117 - 0.998\n",
      "epoch:16 | 0.116 - 0.999\n",
      "epoch:17 | 0.116 - 1.0\n",
      "epoch:18 | 0.116 - 1.0\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch:19 | 0.116 - 1.0\n",
      "============== 10/16 ==============\n",
      "epoch:0 | 0.117 - 0.157\n",
      "epoch:1 | 0.117 - 0.423\n",
      "epoch:2 | 0.117 - 0.759\n",
      "epoch:3 | 0.117 - 0.812\n",
      "epoch:4 | 0.117 - 0.868\n",
      "epoch:5 | 0.117 - 0.891\n",
      "epoch:6 | 0.117 - 0.937\n",
      "epoch:7 | 0.117 - 0.954\n",
      "epoch:8 | 0.117 - 0.977\n",
      "epoch:9 | 0.117 - 0.987\n",
      "epoch:10 | 0.117 - 0.986\n",
      "epoch:11 | 0.117 - 0.991\n",
      "epoch:12 | 0.117 - 0.989\n",
      "epoch:13 | 0.117 - 0.988\n",
      "epoch:14 | 0.117 - 0.994\n",
      "epoch:15 | 0.117 - 0.997\n",
      "epoch:16 | 0.117 - 0.995\n",
      "epoch:17 | 0.117 - 0.996\n",
      "epoch:18 | 0.117 - 0.997\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch:19 | 0.117 - 0.998\n",
      "============== 11/16 ==============\n",
      "epoch:0 | 0.116 - 0.099\n",
      "epoch:1 | 0.117 - 0.511\n",
      "epoch:2 | 0.116 - 0.694\n",
      "epoch:3 | 0.117 - 0.649\n",
      "epoch:4 | 0.117 - 0.839\n",
      "epoch:5 | 0.117 - 0.816\n",
      "epoch:6 | 0.117 - 0.83\n",
      "epoch:7 | 0.117 - 0.871\n",
      "epoch:8 | 0.117 - 0.894\n",
      "epoch:9 | 0.117 - 0.881\n",
      "epoch:10 | 0.117 - 0.898\n",
      "epoch:11 | 0.117 - 0.9\n",
      "epoch:12 | 0.117 - 0.902\n",
      "epoch:13 | 0.117 - 0.903\n",
      "epoch:14 | 0.117 - 0.904\n",
      "epoch:15 | 0.117 - 0.904\n",
      "epoch:16 | 0.117 - 0.895\n",
      "epoch:17 | 0.117 - 0.858\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch:18 | 0.117 - 0.987\n",
      "epoch:19 | 0.116 - 0.988\n",
      "============== 12/16 ==============\n",
      "epoch:0 | 0.117 - 0.122\n",
      "epoch:1 | 0.116 - 0.586\n",
      "epoch:2 | 0.116 - 0.613\n",
      "epoch:3 | 0.116 - 0.667\n",
      "epoch:4 | 0.117 - 0.714\n",
      "epoch:5 | 0.117 - 0.717\n",
      "epoch:6 | 0.116 - 0.739\n",
      "epoch:7 | 0.116 - 0.744\n",
      "epoch:8 | 0.116 - 0.755\n",
      "epoch:9 | 0.116 - 0.734\n",
      "epoch:10 | 0.116 - 0.783\n",
      "epoch:11 | 0.116 - 0.789\n",
      "epoch:12 | 0.116 - 0.796\n",
      "epoch:13 | 0.116 - 0.767\n",
      "epoch:14 | 0.116 - 0.797\n",
      "epoch:15 | 0.116 - 0.738\n",
      "epoch:16 | 0.116 - 0.725\n",
      "epoch:17 | 0.116 - 0.721\n",
      "epoch:18 | 0.116 - 0.801\n",
      "epoch:19 | 0.116 - 0.798\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "============== 13/16 ==============\n",
      "epoch:0 | 0.117 - 0.141\n",
      "epoch:1 | 0.105 - 0.443\n",
      "epoch:2 | 0.117 - 0.617\n",
      "epoch:3 | 0.117 - 0.589\n",
      "epoch:4 | 0.117 - 0.563\n",
      "epoch:5 | 0.117 - 0.685\n",
      "epoch:6 | 0.116 - 0.742\n",
      "epoch:7 | 0.116 - 0.702\n",
      "epoch:8 | 0.116 - 0.766\n",
      "epoch:9 | 0.116 - 0.758\n",
      "epoch:10 | 0.116 - 0.742\n",
      "epoch:11 | 0.116 - 0.712\n",
      "epoch:12 | 0.117 - 0.747\n",
      "epoch:13 | 0.117 - 0.709\n",
      "epoch:14 | 0.116 - 0.703\n",
      "epoch:15 | 0.116 - 0.708\n",
      "epoch:16 | 0.116 - 0.797\n",
      "epoch:17 | 0.116 - 0.799\n",
      "epoch:18 | 0.116 - 0.759\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch:19 | 0.116 - 0.796\n",
      "============== 14/16 ==============\n",
      "epoch:0 | 0.117 - 0.186\n",
      "epoch:1 | 0.099 - 0.413\n",
      "epoch:2 | 0.117 - 0.378\n",
      "epoch:3 | 0.117 - 0.52\n",
      "epoch:4 | 0.117 - 0.549\n",
      "epoch:5 | 0.117 - 0.569\n",
      "epoch:6 | 0.117 - 0.572\n",
      "epoch:7 | 0.117 - 0.578\n",
      "epoch:8 | 0.117 - 0.584\n",
      "epoch:9 | 0.117 - 0.592\n",
      "epoch:10 | 0.117 - 0.593\n",
      "epoch:11 | 0.117 - 0.6\n",
      "epoch:12 | 0.117 - 0.591\n",
      "epoch:13 | 0.117 - 0.596\n",
      "epoch:14 | 0.116 - 0.601\n",
      "epoch:15 | 0.117 - 0.6\n",
      "epoch:16 | 0.117 - 0.602\n",
      "epoch:17 | 0.117 - 0.602\n",
      "epoch:18 | 0.117 - 0.602\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch:19 | 0.117 - 0.582\n",
      "============== 15/16 ==============\n",
      "epoch:0 | 0.087 - 0.097\n",
      "epoch:1 | 0.116 - 0.316\n",
      "epoch:2 | 0.116 - 0.395\n",
      "epoch:3 | 0.117 - 0.477\n",
      "epoch:4 | 0.117 - 0.478\n",
      "epoch:5 | 0.117 - 0.51\n",
      "epoch:6 | 0.117 - 0.568\n",
      "epoch:7 | 0.117 - 0.595\n",
      "epoch:8 | 0.117 - 0.588\n",
      "epoch:9 | 0.117 - 0.602\n",
      "epoch:10 | 0.117 - 0.602\n",
      "epoch:11 | 0.117 - 0.593\n",
      "epoch:12 | 0.117 - 0.585\n",
      "epoch:13 | 0.117 - 0.601\n",
      "epoch:14 | 0.117 - 0.605\n",
      "epoch:15 | 0.117 - 0.602\n",
      "epoch:16 | 0.117 - 0.61\n",
      "epoch:17 | 0.117 - 0.611\n",
      "epoch:18 | 0.117 - 0.604\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch:19 | 0.117 - 0.602\n",
      "============== 16/16 ==============\n",
      "epoch:0 | 0.087 - 0.088\n",
      "epoch:1 | 0.117 - 0.285\n",
      "epoch:2 | 0.117 - 0.337\n",
      "epoch:3 | 0.117 - 0.398\n",
      "epoch:4 | 0.117 - 0.4\n",
      "epoch:5 | 0.117 - 0.388\n",
      "epoch:6 | 0.117 - 0.395\n",
      "epoch:7 | 0.117 - 0.404\n",
      "epoch:8 | 0.117 - 0.411\n",
      "epoch:9 | 0.117 - 0.41\n",
      "epoch:10 | 0.117 - 0.42\n",
      "epoch:11 | 0.117 - 0.416\n",
      "epoch:12 | 0.117 - 0.412\n",
      "epoch:13 | 0.117 - 0.421\n",
      "epoch:14 | 0.117 - 0.426\n",
      "epoch:15 | 0.117 - 0.435\n",
      "epoch:16 | 0.117 - 0.439\n",
      "epoch:17 | 0.117 - 0.463\n",
      "epoch:18 | 0.117 - 0.458\n",
      "epoch:19 | 0.117 - 0.492\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 16 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# coding: utf-8\n",
    "import sys, os\n",
    "sys.path.append(os.pardir)  # 为导入父目录中的文件而进行的设置\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from dataset.mnist import load_mnist\n",
    "from common.multi_layer_net_extend import MultiLayerNetExtend\n",
    "from common.optimizer import SGD, Adam\n",
    "\n",
    "(x_train, t_train), (x_test, t_test) = load_mnist(normalize=True)\n",
    "\n",
    "# 减少学习数据\n",
    "x_train = x_train[:1000]\n",
    "t_train = t_train[:1000]\n",
    "\n",
    "max_epochs = 20\n",
    "train_size = x_train.shape[0]\n",
    "batch_size = 100\n",
    "learning_rate = 0.01\n",
    "\n",
    "\n",
    "def __train(weight_init_std):\n",
    "    bn_network = MultiLayerNetExtend(input_size=784, hidden_size_list=[100, 100, 100, 100, 100], output_size=10, \n",
    "                                    weight_init_std=weight_init_std, use_batchnorm=True)\n",
    "    network = MultiLayerNetExtend(input_size=784, hidden_size_list=[100, 100, 100, 100, 100], output_size=10,\n",
    "                                weight_init_std=weight_init_std)\n",
    "    optimizer = SGD(lr=learning_rate)\n",
    "    \n",
    "    train_acc_list = []\n",
    "    bn_train_acc_list = []\n",
    "    \n",
    "    iter_per_epoch = max(train_size / batch_size, 1)\n",
    "    epoch_cnt = 0\n",
    "    \n",
    "    for i in range(1000000000):\n",
    "        batch_mask = np.random.choice(train_size, batch_size)\n",
    "        x_batch = x_train[batch_mask]\n",
    "        t_batch = t_train[batch_mask]\n",
    "    \n",
    "        for _network in (bn_network, network):\n",
    "            grads = _network.gradient(x_batch, t_batch)\n",
    "            optimizer.update(_network.params, grads)\n",
    "    \n",
    "        if i % iter_per_epoch == 0:\n",
    "            train_acc = network.accuracy(x_train, t_train)\n",
    "            bn_train_acc = bn_network.accuracy(x_train, t_train)\n",
    "            train_acc_list.append(train_acc)\n",
    "            bn_train_acc_list.append(bn_train_acc)\n",
    "    \n",
    "            print(\"epoch:\" + str(epoch_cnt) + \" | \" + str(train_acc) + \" - \" + str(bn_train_acc))\n",
    "    \n",
    "            epoch_cnt += 1\n",
    "            if epoch_cnt >= max_epochs:\n",
    "                break\n",
    "                \n",
    "    return train_acc_list, bn_train_acc_list\n",
    "\n",
    "\n",
    "# 3.绘制图形==========\n",
    "weight_scale_list = np.logspace(0, -4, num=16)\n",
    "x = np.arange(max_epochs)\n",
    "\n",
    "for i, w in enumerate(weight_scale_list):\n",
    "    print( \"============== \" + str(i+1) + \"/16\" + \" ==============\")\n",
    "    train_acc_list, bn_train_acc_list = __train(w)\n",
    "    \n",
    "    plt.subplot(4,4,i+1)\n",
    "    plt.title(\"W:\" + str(w))\n",
    "    if i == 15:\n",
    "        plt.plot(x, bn_train_acc_list, label='Batch Normalization', markevery=2)\n",
    "        plt.plot(x, train_acc_list, linestyle = \"--\", label='Normal(without BatchNorm)', markevery=2)\n",
    "    else:\n",
    "        plt.plot(x, bn_train_acc_list, markevery=2)\n",
    "        plt.plot(x, train_acc_list, linestyle=\"--\", markevery=2)\n",
    "\n",
    "    plt.ylim(0, 1.0)\n",
    "    if i % 4:\n",
    "        plt.yticks([])\n",
    "    else:\n",
    "        plt.ylabel(\"accuracy\")\n",
    "    if i < 12:\n",
    "        plt.xticks([])\n",
    "    else:\n",
    "        plt.xlabel(\"epochs\")\n",
    "    plt.legend(loc='lower right')\n",
    "    \n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 检查Batch Norm梯度\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch/blob/master/ch06/batch_norm_gradient_check.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "W1:0.0\n",
      "b1:0.0\n",
      "gamma1:0.0\n",
      "beta1:0.0\n",
      "W2:0.0\n",
      "b2:0.0\n",
      "gamma2:0.0\n",
      "beta2:0.04921507676547421\n",
      "W3:0.0\n",
      "b3:1.7990402263745597e-07\n"
     ]
    }
   ],
   "source": [
    "# coding: utf-8\n",
    "import sys, os\n",
    "sys.path.append(os.pardir)  # 为导入父目录中的文件而进行的设置\n",
    "import numpy as np\n",
    "from dataset.mnist import load_mnist\n",
    "from common.multi_layer_net_extend import MultiLayerNetExtend\n",
    "\n",
    "# 读入数据\n",
    "(x_train, t_train), (x_test, t_test) = load_mnist(normalize=True, one_hot_label=True)\n",
    "\n",
    "network = MultiLayerNetExtend(input_size=784, hidden_size_list=[100, 100], output_size=10,\n",
    "                              use_batchnorm=True)\n",
    "\n",
    "x_batch = x_train[:1]\n",
    "t_batch = t_train[:1]\n",
    "\n",
    "grad_backprop = network.gradient(x_batch, t_batch)\n",
    "grad_numerical = network.numerical_gradient(x_batch, t_batch)\n",
    "\n",
    "for key in grad_numerical.keys():\n",
    "    diff = np.average( np.abs(grad_backprop[key] - grad_numerical[key]) )\n",
    "    print(key + \":\" + str(diff))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 过拟合\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch/blob/master/ch06/overfit_weight_decay.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch:0, train acc:0.11, test acc:0.1008\n",
      "epoch:1, train acc:0.11, test acc:0.1008\n",
      "epoch:2, train acc:0.11333333333333333, test acc:0.1013\n",
      "epoch:3, train acc:0.12, test acc:0.102\n",
      "epoch:4, train acc:0.12333333333333334, test acc:0.1027\n",
      "epoch:5, train acc:0.12, test acc:0.1015\n",
      "epoch:6, train acc:0.13, test acc:0.1034\n",
      "epoch:7, train acc:0.13333333333333333, test acc:0.1045\n",
      "epoch:8, train acc:0.13333333333333333, test acc:0.1071\n",
      "epoch:9, train acc:0.14333333333333334, test acc:0.111\n",
      "epoch:10, train acc:0.14333333333333334, test acc:0.1152\n",
      "epoch:11, train acc:0.15666666666666668, test acc:0.1182\n",
      "epoch:12, train acc:0.18666666666666668, test acc:0.1316\n",
      "epoch:13, train acc:0.19666666666666666, test acc:0.1414\n",
      "epoch:14, train acc:0.22666666666666666, test acc:0.1699\n",
      "epoch:15, train acc:0.22333333333333333, test acc:0.1662\n",
      "epoch:16, train acc:0.24333333333333335, test acc:0.1779\n",
      "epoch:17, train acc:0.25666666666666665, test acc:0.1901\n",
      "epoch:18, train acc:0.26666666666666666, test acc:0.2145\n",
      "epoch:19, train acc:0.2833333333333333, test acc:0.2255\n",
      "epoch:20, train acc:0.30333333333333334, test acc:0.2372\n",
      "epoch:21, train acc:0.33, test acc:0.2613\n",
      "epoch:22, train acc:0.34, test acc:0.2696\n",
      "epoch:23, train acc:0.3566666666666667, test acc:0.2805\n",
      "epoch:24, train acc:0.38333333333333336, test acc:0.2971\n",
      "epoch:25, train acc:0.38666666666666666, test acc:0.3064\n",
      "epoch:26, train acc:0.41, test acc:0.3181\n",
      "epoch:27, train acc:0.4266666666666667, test acc:0.3216\n",
      "epoch:28, train acc:0.42, test acc:0.3256\n",
      "epoch:29, train acc:0.45, test acc:0.3404\n",
      "epoch:30, train acc:0.4666666666666667, test acc:0.3458\n",
      "epoch:31, train acc:0.47333333333333333, test acc:0.3393\n",
      "epoch:32, train acc:0.48, test acc:0.3452\n",
      "epoch:33, train acc:0.5066666666666667, test acc:0.3634\n",
      "epoch:34, train acc:0.5033333333333333, test acc:0.3675\n",
      "epoch:35, train acc:0.53, test acc:0.3807\n",
      "epoch:36, train acc:0.5433333333333333, test acc:0.3857\n",
      "epoch:37, train acc:0.5466666666666666, test acc:0.4015\n",
      "epoch:38, train acc:0.5433333333333333, test acc:0.3932\n",
      "epoch:39, train acc:0.5833333333333334, test acc:0.4149\n",
      "epoch:40, train acc:0.5933333333333334, test acc:0.4191\n",
      "epoch:41, train acc:0.61, test acc:0.4307\n",
      "epoch:42, train acc:0.62, test acc:0.4477\n",
      "epoch:43, train acc:0.6233333333333333, test acc:0.453\n",
      "epoch:44, train acc:0.62, test acc:0.4476\n",
      "epoch:45, train acc:0.6166666666666667, test acc:0.4488\n",
      "epoch:46, train acc:0.6333333333333333, test acc:0.4555\n",
      "epoch:47, train acc:0.6466666666666666, test acc:0.4701\n",
      "epoch:48, train acc:0.66, test acc:0.4884\n",
      "epoch:49, train acc:0.66, test acc:0.49\n",
      "epoch:50, train acc:0.6666666666666666, test acc:0.5028\n",
      "epoch:51, train acc:0.6666666666666666, test acc:0.503\n",
      "epoch:52, train acc:0.6633333333333333, test acc:0.5014\n",
      "epoch:53, train acc:0.6633333333333333, test acc:0.5056\n",
      "epoch:54, train acc:0.6766666666666666, test acc:0.5144\n",
      "epoch:55, train acc:0.6866666666666666, test acc:0.5226\n",
      "epoch:56, train acc:0.69, test acc:0.5285\n",
      "epoch:57, train acc:0.6966666666666667, test acc:0.5395\n",
      "epoch:58, train acc:0.69, test acc:0.5297\n",
      "epoch:59, train acc:0.69, test acc:0.537\n",
      "epoch:60, train acc:0.71, test acc:0.5548\n",
      "epoch:61, train acc:0.7133333333333334, test acc:0.5546\n",
      "epoch:62, train acc:0.6966666666666667, test acc:0.5531\n",
      "epoch:63, train acc:0.7033333333333334, test acc:0.5474\n",
      "epoch:64, train acc:0.7266666666666667, test acc:0.5649\n",
      "epoch:65, train acc:0.74, test acc:0.5666\n",
      "epoch:66, train acc:0.7433333333333333, test acc:0.5774\n",
      "epoch:67, train acc:0.7333333333333333, test acc:0.5778\n",
      "epoch:68, train acc:0.7366666666666667, test acc:0.5835\n",
      "epoch:69, train acc:0.76, test acc:0.5916\n",
      "epoch:70, train acc:0.77, test acc:0.6031\n",
      "epoch:71, train acc:0.7633333333333333, test acc:0.6039\n",
      "epoch:72, train acc:0.7533333333333333, test acc:0.5991\n",
      "epoch:73, train acc:0.7666666666666667, test acc:0.6153\n",
      "epoch:74, train acc:0.7833333333333333, test acc:0.6188\n",
      "epoch:75, train acc:0.7866666666666666, test acc:0.6269\n",
      "epoch:76, train acc:0.7633333333333333, test acc:0.6244\n",
      "epoch:77, train acc:0.7733333333333333, test acc:0.628\n",
      "epoch:78, train acc:0.7733333333333333, test acc:0.6209\n",
      "epoch:79, train acc:0.7633333333333333, test acc:0.6147\n",
      "epoch:80, train acc:0.7766666666666666, test acc:0.6346\n",
      "epoch:81, train acc:0.7866666666666666, test acc:0.6342\n",
      "epoch:82, train acc:0.8033333333333333, test acc:0.6336\n",
      "epoch:83, train acc:0.8066666666666666, test acc:0.6361\n",
      "epoch:84, train acc:0.79, test acc:0.6381\n",
      "epoch:85, train acc:0.8, test acc:0.6325\n",
      "epoch:86, train acc:0.8166666666666667, test acc:0.6398\n",
      "epoch:87, train acc:0.82, test acc:0.6358\n",
      "epoch:88, train acc:0.8133333333333334, test acc:0.6394\n",
      "epoch:89, train acc:0.8533333333333334, test acc:0.6553\n",
      "epoch:90, train acc:0.8166666666666667, test acc:0.6477\n",
      "epoch:91, train acc:0.82, test acc:0.6477\n",
      "epoch:92, train acc:0.8333333333333334, test acc:0.6423\n",
      "epoch:93, train acc:0.8266666666666667, test acc:0.6468\n",
      "epoch:94, train acc:0.85, test acc:0.653\n",
      "epoch:95, train acc:0.83, test acc:0.648\n",
      "epoch:96, train acc:0.85, test acc:0.6576\n",
      "epoch:97, train acc:0.85, test acc:0.656\n",
      "epoch:98, train acc:0.8466666666666667, test acc:0.6548\n",
      "epoch:99, train acc:0.8433333333333334, test acc:0.662\n",
      "epoch:100, train acc:0.8233333333333334, test acc:0.6691\n",
      "epoch:101, train acc:0.8666666666666667, test acc:0.6624\n",
      "epoch:102, train acc:0.84, test acc:0.6563\n",
      "epoch:103, train acc:0.8533333333333334, test acc:0.664\n",
      "epoch:104, train acc:0.8666666666666667, test acc:0.6683\n",
      "epoch:105, train acc:0.85, test acc:0.6701\n",
      "epoch:106, train acc:0.8633333333333333, test acc:0.6707\n",
      "epoch:107, train acc:0.85, test acc:0.6731\n",
      "epoch:108, train acc:0.84, test acc:0.6706\n",
      "epoch:109, train acc:0.8766666666666667, test acc:0.678\n",
      "epoch:110, train acc:0.8533333333333334, test acc:0.6751\n",
      "epoch:111, train acc:0.86, test acc:0.6776\n",
      "epoch:112, train acc:0.87, test acc:0.6729\n",
      "epoch:113, train acc:0.88, test acc:0.678\n",
      "epoch:114, train acc:0.8766666666666667, test acc:0.6788\n",
      "epoch:115, train acc:0.87, test acc:0.6827\n",
      "epoch:116, train acc:0.8666666666666667, test acc:0.6778\n",
      "epoch:117, train acc:0.87, test acc:0.6681\n",
      "epoch:118, train acc:0.8766666666666667, test acc:0.6788\n",
      "epoch:119, train acc:0.86, test acc:0.6733\n",
      "epoch:120, train acc:0.8733333333333333, test acc:0.6823\n",
      "epoch:121, train acc:0.88, test acc:0.6843\n",
      "epoch:122, train acc:0.8766666666666667, test acc:0.6885\n",
      "epoch:123, train acc:0.8666666666666667, test acc:0.6896\n",
      "epoch:124, train acc:0.8866666666666667, test acc:0.6894\n",
      "epoch:125, train acc:0.8766666666666667, test acc:0.6827\n",
      "epoch:126, train acc:0.8733333333333333, test acc:0.6897\n",
      "epoch:127, train acc:0.88, test acc:0.6891\n",
      "epoch:128, train acc:0.8933333333333333, test acc:0.692\n",
      "epoch:129, train acc:0.88, test acc:0.6956\n",
      "epoch:130, train acc:0.8833333333333333, test acc:0.6873\n",
      "epoch:131, train acc:0.8866666666666667, test acc:0.6885\n",
      "epoch:132, train acc:0.8866666666666667, test acc:0.6826\n",
      "epoch:133, train acc:0.88, test acc:0.6866\n",
      "epoch:134, train acc:0.8866666666666667, test acc:0.6933\n",
      "epoch:135, train acc:0.8833333333333333, test acc:0.6911\n",
      "epoch:136, train acc:0.8866666666666667, test acc:0.6887\n",
      "epoch:137, train acc:0.89, test acc:0.6959\n",
      "epoch:138, train acc:0.88, test acc:0.6989\n",
      "epoch:139, train acc:0.8833333333333333, test acc:0.6956\n",
      "epoch:140, train acc:0.8833333333333333, test acc:0.6974\n",
      "epoch:141, train acc:0.9, test acc:0.6998\n",
      "epoch:142, train acc:0.8833333333333333, test acc:0.6902\n",
      "epoch:143, train acc:0.89, test acc:0.7007\n",
      "epoch:144, train acc:0.8966666666666666, test acc:0.6954\n",
      "epoch:145, train acc:0.9033333333333333, test acc:0.7033\n",
      "epoch:146, train acc:0.8966666666666666, test acc:0.7\n",
      "epoch:147, train acc:0.8966666666666666, test acc:0.7013\n",
      "epoch:148, train acc:0.8866666666666667, test acc:0.695\n",
      "epoch:149, train acc:0.8966666666666666, test acc:0.7016\n",
      "epoch:150, train acc:0.8933333333333333, test acc:0.6968\n",
      "epoch:151, train acc:0.9, test acc:0.6987\n",
      "epoch:152, train acc:0.8933333333333333, test acc:0.6938\n",
      "epoch:153, train acc:0.8833333333333333, test acc:0.6907\n",
      "epoch:154, train acc:0.9, test acc:0.7059\n",
      "epoch:155, train acc:0.8933333333333333, test acc:0.6996\n",
      "epoch:156, train acc:0.8966666666666666, test acc:0.7067\n",
      "epoch:157, train acc:0.9, test acc:0.7086\n",
      "epoch:158, train acc:0.8933333333333333, test acc:0.7059\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch:159, train acc:0.8966666666666666, test acc:0.7016\n",
      "epoch:160, train acc:0.8933333333333333, test acc:0.7027\n",
      "epoch:161, train acc:0.8933333333333333, test acc:0.705\n",
      "epoch:162, train acc:0.8966666666666666, test acc:0.7008\n",
      "epoch:163, train acc:0.8966666666666666, test acc:0.703\n",
      "epoch:164, train acc:0.9066666666666666, test acc:0.7119\n",
      "epoch:165, train acc:0.9, test acc:0.707\n",
      "epoch:166, train acc:0.89, test acc:0.7078\n",
      "epoch:167, train acc:0.8966666666666666, test acc:0.7088\n",
      "epoch:168, train acc:0.9033333333333333, test acc:0.7093\n",
      "epoch:169, train acc:0.91, test acc:0.7115\n",
      "epoch:170, train acc:0.8933333333333333, test acc:0.7\n",
      "epoch:171, train acc:0.8966666666666666, test acc:0.7048\n",
      "epoch:172, train acc:0.9, test acc:0.7079\n",
      "epoch:173, train acc:0.9133333333333333, test acc:0.7087\n",
      "epoch:174, train acc:0.8933333333333333, test acc:0.7082\n",
      "epoch:175, train acc:0.9066666666666666, test acc:0.7119\n",
      "epoch:176, train acc:0.9133333333333333, test acc:0.7172\n",
      "epoch:177, train acc:0.9, test acc:0.7101\n",
      "epoch:178, train acc:0.9066666666666666, test acc:0.7105\n",
      "epoch:179, train acc:0.9033333333333333, test acc:0.7082\n",
      "epoch:180, train acc:0.9033333333333333, test acc:0.7157\n",
      "epoch:181, train acc:0.9, test acc:0.7062\n",
      "epoch:182, train acc:0.91, test acc:0.7112\n",
      "epoch:183, train acc:0.9166666666666666, test acc:0.7154\n",
      "epoch:184, train acc:0.8933333333333333, test acc:0.7075\n",
      "epoch:185, train acc:0.9033333333333333, test acc:0.7101\n",
      "epoch:186, train acc:0.9033333333333333, test acc:0.7122\n",
      "epoch:187, train acc:0.9066666666666666, test acc:0.7105\n",
      "epoch:188, train acc:0.9066666666666666, test acc:0.7101\n",
      "epoch:189, train acc:0.9, test acc:0.7074\n",
      "epoch:190, train acc:0.9033333333333333, test acc:0.7115\n",
      "epoch:191, train acc:0.91, test acc:0.7124\n",
      "epoch:192, train acc:0.8866666666666667, test acc:0.705\n",
      "epoch:193, train acc:0.91, test acc:0.7132\n",
      "epoch:194, train acc:0.9066666666666666, test acc:0.7139\n",
      "epoch:195, train acc:0.91, test acc:0.7088\n",
      "epoch:196, train acc:0.91, test acc:0.7081\n",
      "epoch:197, train acc:0.9133333333333333, test acc:0.7081\n",
      "epoch:198, train acc:0.9033333333333333, test acc:0.7152\n",
      "epoch:199, train acc:0.9066666666666666, test acc:0.7096\n",
      "epoch:200, train acc:0.91, test acc:0.718\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"
    }
   ],
   "source": [
    "# coding: utf-8\n",
    "import os\n",
    "import sys\n",
    "\n",
    "sys.path.append(os.pardir)  # 为导入父目录中的文件而进行的设置\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from dataset.mnist import load_mnist\n",
    "from common.multi_layer_net import MultiLayerNet\n",
    "from common.optimizer import SGD\n",
    "\n",
    "(x_train, t_train), (x_test, t_test) = load_mnist(normalize=True)\n",
    "\n",
    "# 为了重现过拟合，减少学习数据\n",
    "x_train = x_train[:300]\n",
    "t_train = t_train[:300]\n",
    "\n",
    "# 设置weight decay（权重衰减） =======================\n",
    "# weight_decay_lambda = 0 # 不使用weight decay的情况\n",
    "weight_decay_lambda = 0.1\n",
    "# ====================================================\n",
    "\n",
    "network = MultiLayerNet(input_size=784, hidden_size_list=[100, 100, 100, 100, 100, 100], output_size=10,\n",
    "                        weight_decay_lambda=weight_decay_lambda)\n",
    "optimizer = SGD(lr=0.01)\n",
    "\n",
    "max_epochs = 201\n",
    "train_size = x_train.shape[0]\n",
    "batch_size = 100\n",
    "\n",
    "train_loss_list = []\n",
    "train_acc_list = []\n",
    "test_acc_list = []\n",
    "\n",
    "iter_per_epoch = max(train_size / batch_size, 1)\n",
    "epoch_cnt = 0\n",
    "\n",
    "for i in range(1000000000):\n",
    "    batch_mask = np.random.choice(train_size, batch_size)\n",
    "    x_batch = x_train[batch_mask]\n",
    "    t_batch = t_train[batch_mask]\n",
    "\n",
    "    grads = network.gradient(x_batch, t_batch)\n",
    "    optimizer.update(network.params, grads)\n",
    "\n",
    "    if i % iter_per_epoch == 0:\n",
    "        train_acc = network.accuracy(x_train, t_train)\n",
    "        test_acc = network.accuracy(x_test, t_test)\n",
    "        train_acc_list.append(train_acc)\n",
    "        test_acc_list.append(test_acc)\n",
    "\n",
    "        print(\"epoch:\" + str(epoch_cnt) + \", train acc:\" + str(train_acc) + \", test acc:\" + str(test_acc))\n",
    "\n",
    "        epoch_cnt += 1\n",
    "        if epoch_cnt >= max_epochs:\n",
    "            break\n",
    "\n",
    "\n",
    "# 3.绘制图形==========\n",
    "markers = {'train': 'o', 'test': 's'}\n",
    "x = np.arange(max_epochs)\n",
    "plt.plot(x, train_acc_list, marker='o', label='train', markevery=10)\n",
    "plt.plot(x, test_acc_list, marker='s', label='test', markevery=10)\n",
    "plt.xlabel(\"epochs\")\n",
    "plt.ylabel(\"accuracy\")\n",
    "plt.ylim(0, 1.0)\n",
    "plt.legend(loc='lower right')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# dropout\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch/blob/master/ch06/overfit_dropout.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "train loss:2.2940388257875033\n",
      "=== epoch:1, train acc:0.08333333333333333, test acc:0.1032 ===\n",
      "train loss:2.308232859975659\n",
      "train loss:2.3103346159001243\n",
      "train loss:2.2997437891101793\n",
      "=== epoch:2, train acc:0.08666666666666667, test acc:0.1044 ===\n",
      "train loss:2.2990201182070833\n",
      "train loss:2.3063753163964673\n",
      "train loss:2.301757694719673\n",
      "=== epoch:3, train acc:0.08666666666666667, test acc:0.1085 ===\n",
      "train loss:2.289907493579498\n",
      "train loss:2.3001036707568727\n",
      "train loss:2.3065279836350445\n",
      "=== epoch:4, train acc:0.08666666666666667, test acc:0.1104 ===\n",
      "train loss:2.3011571609762025\n",
      "train loss:2.302940677033991\n",
      "train loss:2.307538649520236\n",
      "=== epoch:5, train acc:0.09, test acc:0.1104 ===\n",
      "train loss:2.3007307039051046\n",
      "train loss:2.291755477734266\n",
      "train loss:2.3038531548484884\n",
      "=== epoch:6, train acc:0.10333333333333333, test acc:0.1121 ===\n",
      "train loss:2.294566543524385\n",
      "train loss:2.3052050176727747\n",
      "train loss:2.300792401627994\n",
      "=== epoch:7, train acc:0.11666666666666667, test acc:0.1142 ===\n",
      "train loss:2.2995501634857165\n",
      "train loss:2.2954389959312835\n",
      "train loss:2.2981022011318153\n",
      "=== epoch:8, train acc:0.11666666666666667, test acc:0.1151 ===\n",
      "train loss:2.2896184383199523\n",
      "train loss:2.3008491885695417\n",
      "train loss:2.285244675201674\n",
      "=== epoch:9, train acc:0.12, test acc:0.1156 ===\n",
      "train loss:2.2904687302399114\n",
      "train loss:2.296682766096851\n",
      "train loss:2.2896129732008625\n",
      "=== epoch:10, train acc:0.11666666666666667, test acc:0.1195 ===\n",
      "train loss:2.2873895684863417\n",
      "train loss:2.301087008500582\n",
      "train loss:2.2883772604118042\n",
      "=== epoch:11, train acc:0.13, test acc:0.1238 ===\n",
      "train loss:2.2884586114828713\n",
      "train loss:2.289438349650184\n",
      "train loss:2.295089093130189\n",
      "=== epoch:12, train acc:0.12666666666666668, test acc:0.1259 ===\n",
      "train loss:2.301516864034868\n",
      "train loss:2.291766044760991\n",
      "train loss:2.2912910834329305\n",
      "=== epoch:13, train acc:0.14, test acc:0.1285 ===\n",
      "train loss:2.283117469783994\n",
      "train loss:2.294188794303426\n",
      "train loss:2.2991622779835534\n",
      "=== epoch:14, train acc:0.14666666666666667, test acc:0.1311 ===\n",
      "train loss:2.2938177243230817\n",
      "train loss:2.292581838183246\n",
      "train loss:2.2922324238351743\n",
      "=== epoch:15, train acc:0.15, test acc:0.133 ===\n",
      "train loss:2.296740091765231\n",
      "train loss:2.2829731988923485\n",
      "train loss:2.287927914142226\n",
      "=== epoch:16, train acc:0.15666666666666668, test acc:0.135 ===\n",
      "train loss:2.2936505634091238\n",
      "train loss:2.289811887790259\n",
      "train loss:2.2903162431135216\n",
      "=== epoch:17, train acc:0.16, test acc:0.138 ===\n",
      "train loss:2.2802760168530996\n",
      "train loss:2.2860105166950366\n",
      "train loss:2.2802925542800865\n",
      "=== epoch:18, train acc:0.16333333333333333, test acc:0.14 ===\n",
      "train loss:2.3026023211830324\n",
      "train loss:2.2810694449691273\n",
      "train loss:2.2830472338922805\n",
      "=== epoch:19, train acc:0.16, test acc:0.1425 ===\n",
      "train loss:2.289077013780289\n",
      "train loss:2.2875380297452015\n",
      "train loss:2.291173038270291\n",
      "=== epoch:20, train acc:0.17333333333333334, test acc:0.1457 ===\n",
      "train loss:2.284365515459418\n",
      "train loss:2.287938403335496\n",
      "train loss:2.2879100418070495\n",
      "=== epoch:21, train acc:0.17333333333333334, test acc:0.1505 ===\n",
      "train loss:2.2727949822238607\n",
      "train loss:2.277220007011094\n",
      "train loss:2.288543359848358\n",
      "=== epoch:22, train acc:0.18, test acc:0.1533 ===\n",
      "train loss:2.2877800251502762\n",
      "train loss:2.294194523699717\n",
      "train loss:2.2720740283575966\n",
      "=== epoch:23, train acc:0.17333333333333334, test acc:0.1557 ===\n",
      "train loss:2.282827555771313\n",
      "train loss:2.291612310783357\n",
      "train loss:2.279713840643862\n",
      "=== epoch:24, train acc:0.17666666666666667, test acc:0.1574 ===\n",
      "train loss:2.2796747524253655\n",
      "train loss:2.273563005046425\n",
      "train loss:2.2786691870165026\n",
      "=== epoch:25, train acc:0.18, test acc:0.1593 ===\n",
      "train loss:2.2815155083365015\n",
      "train loss:2.2808955183493165\n",
      "train loss:2.2796705181791777\n",
      "=== epoch:26, train acc:0.17666666666666667, test acc:0.1646 ===\n",
      "train loss:2.2801418220194005\n",
      "train loss:2.290999363496356\n",
      "train loss:2.278808534292977\n",
      "=== epoch:27, train acc:0.18333333333333332, test acc:0.1731 ===\n",
      "train loss:2.2873837250504763\n",
      "train loss:2.2836303559549296\n",
      "train loss:2.288643479154207\n",
      "=== epoch:28, train acc:0.18666666666666668, test acc:0.1736 ===\n",
      "train loss:2.27970412995153\n",
      "train loss:2.2715986121187233\n",
      "train loss:2.272028874054088\n",
      "=== epoch:29, train acc:0.19666666666666666, test acc:0.1707 ===\n",
      "train loss:2.2743738149825985\n",
      "train loss:2.2774196587195608\n",
      "train loss:2.2774113859069094\n",
      "=== epoch:30, train acc:0.19333333333333333, test acc:0.1718 ===\n",
      "train loss:2.276479880241483\n",
      "train loss:2.2856720005161817\n",
      "train loss:2.2712569083161527\n",
      "=== epoch:31, train acc:0.20333333333333334, test acc:0.1784 ===\n",
      "train loss:2.2731671870294052\n",
      "train loss:2.269360061953069\n",
      "train loss:2.2797120763115477\n",
      "=== epoch:32, train acc:0.20333333333333334, test acc:0.1728 ===\n",
      "train loss:2.2760063764397698\n",
      "train loss:2.270571742436695\n",
      "train loss:2.25678342546172\n",
      "=== epoch:33, train acc:0.20666666666666667, test acc:0.1749 ===\n",
      "train loss:2.2747803390254586\n",
      "train loss:2.273075610189027\n",
      "train loss:2.2671414420836893\n",
      "=== epoch:34, train acc:0.20333333333333334, test acc:0.1738 ===\n",
      "train loss:2.2655556699337924\n",
      "train loss:2.2778299292395827\n",
      "train loss:2.2634915438263943\n",
      "=== epoch:35, train acc:0.20333333333333334, test acc:0.173 ===\n",
      "train loss:2.273370332035687\n",
      "train loss:2.263015319264545\n",
      "train loss:2.2795416466804954\n",
      "=== epoch:36, train acc:0.21666666666666667, test acc:0.1769 ===\n",
      "train loss:2.277765562059763\n",
      "train loss:2.26216974459006\n",
      "train loss:2.268110403834927\n",
      "=== epoch:37, train acc:0.21666666666666667, test acc:0.1823 ===\n",
      "train loss:2.272795136089243\n",
      "train loss:2.2688743149093566\n",
      "train loss:2.2713931135774685\n",
      "=== epoch:38, train acc:0.22666666666666666, test acc:0.1878 ===\n",
      "train loss:2.261097848204337\n",
      "train loss:2.268456727392203\n",
      "train loss:2.2695607467828136\n",
      "=== epoch:39, train acc:0.23, test acc:0.1923 ===\n",
      "train loss:2.2591185586428084\n",
      "train loss:2.265363378756387\n",
      "train loss:2.2713374540598346\n",
      "=== epoch:40, train acc:0.24, test acc:0.1953 ===\n",
      "train loss:2.2743967913730967\n",
      "train loss:2.252193616561174\n",
      "train loss:2.271756305125626\n",
      "=== epoch:41, train acc:0.24333333333333335, test acc:0.1986 ===\n",
      "train loss:2.2666665726839867\n",
      "train loss:2.271703994525177\n",
      "train loss:2.2691965320976557\n",
      "=== epoch:42, train acc:0.23333333333333334, test acc:0.1961 ===\n",
      "train loss:2.265197356873885\n",
      "train loss:2.2568605407040563\n",
      "train loss:2.276638964714521\n",
      "=== epoch:43, train acc:0.24, test acc:0.2013 ===\n",
      "train loss:2.264777598554592\n",
      "train loss:2.2777729327441008\n",
      "train loss:2.248104299288967\n",
      "=== epoch:44, train acc:0.24, test acc:0.2008 ===\n",
      "train loss:2.267406976889717\n",
      "train loss:2.2777569001919287\n",
      "train loss:2.2794335602641693\n",
      "=== epoch:45, train acc:0.25, test acc:0.2093 ===\n",
      "train loss:2.264390684987072\n",
      "train loss:2.2775561632872634\n",
      "train loss:2.271574237005641\n",
      "=== epoch:46, train acc:0.25666666666666665, test acc:0.213 ===\n",
      "train loss:2.2620738236521003\n",
      "train loss:2.251038822885855\n",
      "train loss:2.266808154833347\n",
      "=== epoch:47, train acc:0.2633333333333333, test acc:0.2146 ===\n",
      "train loss:2.250713889573957\n",
      "train loss:2.2474316652328916\n",
      "train loss:2.253052024097263\n",
      "=== epoch:48, train acc:0.2866666666666667, test acc:0.2229 ===\n",
      "train loss:2.249326809523824\n",
      "train loss:2.254141004672343\n",
      "train loss:2.2604954221228306\n",
      "=== epoch:49, train acc:0.2966666666666667, test acc:0.2273 ===\n",
      "train loss:2.258508364976658\n",
      "train loss:2.262900836604524\n",
      "train loss:2.2683356104091783\n",
      "=== epoch:50, train acc:0.28, test acc:0.2296 ===\n",
      "train loss:2.2473984222148897\n",
      "train loss:2.2585412280362327\n",
      "train loss:2.2410750432071422\n",
      "=== epoch:51, train acc:0.29333333333333333, test acc:0.2251 ===\n",
      "train loss:2.2525280418247697\n",
      "train loss:2.2620614567961974\n",
      "train loss:2.2625482179752496\n",
      "=== epoch:52, train acc:0.2966666666666667, test acc:0.2274 ===\n",
      "train loss:2.244891234334391\n",
      "train loss:2.2556098644933598\n",
      "train loss:2.241503615100711\n",
      "=== epoch:53, train acc:0.27666666666666667, test acc:0.2268 ===\n",
      "train loss:2.2761858484822644\n",
      "train loss:2.237776639613375\n",
      "train loss:2.2403054498459753\n",
      "=== epoch:54, train acc:0.27666666666666667, test acc:0.2297 ===\n",
      "train loss:2.2454033626919894\n",
      "train loss:2.244084473754702\n",
      "train loss:2.242211099248243\n",
      "=== epoch:55, train acc:0.2866666666666667, test acc:0.2331 ===\n",
      "train loss:2.2574331933733927\n",
      "train loss:2.24895155514276\n",
      "train loss:2.2639550598372686\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "=== epoch:56, train acc:0.27666666666666667, test acc:0.2346 ===\n",
      "train loss:2.2540982893355905\n",
      "train loss:2.2523575474401856\n",
      "train loss:2.2382248665400364\n",
      "=== epoch:57, train acc:0.2833333333333333, test acc:0.2294 ===\n",
      "train loss:2.2680256659301925\n",
      "train loss:2.25106843490167\n",
      "train loss:2.246856310448618\n",
      "=== epoch:58, train acc:0.2833333333333333, test acc:0.235 ===\n",
      "train loss:2.2208335723115753\n",
      "train loss:2.247160006058844\n",
      "train loss:2.233602159055075\n",
      "=== epoch:59, train acc:0.2866666666666667, test acc:0.236 ===\n",
      "train loss:2.2527706932962626\n",
      "train loss:2.2444268530140667\n",
      "train loss:2.275639557917651\n",
      "=== epoch:60, train acc:0.2866666666666667, test acc:0.2344 ===\n",
      "train loss:2.238042764591466\n",
      "train loss:2.2530785116411187\n",
      "train loss:2.2471487005124753\n",
      "=== epoch:61, train acc:0.2866666666666667, test acc:0.2401 ===\n",
      "train loss:2.2519439456511083\n",
      "train loss:2.2227210095272674\n",
      "train loss:2.240754767804656\n",
      "=== epoch:62, train acc:0.2966666666666667, test acc:0.243 ===\n",
      "train loss:2.2555776539156005\n",
      "train loss:2.2313675926510244\n",
      "train loss:2.23284565191659\n",
      "=== epoch:63, train acc:0.29333333333333333, test acc:0.2448 ===\n",
      "train loss:2.234647017032898\n",
      "train loss:2.2357726403491993\n",
      "train loss:2.237490004056358\n",
      "=== epoch:64, train acc:0.2866666666666667, test acc:0.2456 ===\n",
      "train loss:2.2464893423602184\n",
      "train loss:2.232264766109565\n",
      "train loss:2.2268646578882705\n",
      "=== epoch:65, train acc:0.2866666666666667, test acc:0.245 ===\n",
      "train loss:2.225311124787163\n",
      "train loss:2.2730438867869616\n",
      "train loss:2.2301504169368633\n",
      "=== epoch:66, train acc:0.2866666666666667, test acc:0.2479 ===\n",
      "train loss:2.2381190914120994\n",
      "train loss:2.245626048776254\n",
      "train loss:2.260812816144218\n",
      "=== epoch:67, train acc:0.3, test acc:0.2543 ===\n",
      "train loss:2.258652703549779\n",
      "train loss:2.2343236005073637\n",
      "train loss:2.232451878624261\n",
      "=== epoch:68, train acc:0.29333333333333333, test acc:0.2531 ===\n",
      "train loss:2.2378062965080856\n",
      "train loss:2.2466645133733443\n",
      "train loss:2.2180920410979517\n",
      "=== epoch:69, train acc:0.3, test acc:0.2536 ===\n",
      "train loss:2.238447755648657\n",
      "train loss:2.230542416189077\n",
      "train loss:2.2681925230048345\n",
      "=== epoch:70, train acc:0.31, test acc:0.2601 ===\n",
      "train loss:2.2277158535195367\n",
      "train loss:2.237227030971297\n",
      "train loss:2.210411553023221\n",
      "=== epoch:71, train acc:0.31, test acc:0.2621 ===\n",
      "train loss:2.2320607446747394\n",
      "train loss:2.2346064593518484\n",
      "train loss:2.252362845293699\n",
      "=== epoch:72, train acc:0.31333333333333335, test acc:0.2637 ===\n",
      "train loss:2.2369223217102974\n",
      "train loss:2.242998776244173\n",
      "train loss:2.2259673658619175\n",
      "=== epoch:73, train acc:0.31333333333333335, test acc:0.2652 ===\n",
      "train loss:2.2256663681034574\n",
      "train loss:2.232239443918821\n",
      "train loss:2.2024472891127336\n",
      "=== epoch:74, train acc:0.31, test acc:0.2606 ===\n",
      "train loss:2.2151789083816276\n",
      "train loss:2.209503470586021\n",
      "train loss:2.202365584496756\n",
      "=== epoch:75, train acc:0.31, test acc:0.2598 ===\n",
      "train loss:2.225832571313714\n",
      "train loss:2.234320165158971\n",
      "train loss:2.2416936709752098\n",
      "=== epoch:76, train acc:0.31666666666666665, test acc:0.2618 ===\n",
      "train loss:2.2402449009743903\n",
      "train loss:2.2232561606528405\n",
      "train loss:2.2681355370019545\n",
      "=== epoch:77, train acc:0.32, test acc:0.2674 ===\n",
      "train loss:2.2311833794951865\n",
      "train loss:2.22563167282272\n",
      "train loss:2.232472231789971\n",
      "=== epoch:78, train acc:0.3233333333333333, test acc:0.2675 ===\n",
      "train loss:2.235403422094662\n",
      "train loss:2.206416014841308\n",
      "train loss:2.222626974906757\n",
      "=== epoch:79, train acc:0.31666666666666665, test acc:0.2688 ===\n",
      "train loss:2.193193854720276\n",
      "train loss:2.226237190031903\n",
      "train loss:2.243233714385899\n",
      "=== epoch:80, train acc:0.3233333333333333, test acc:0.2671 ===\n",
      "train loss:2.222121696811425\n",
      "train loss:2.196131968775633\n",
      "train loss:2.232236233787413\n",
      "=== epoch:81, train acc:0.32, test acc:0.2692 ===\n",
      "train loss:2.2121800566997876\n",
      "train loss:2.233914497183375\n",
      "train loss:2.223757078123282\n",
      "=== epoch:82, train acc:0.32666666666666666, test acc:0.2722 ===\n",
      "train loss:2.2239091157757263\n",
      "train loss:2.229609040409346\n",
      "train loss:2.2330042875828666\n",
      "=== epoch:83, train acc:0.33, test acc:0.2778 ===\n",
      "train loss:2.2243916578893037\n",
      "train loss:2.205058681368256\n",
      "train loss:2.21842976872871\n",
      "=== epoch:84, train acc:0.33, test acc:0.2794 ===\n",
      "train loss:2.2194022101615247\n",
      "train loss:2.2384615586939995\n",
      "train loss:2.245740923884487\n",
      "=== epoch:85, train acc:0.33, test acc:0.2836 ===\n",
      "train loss:2.222889601848725\n",
      "train loss:2.2153006174083476\n",
      "train loss:2.207063801992222\n",
      "=== epoch:86, train acc:0.3433333333333333, test acc:0.2892 ===\n",
      "train loss:2.2067444909615848\n",
      "train loss:2.2209625992188995\n",
      "train loss:2.2264961560830305\n",
      "=== epoch:87, train acc:0.35333333333333333, test acc:0.2927 ===\n",
      "train loss:2.1932376920031533\n",
      "train loss:2.223740695921288\n",
      "train loss:2.2184951196806932\n",
      "=== epoch:88, train acc:0.35, test acc:0.2962 ===\n",
      "train loss:2.2035151717143977\n",
      "train loss:2.2054516090704976\n",
      "train loss:2.2046546906591598\n",
      "=== epoch:89, train acc:0.3466666666666667, test acc:0.2939 ===\n",
      "train loss:2.205769653379846\n",
      "train loss:2.206529646696508\n",
      "train loss:2.203009952205241\n",
      "=== epoch:90, train acc:0.34, test acc:0.2892 ===\n",
      "train loss:2.2195070658279414\n",
      "train loss:2.1945320814143012\n",
      "train loss:2.206638563237349\n",
      "=== epoch:91, train acc:0.3566666666666667, test acc:0.2941 ===\n",
      "train loss:2.1782875925324956\n",
      "train loss:2.2046282096474665\n",
      "train loss:2.197133851419786\n",
      "=== epoch:92, train acc:0.3466666666666667, test acc:0.2914 ===\n",
      "train loss:2.219965210895963\n",
      "train loss:2.204402863312795\n",
      "train loss:2.201826336999334\n",
      "=== epoch:93, train acc:0.35, test acc:0.2888 ===\n",
      "train loss:2.183965304879573\n",
      "train loss:2.1781681307606964\n",
      "train loss:2.1983654799474666\n",
      "=== epoch:94, train acc:0.33666666666666667, test acc:0.2874 ===\n",
      "train loss:2.204374572897437\n",
      "train loss:2.2116018516702423\n",
      "train loss:2.1858182385516134\n",
      "=== epoch:95, train acc:0.3433333333333333, test acc:0.2888 ===\n",
      "train loss:2.1733492678240167\n",
      "train loss:2.210158427784488\n",
      "train loss:2.1806150907745896\n",
      "=== epoch:96, train acc:0.3433333333333333, test acc:0.2883 ===\n",
      "train loss:2.2045325144565733\n",
      "train loss:2.1723312660834644\n",
      "train loss:2.190689674703287\n",
      "=== epoch:97, train acc:0.34, test acc:0.2881 ===\n",
      "train loss:2.1774318724566326\n",
      "train loss:2.1719177827487886\n",
      "train loss:2.1932210809623163\n",
      "=== epoch:98, train acc:0.3466666666666667, test acc:0.2909 ===\n",
      "train loss:2.1682833680105302\n",
      "train loss:2.1702815665750355\n",
      "train loss:2.2025045832125194\n",
      "=== epoch:99, train acc:0.3433333333333333, test acc:0.2897 ===\n",
      "train loss:2.205752745757883\n",
      "train loss:2.2151741304718646\n",
      "train loss:2.1886363196209637\n",
      "=== epoch:100, train acc:0.34, test acc:0.2888 ===\n",
      "train loss:2.149459629538387\n",
      "train loss:2.1677313908389766\n",
      "train loss:2.1848882084230725\n",
      "=== epoch:101, train acc:0.32666666666666666, test acc:0.285 ===\n",
      "train loss:2.2010588772168638\n",
      "train loss:2.123156279530495\n",
      "train loss:2.1703419813710836\n",
      "=== epoch:102, train acc:0.32666666666666666, test acc:0.2806 ===\n",
      "train loss:2.1780635394023147\n",
      "train loss:2.186652868549929\n",
      "train loss:2.195927186998315\n",
      "=== epoch:103, train acc:0.33666666666666667, test acc:0.2868 ===\n",
      "train loss:2.1919976584762373\n",
      "train loss:2.1750439990272743\n",
      "train loss:2.1724988855298\n",
      "=== epoch:104, train acc:0.3433333333333333, test acc:0.2874 ===\n",
      "train loss:2.1297813069203486\n",
      "train loss:2.14758122293084\n",
      "train loss:2.1511387843959238\n",
      "=== epoch:105, train acc:0.33666666666666667, test acc:0.2836 ===\n",
      "train loss:2.1885531990778113\n",
      "train loss:2.1898448585848835\n",
      "train loss:2.1695727813283856\n",
      "=== epoch:106, train acc:0.3466666666666667, test acc:0.2907 ===\n",
      "train loss:2.1762794321123966\n",
      "train loss:2.155674350133807\n",
      "train loss:2.1744835758106387\n",
      "=== epoch:107, train acc:0.34, test acc:0.2927 ===\n",
      "train loss:2.193455043723022\n",
      "train loss:2.193039210194324\n",
      "train loss:2.167388323350167\n",
      "=== epoch:108, train acc:0.35333333333333333, test acc:0.2986 ===\n",
      "train loss:2.1642783889965402\n",
      "train loss:2.15653673811631\n",
      "train loss:2.149895964091553\n",
      "=== epoch:109, train acc:0.35333333333333333, test acc:0.2983 ===\n",
      "train loss:2.201339624993566\n",
      "train loss:2.187521076431666\n",
      "train loss:2.18211876221541\n",
      "=== epoch:110, train acc:0.3566666666666667, test acc:0.3027 ===\n",
      "train loss:2.1308500476064243\n",
      "train loss:2.1880622454698115\n",
      "train loss:2.144424157541757\n",
      "=== epoch:111, train acc:0.36333333333333334, test acc:0.3039 ===\n",
      "train loss:2.133665611999049\n",
      "train loss:2.148384591254537\n",
      "train loss:2.1446439599988456\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "=== epoch:112, train acc:0.36666666666666664, test acc:0.3035 ===\n",
      "train loss:2.093672023480318\n",
      "train loss:2.182533261989657\n",
      "train loss:2.142059259377817\n",
      "=== epoch:113, train acc:0.3566666666666667, test acc:0.2997 ===\n",
      "train loss:2.1910712892464748\n",
      "train loss:2.136120359652369\n",
      "train loss:2.149411955162862\n",
      "=== epoch:114, train acc:0.36333333333333334, test acc:0.3026 ===\n",
      "train loss:2.130463106018682\n",
      "train loss:2.1456786289268526\n",
      "train loss:2.1331480738086936\n",
      "=== epoch:115, train acc:0.36333333333333334, test acc:0.303 ===\n",
      "train loss:2.1432157005892565\n",
      "train loss:2.180516082406969\n",
      "train loss:2.1705505080841294\n",
      "=== epoch:116, train acc:0.37, test acc:0.3057 ===\n",
      "train loss:2.1608279614214503\n",
      "train loss:2.158276306488213\n",
      "train loss:2.0984287042367655\n",
      "=== epoch:117, train acc:0.37, test acc:0.3084 ===\n",
      "train loss:2.0860976064463266\n",
      "train loss:2.1258847920971444\n",
      "train loss:2.156240298208153\n",
      "=== epoch:118, train acc:0.36, test acc:0.3031 ===\n",
      "train loss:2.1230979563973587\n",
      "train loss:2.153643118672767\n",
      "train loss:2.1330703284938615\n",
      "=== epoch:119, train acc:0.36, test acc:0.3034 ===\n",
      "train loss:2.119511034561465\n",
      "train loss:2.086453744624581\n",
      "train loss:2.171516018959032\n",
      "=== epoch:120, train acc:0.3566666666666667, test acc:0.304 ===\n",
      "train loss:2.151020401560259\n",
      "train loss:2.1301296272344743\n",
      "train loss:2.16180342417784\n",
      "=== epoch:121, train acc:0.36333333333333334, test acc:0.3085 ===\n",
      "train loss:2.074191820987335\n",
      "train loss:2.1279907798657907\n",
      "train loss:2.1238711650222357\n",
      "=== epoch:122, train acc:0.36, test acc:0.3082 ===\n",
      "train loss:2.146936114620249\n",
      "train loss:2.125003864769242\n",
      "train loss:2.1660674771538417\n",
      "=== epoch:123, train acc:0.37, test acc:0.3165 ===\n",
      "train loss:2.1353817945860856\n",
      "train loss:2.0484401914658137\n",
      "train loss:2.1550454774877865\n",
      "=== epoch:124, train acc:0.38, test acc:0.322 ===\n",
      "train loss:2.1174758354824452\n",
      "train loss:2.1430215551299048\n",
      "train loss:2.1286273517337175\n",
      "=== epoch:125, train acc:0.39, test acc:0.323 ===\n",
      "train loss:2.060481096927207\n",
      "train loss:2.120527115495384\n",
      "train loss:2.077877364747489\n",
      "=== epoch:126, train acc:0.37, test acc:0.3197 ===\n",
      "train loss:2.109535904503951\n",
      "train loss:2.0947606657349755\n",
      "train loss:2.1014384981090997\n",
      "=== epoch:127, train acc:0.37333333333333335, test acc:0.3208 ===\n",
      "train loss:2.1168963571452584\n",
      "train loss:2.0786189794771106\n",
      "train loss:2.1162408059215494\n",
      "=== epoch:128, train acc:0.37333333333333335, test acc:0.3217 ===\n",
      "train loss:2.130590169342494\n",
      "train loss:2.146348250241871\n",
      "train loss:2.1431019309943657\n",
      "=== epoch:129, train acc:0.39666666666666667, test acc:0.333 ===\n",
      "train loss:2.102715988821611\n",
      "train loss:2.116041225963425\n",
      "train loss:2.1146125418185817\n",
      "=== epoch:130, train acc:0.4033333333333333, test acc:0.3376 ===\n",
      "train loss:2.1502990128262036\n",
      "train loss:2.1062687725673\n",
      "train loss:2.1285429571209002\n",
      "=== epoch:131, train acc:0.4033333333333333, test acc:0.3401 ===\n",
      "train loss:2.0610920521548994\n",
      "train loss:2.1019222391509182\n",
      "train loss:2.091327433661026\n",
      "=== epoch:132, train acc:0.3933333333333333, test acc:0.3339 ===\n",
      "train loss:2.1098374853126707\n",
      "train loss:2.1562415779335846\n",
      "train loss:2.063775018667529\n",
      "=== epoch:133, train acc:0.39666666666666667, test acc:0.3385 ===\n",
      "train loss:2.091891548338868\n",
      "train loss:2.0936007961617755\n",
      "train loss:2.1094385180102737\n",
      "=== epoch:134, train acc:0.4033333333333333, test acc:0.3452 ===\n",
      "train loss:2.0904013821489076\n",
      "train loss:2.084659583882458\n",
      "train loss:2.076267926866624\n",
      "=== epoch:135, train acc:0.4033333333333333, test acc:0.3455 ===\n",
      "train loss:2.133569213821803\n",
      "train loss:2.1025026283636934\n",
      "train loss:2.10482891644319\n",
      "=== epoch:136, train acc:0.41, test acc:0.3496 ===\n",
      "train loss:2.0911025763372817\n",
      "train loss:2.0749707716159675\n",
      "train loss:2.094676285245637\n",
      "=== epoch:137, train acc:0.41, test acc:0.3493 ===\n",
      "train loss:2.0820079069354747\n",
      "train loss:2.0683518316002742\n",
      "train loss:2.040898546770586\n",
      "=== epoch:138, train acc:0.4166666666666667, test acc:0.3475 ===\n",
      "train loss:2.05549666310735\n",
      "train loss:2.103363438369761\n",
      "train loss:2.1039169217758844\n",
      "=== epoch:139, train acc:0.4166666666666667, test acc:0.3518 ===\n",
      "train loss:2.117605697891185\n",
      "train loss:2.1173765529586643\n",
      "train loss:2.0814807377464764\n",
      "=== epoch:140, train acc:0.43333333333333335, test acc:0.3598 ===\n",
      "train loss:2.023986858091332\n",
      "train loss:2.06800696654914\n",
      "train loss:2.123130789930023\n",
      "=== epoch:141, train acc:0.44666666666666666, test acc:0.3634 ===\n",
      "train loss:2.0486362894164842\n",
      "train loss:2.0871786122908462\n",
      "train loss:2.0223281260166925\n",
      "=== epoch:142, train acc:0.44666666666666666, test acc:0.365 ===\n",
      "train loss:2.0758193578103006\n",
      "train loss:2.0543206406106065\n",
      "train loss:2.0780738515659145\n",
      "=== epoch:143, train acc:0.44666666666666666, test acc:0.3668 ===\n",
      "train loss:2.0437634619213294\n",
      "train loss:2.0363782519677067\n",
      "train loss:2.0161513852098385\n",
      "=== epoch:144, train acc:0.44666666666666666, test acc:0.3692 ===\n",
      "train loss:2.069212590789773\n",
      "train loss:2.0537065719737386\n",
      "train loss:2.0878525554993876\n",
      "=== epoch:145, train acc:0.45666666666666667, test acc:0.3688 ===\n",
      "train loss:2.088304622348351\n",
      "train loss:2.075264108774885\n",
      "train loss:2.0315867217421917\n",
      "=== epoch:146, train acc:0.4533333333333333, test acc:0.3718 ===\n",
      "train loss:2.0031304449606906\n",
      "train loss:2.104436323046336\n",
      "train loss:2.0672095585306973\n",
      "=== epoch:147, train acc:0.45666666666666667, test acc:0.374 ===\n",
      "train loss:2.142813746001985\n",
      "train loss:2.036919534000181\n",
      "train loss:1.9720686608754745\n",
      "=== epoch:148, train acc:0.4766666666666667, test acc:0.3769 ===\n",
      "train loss:2.1053264364009174\n",
      "train loss:2.060613678215139\n",
      "train loss:1.9487353438951809\n",
      "=== epoch:149, train acc:0.4666666666666667, test acc:0.3738 ===\n",
      "train loss:2.06981648391593\n",
      "train loss:2.0456633655881866\n",
      "train loss:2.073866873870371\n",
      "=== epoch:150, train acc:0.47, test acc:0.377 ===\n",
      "train loss:2.024324521997301\n",
      "train loss:2.045920940147481\n",
      "train loss:2.070308748514818\n",
      "=== epoch:151, train acc:0.4766666666666667, test acc:0.3819 ===\n",
      "train loss:2.0577371319644753\n",
      "train loss:2.0427680602126195\n",
      "train loss:2.0396424207843586\n",
      "=== epoch:152, train acc:0.48, test acc:0.3871 ===\n",
      "train loss:2.0088756750362413\n",
      "train loss:2.0689472034148353\n",
      "train loss:2.0284451466091062\n",
      "=== epoch:153, train acc:0.48333333333333334, test acc:0.3853 ===\n",
      "train loss:2.067882811169831\n",
      "train loss:1.9719834333078288\n",
      "train loss:2.0618920797088505\n",
      "=== epoch:154, train acc:0.4866666666666667, test acc:0.3858 ===\n",
      "train loss:2.0392377584373635\n",
      "train loss:2.072626220058682\n",
      "train loss:2.0900314694433466\n",
      "=== epoch:155, train acc:0.49666666666666665, test acc:0.3879 ===\n",
      "train loss:2.073023827451628\n",
      "train loss:2.0216932080976564\n",
      "train loss:1.9815531646400104\n",
      "=== epoch:156, train acc:0.5, test acc:0.3906 ===\n",
      "train loss:2.0556048364129715\n",
      "train loss:2.0695501412936204\n",
      "train loss:2.0835477106019256\n",
      "=== epoch:157, train acc:0.49666666666666665, test acc:0.3967 ===\n",
      "train loss:2.05241298565474\n",
      "train loss:2.027215251741732\n",
      "train loss:2.019239179771079\n",
      "=== epoch:158, train acc:0.5, test acc:0.4002 ===\n",
      "train loss:1.952590879715675\n",
      "train loss:1.9405008068285914\n",
      "train loss:2.026106841717797\n",
      "=== epoch:159, train acc:0.5, test acc:0.3985 ===\n",
      "train loss:2.032003570594238\n",
      "train loss:2.0318274842730504\n",
      "train loss:1.9953318443527488\n",
      "=== epoch:160, train acc:0.5066666666666667, test acc:0.4023 ===\n",
      "train loss:1.9694556890147814\n",
      "train loss:1.928274237695911\n",
      "train loss:1.912108483331319\n",
      "=== epoch:161, train acc:0.5033333333333333, test acc:0.4005 ===\n",
      "train loss:1.9734585686449568\n",
      "train loss:2.014584444505148\n",
      "train loss:2.062048085347076\n",
      "=== epoch:162, train acc:0.51, test acc:0.4032 ===\n",
      "train loss:2.005646930437524\n",
      "train loss:1.9832835535743885\n",
      "train loss:2.0184412511761773\n",
      "=== epoch:163, train acc:0.5133333333333333, test acc:0.4083 ===\n",
      "train loss:1.9577104841015938\n",
      "train loss:1.9897304944098644\n",
      "train loss:1.9777590883294751\n",
      "=== epoch:164, train acc:0.52, test acc:0.4057 ===\n",
      "train loss:1.9534270612700968\n",
      "train loss:1.958226490256217\n",
      "train loss:1.9571077257206235\n",
      "=== epoch:165, train acc:0.5233333333333333, test acc:0.4081 ===\n",
      "train loss:2.0700683100285575\n",
      "train loss:1.9732138558864818\n",
      "train loss:2.00881799904254\n",
      "=== epoch:166, train acc:0.5233333333333333, test acc:0.4092 ===\n",
      "train loss:2.0052559023105623\n",
      "train loss:2.0585975599069646\n",
      "train loss:2.0048327786587357\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "=== epoch:167, train acc:0.5333333333333333, test acc:0.4117 ===\n",
      "train loss:1.8807294669335914\n",
      "train loss:1.973508411560971\n",
      "train loss:2.0354096850637204\n",
      "=== epoch:168, train acc:0.5333333333333333, test acc:0.4135 ===\n",
      "train loss:1.8900315108070425\n",
      "train loss:1.9307757252088293\n",
      "train loss:2.009960131032277\n",
      "=== epoch:169, train acc:0.5366666666666666, test acc:0.41 ===\n",
      "train loss:1.9780708691721203\n",
      "train loss:1.8538066412003302\n",
      "train loss:1.9142305004006122\n",
      "=== epoch:170, train acc:0.53, test acc:0.4101 ===\n",
      "train loss:1.995673646520411\n",
      "train loss:1.8956134452539315\n",
      "train loss:1.956321165307772\n",
      "=== epoch:171, train acc:0.5233333333333333, test acc:0.4124 ===\n",
      "train loss:1.8634261750305305\n",
      "train loss:1.9587078042977177\n",
      "train loss:1.9150976216606697\n",
      "=== epoch:172, train acc:0.52, test acc:0.4097 ===\n",
      "train loss:1.980279784453942\n",
      "train loss:1.903121153671874\n",
      "train loss:1.976911637009236\n",
      "=== epoch:173, train acc:0.5266666666666666, test acc:0.4081 ===\n",
      "train loss:1.9165134613035135\n",
      "train loss:1.9905734224920915\n",
      "train loss:2.019531823400102\n",
      "=== epoch:174, train acc:0.5233333333333333, test acc:0.4053 ===\n",
      "train loss:1.86646290542105\n",
      "train loss:1.928252804627669\n",
      "train loss:1.9640586329961889\n",
      "=== epoch:175, train acc:0.5266666666666666, test acc:0.4052 ===\n",
      "train loss:1.9021010031003658\n",
      "train loss:1.9981346203022392\n",
      "train loss:1.9549462702793894\n",
      "=== epoch:176, train acc:0.5233333333333333, test acc:0.4118 ===\n",
      "train loss:1.9411817856745768\n",
      "train loss:1.9744510391914205\n",
      "train loss:1.9040870460359551\n",
      "=== epoch:177, train acc:0.5233333333333333, test acc:0.4157 ===\n",
      "train loss:1.9360038828261354\n",
      "train loss:1.900854836556505\n",
      "train loss:1.9013715331570773\n",
      "=== epoch:178, train acc:0.5333333333333333, test acc:0.4162 ===\n",
      "train loss:1.9573310293435038\n",
      "train loss:1.8778385330825083\n",
      "train loss:1.9253916421726331\n",
      "=== epoch:179, train acc:0.5333333333333333, test acc:0.4154 ===\n",
      "train loss:1.895729573459572\n",
      "train loss:1.975277002849779\n",
      "train loss:1.918230244492072\n",
      "=== epoch:180, train acc:0.5333333333333333, test acc:0.4192 ===\n",
      "train loss:1.9788567019402747\n",
      "train loss:1.9134117796216907\n",
      "train loss:1.7694250096682067\n",
      "=== epoch:181, train acc:0.5333333333333333, test acc:0.4189 ===\n",
      "train loss:1.8763148362560478\n",
      "train loss:1.9044789743288737\n",
      "train loss:1.9228625811755313\n",
      "=== epoch:182, train acc:0.5466666666666666, test acc:0.4199 ===\n",
      "train loss:1.917520295932201\n",
      "train loss:1.922714548028786\n",
      "train loss:1.938752495823576\n",
      "=== epoch:183, train acc:0.5433333333333333, test acc:0.4239 ===\n",
      "train loss:1.8822481703486584\n",
      "train loss:1.9566448075137608\n",
      "train loss:1.9968845235555415\n",
      "=== epoch:184, train acc:0.5366666666666666, test acc:0.4244 ===\n",
      "train loss:1.9784744188034171\n",
      "train loss:1.7792537940021393\n",
      "train loss:2.0278074027351165\n",
      "=== epoch:185, train acc:0.5366666666666666, test acc:0.4253 ===\n",
      "train loss:1.9026219527228805\n",
      "train loss:1.817581301769838\n",
      "train loss:1.8741446092770289\n",
      "=== epoch:186, train acc:0.5533333333333333, test acc:0.4242 ===\n",
      "train loss:1.9126201271096488\n",
      "train loss:1.9722580477467055\n",
      "train loss:1.9968982107024578\n",
      "=== epoch:187, train acc:0.55, test acc:0.4259 ===\n",
      "train loss:1.932307978672737\n",
      "train loss:1.9274353814178946\n",
      "train loss:1.9580985491645575\n",
      "=== epoch:188, train acc:0.5533333333333333, test acc:0.4251 ===\n",
      "train loss:1.8940458732249454\n",
      "train loss:1.8495404283812278\n",
      "train loss:1.8353286387758163\n",
      "=== epoch:189, train acc:0.5533333333333333, test acc:0.4221 ===\n",
      "train loss:1.8801305658294905\n",
      "train loss:1.7105020925857244\n",
      "train loss:1.8650003561585482\n",
      "=== epoch:190, train acc:0.55, test acc:0.4192 ===\n",
      "train loss:1.8804823712699466\n",
      "train loss:1.9430675876923125\n",
      "train loss:1.8267522241718823\n",
      "=== epoch:191, train acc:0.5633333333333334, test acc:0.4204 ===\n",
      "train loss:1.8397443070847521\n",
      "train loss:1.7752826521029184\n",
      "train loss:1.8855931588927677\n",
      "=== epoch:192, train acc:0.5633333333333334, test acc:0.4181 ===\n",
      "train loss:1.9554181379651867\n",
      "train loss:1.9242441437366766\n",
      "train loss:1.8811172201097321\n",
      "=== epoch:193, train acc:0.5533333333333333, test acc:0.4212 ===\n",
      "train loss:1.852963034477088\n",
      "train loss:1.871788528217204\n",
      "train loss:1.8406133692278075\n",
      "=== epoch:194, train acc:0.5566666666666666, test acc:0.4212 ===\n",
      "train loss:1.903275846046796\n",
      "train loss:1.7614998194373528\n",
      "train loss:1.8973461248741443\n",
      "=== epoch:195, train acc:0.5733333333333334, test acc:0.4235 ===\n",
      "train loss:1.8485305588854617\n",
      "train loss:1.888284126106751\n",
      "train loss:1.8557450778142155\n",
      "=== epoch:196, train acc:0.5766666666666667, test acc:0.4236 ===\n",
      "train loss:1.9187725911398414\n",
      "train loss:1.9153180799129992\n",
      "train loss:1.9422064691541245\n",
      "=== epoch:197, train acc:0.5733333333333334, test acc:0.4244 ===\n",
      "train loss:1.8010376944522966\n",
      "train loss:1.7326772974049218\n",
      "train loss:1.9035161764063215\n",
      "=== epoch:198, train acc:0.57, test acc:0.4193 ===\n",
      "train loss:1.9213085332804707\n",
      "train loss:1.7690357701756878\n",
      "train loss:1.8703104209634631\n",
      "=== epoch:199, train acc:0.57, test acc:0.4242 ===\n",
      "train loss:1.7562570675915752\n",
      "train loss:1.7992923486445553\n",
      "train loss:1.8218661968150842\n",
      "=== epoch:200, train acc:0.5733333333333334, test acc:0.4249 ===\n",
      "train loss:1.8361685950914666\n",
      "train loss:1.7815589700580636\n",
      "train loss:1.8806908476704407\n",
      "=== epoch:201, train acc:0.5633333333333334, test acc:0.4236 ===\n",
      "train loss:1.8272503331824823\n",
      "train loss:1.7306229263407413\n",
      "train loss:1.7629658115035787\n",
      "=== epoch:202, train acc:0.5733333333333334, test acc:0.4259 ===\n",
      "train loss:1.7637113650997938\n",
      "train loss:1.758828116877958\n",
      "train loss:1.6709555003707228\n",
      "=== epoch:203, train acc:0.58, test acc:0.429 ===\n",
      "train loss:1.7824100516326564\n",
      "train loss:1.7651330326544648\n",
      "train loss:1.8382363749463042\n",
      "=== epoch:204, train acc:0.58, test acc:0.4301 ===\n",
      "train loss:1.8465828429579216\n",
      "train loss:1.737926834886906\n",
      "train loss:1.8293097931885531\n",
      "=== epoch:205, train acc:0.58, test acc:0.4331 ===\n",
      "train loss:1.8569771148254548\n",
      "train loss:1.792687919266071\n",
      "train loss:1.8405931095668173\n",
      "=== epoch:206, train acc:0.5833333333333334, test acc:0.4384 ===\n",
      "train loss:1.8913993120484063\n",
      "train loss:1.6873771901993573\n",
      "train loss:1.8118988398778257\n",
      "=== epoch:207, train acc:0.5833333333333334, test acc:0.4404 ===\n",
      "train loss:1.7666714396810057\n",
      "train loss:1.6275013334112582\n",
      "train loss:1.7357665464612029\n",
      "=== epoch:208, train acc:0.5766666666666667, test acc:0.4399 ===\n",
      "train loss:1.8057298101396164\n",
      "train loss:1.7258495290898122\n",
      "train loss:1.8444284019228772\n",
      "=== epoch:209, train acc:0.5833333333333334, test acc:0.4425 ===\n",
      "train loss:1.8423126107741348\n",
      "train loss:1.8215367023395541\n",
      "train loss:1.7275556334958222\n",
      "=== epoch:210, train acc:0.58, test acc:0.4445 ===\n",
      "train loss:1.7120219166323114\n",
      "train loss:1.6969730440949966\n",
      "train loss:1.8303657775665738\n",
      "=== epoch:211, train acc:0.5833333333333334, test acc:0.4451 ===\n",
      "train loss:1.75398928674155\n",
      "train loss:1.7199496603306503\n",
      "train loss:1.6619854250582535\n",
      "=== epoch:212, train acc:0.58, test acc:0.4444 ===\n",
      "train loss:1.782995444351213\n",
      "train loss:1.5978860728251607\n",
      "train loss:1.7485753464491032\n",
      "=== epoch:213, train acc:0.57, test acc:0.4416 ===\n",
      "train loss:1.7668162511254732\n",
      "train loss:1.778529808359382\n",
      "train loss:1.7161093406740158\n",
      "=== epoch:214, train acc:0.58, test acc:0.4428 ===\n",
      "train loss:1.7813740207564117\n",
      "train loss:1.7314072389480808\n",
      "train loss:1.698908393673251\n",
      "=== epoch:215, train acc:0.5833333333333334, test acc:0.4444 ===\n",
      "train loss:1.6882909407904143\n",
      "train loss:1.7557293632495519\n",
      "train loss:1.7413452073888078\n",
      "=== epoch:216, train acc:0.5866666666666667, test acc:0.4437 ===\n",
      "train loss:1.6967541242479387\n",
      "train loss:1.7029829475451348\n",
      "train loss:1.846091306383363\n",
      "=== epoch:217, train acc:0.5933333333333334, test acc:0.4469 ===\n",
      "train loss:1.7019004537207172\n",
      "train loss:1.7988723957518347\n",
      "train loss:1.7262717949176531\n",
      "=== epoch:218, train acc:0.5966666666666667, test acc:0.4484 ===\n",
      "train loss:1.6579020724900744\n",
      "train loss:1.6399654516285025\n",
      "train loss:1.6647730561279888\n",
      "=== epoch:219, train acc:0.59, test acc:0.4468 ===\n",
      "train loss:1.6148159788568341\n",
      "train loss:1.6662191232721497\n",
      "train loss:1.7737089899550935\n",
      "=== epoch:220, train acc:0.5866666666666667, test acc:0.4457 ===\n",
      "train loss:1.752727143688017\n",
      "train loss:1.68530839692251\n",
      "train loss:1.7742356907784387\n",
      "=== epoch:221, train acc:0.5833333333333334, test acc:0.4473 ===\n",
      "train loss:1.7154477722128099\n",
      "train loss:1.7131420011456109\n",
      "train loss:1.7350247245250574\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "=== epoch:222, train acc:0.59, test acc:0.4488 ===\n",
      "train loss:1.6616899836534826\n",
      "train loss:1.7280742994804714\n",
      "train loss:1.7379793379590394\n",
      "=== epoch:223, train acc:0.5933333333333334, test acc:0.453 ===\n",
      "train loss:1.739521651233649\n",
      "train loss:1.6629568565295154\n",
      "train loss:1.7275880688654193\n",
      "=== epoch:224, train acc:0.5933333333333334, test acc:0.4578 ===\n",
      "train loss:1.7298314067357012\n",
      "train loss:1.6671879155520852\n",
      "train loss:1.623546772979323\n",
      "=== epoch:225, train acc:0.5966666666666667, test acc:0.4548 ===\n",
      "train loss:1.6598364738053208\n",
      "train loss:1.6073278684824055\n",
      "train loss:1.6970573527375257\n",
      "=== epoch:226, train acc:0.59, test acc:0.4531 ===\n",
      "train loss:1.7601016490007941\n",
      "train loss:1.7342836834950424\n",
      "train loss:1.5957049545585935\n",
      "=== epoch:227, train acc:0.5966666666666667, test acc:0.4544 ===\n",
      "train loss:1.6650547093929418\n",
      "train loss:1.8427239670695243\n",
      "train loss:1.633299521940792\n",
      "=== epoch:228, train acc:0.59, test acc:0.4587 ===\n",
      "train loss:1.7164912828433707\n",
      "train loss:1.640203779359606\n",
      "train loss:1.7066077586726356\n",
      "=== epoch:229, train acc:0.5933333333333334, test acc:0.4618 ===\n",
      "train loss:1.5406714888312931\n",
      "train loss:1.5562182509414237\n",
      "train loss:1.580319328010256\n",
      "=== epoch:230, train acc:0.5933333333333334, test acc:0.4618 ===\n",
      "train loss:1.700604832612424\n",
      "train loss:1.662919326439547\n",
      "train loss:1.6215868741865296\n",
      "=== epoch:231, train acc:0.5966666666666667, test acc:0.4638 ===\n",
      "train loss:1.737132038430078\n",
      "train loss:1.6258710816631001\n",
      "train loss:1.6967129387468947\n",
      "=== epoch:232, train acc:0.6033333333333334, test acc:0.4685 ===\n",
      "train loss:1.7800323881977347\n",
      "train loss:1.7402972411183533\n",
      "train loss:1.6227864598603414\n",
      "=== epoch:233, train acc:0.6066666666666667, test acc:0.4697 ===\n",
      "train loss:1.6906177922187176\n",
      "train loss:1.655815716725181\n",
      "train loss:1.5955080079128756\n",
      "=== epoch:234, train acc:0.6066666666666667, test acc:0.4685 ===\n",
      "train loss:1.6469911916437\n",
      "train loss:1.7990559342513979\n",
      "train loss:1.6575569219214321\n",
      "=== epoch:235, train acc:0.6033333333333334, test acc:0.4735 ===\n",
      "train loss:1.5709223135222652\n",
      "train loss:1.6249538238119885\n",
      "train loss:1.699832037534488\n",
      "=== epoch:236, train acc:0.6133333333333333, test acc:0.4779 ===\n",
      "train loss:1.6695947692531976\n",
      "train loss:1.7602891301009773\n",
      "train loss:1.7184388732910094\n",
      "=== epoch:237, train acc:0.6233333333333333, test acc:0.4818 ===\n",
      "train loss:1.6935316758863246\n",
      "train loss:1.6079809476136129\n",
      "train loss:1.7186015864901063\n",
      "=== epoch:238, train acc:0.6166666666666667, test acc:0.4792 ===\n",
      "train loss:1.6359760732260313\n",
      "train loss:1.6851565209910067\n",
      "train loss:1.648692830074021\n",
      "=== epoch:239, train acc:0.6266666666666667, test acc:0.4813 ===\n",
      "train loss:1.6396710801400893\n",
      "train loss:1.6144397591919626\n",
      "train loss:1.6244363475213683\n",
      "=== epoch:240, train acc:0.62, test acc:0.4815 ===\n",
      "train loss:1.6680910223774412\n",
      "train loss:1.542480644180567\n",
      "train loss:1.6141303340445736\n",
      "=== epoch:241, train acc:0.6233333333333333, test acc:0.4836 ===\n",
      "train loss:1.633653858501489\n",
      "train loss:1.635047888348194\n",
      "train loss:1.6515677095157684\n",
      "=== epoch:242, train acc:0.63, test acc:0.4868 ===\n",
      "train loss:1.6133719431954519\n",
      "train loss:1.6942020622295408\n",
      "train loss:1.4198863052453736\n",
      "=== epoch:243, train acc:0.6266666666666667, test acc:0.4873 ===\n",
      "train loss:1.5382218138524888\n",
      "train loss:1.598356363536624\n",
      "train loss:1.5414782895578265\n",
      "=== epoch:244, train acc:0.6266666666666667, test acc:0.4838 ===\n",
      "train loss:1.602803884581139\n",
      "train loss:1.6631415909844365\n",
      "train loss:1.4779108421071527\n",
      "=== epoch:245, train acc:0.6233333333333333, test acc:0.4832 ===\n",
      "train loss:1.617972383489422\n",
      "train loss:1.5394778170775631\n",
      "train loss:1.6799331155536408\n",
      "=== epoch:246, train acc:0.63, test acc:0.488 ===\n",
      "train loss:1.461059040893922\n",
      "train loss:1.5574348259182156\n",
      "train loss:1.5069239668442607\n",
      "=== epoch:247, train acc:0.6366666666666667, test acc:0.4906 ===\n",
      "train loss:1.487058758298833\n",
      "train loss:1.603408108694286\n",
      "train loss:1.5171905335913352\n",
      "=== epoch:248, train acc:0.64, test acc:0.491 ===\n",
      "train loss:1.4544651675763942\n",
      "train loss:1.6241311479806486\n",
      "train loss:1.5182529610372506\n",
      "=== epoch:249, train acc:0.64, test acc:0.4888 ===\n",
      "train loss:1.6222427006055964\n",
      "train loss:1.470098794515429\n",
      "train loss:1.5889139460303427\n",
      "=== epoch:250, train acc:0.65, test acc:0.4924 ===\n",
      "train loss:1.6694769422608573\n",
      "train loss:1.595815156254652\n",
      "train loss:1.6165564957784628\n",
      "=== epoch:251, train acc:0.65, test acc:0.4948 ===\n",
      "train loss:1.54393649331692\n",
      "train loss:1.6013601912343367\n",
      "train loss:1.48444702718377\n",
      "=== epoch:252, train acc:0.6466666666666666, test acc:0.499 ===\n",
      "train loss:1.49551954529353\n",
      "train loss:1.5914104377676244\n",
      "train loss:1.432649559801713\n",
      "=== epoch:253, train acc:0.6466666666666666, test acc:0.5001 ===\n",
      "train loss:1.531425992741849\n",
      "train loss:1.5668335189751872\n",
      "train loss:1.4836045042360817\n",
      "=== epoch:254, train acc:0.6366666666666667, test acc:0.5008 ===\n",
      "train loss:1.435451108165075\n",
      "train loss:1.4825356166745811\n",
      "train loss:1.44508062248743\n",
      "=== epoch:255, train acc:0.65, test acc:0.4992 ===\n",
      "train loss:1.4229305911333592\n",
      "train loss:1.502347202706883\n",
      "train loss:1.5527368131008266\n",
      "=== epoch:256, train acc:0.65, test acc:0.4996 ===\n",
      "train loss:1.6072596704927506\n",
      "train loss:1.5181559542431662\n",
      "train loss:1.5279078710212644\n",
      "=== epoch:257, train acc:0.65, test acc:0.5036 ===\n",
      "train loss:1.4964900708184825\n",
      "train loss:1.465837369570189\n",
      "train loss:1.4061118894559264\n",
      "=== epoch:258, train acc:0.65, test acc:0.5032 ===\n",
      "train loss:1.399070124979022\n",
      "train loss:1.4494689871677244\n",
      "train loss:1.3839572767737587\n",
      "=== epoch:259, train acc:0.65, test acc:0.5037 ===\n",
      "train loss:1.5338603550739953\n",
      "train loss:1.529393825596952\n",
      "train loss:1.3998585144858289\n",
      "=== epoch:260, train acc:0.65, test acc:0.5019 ===\n",
      "train loss:1.5537806840177582\n",
      "train loss:1.5244711998982503\n",
      "train loss:1.5655324845164755\n",
      "=== epoch:261, train acc:0.6433333333333333, test acc:0.5032 ===\n",
      "train loss:1.4962715752734903\n",
      "train loss:1.5448736001823782\n",
      "train loss:1.3534072043569365\n",
      "=== epoch:262, train acc:0.6466666666666666, test acc:0.5051 ===\n",
      "train loss:1.5273943650199384\n",
      "train loss:1.420433468519905\n",
      "train loss:1.3701594519086795\n",
      "=== epoch:263, train acc:0.6466666666666666, test acc:0.5051 ===\n",
      "train loss:1.498463775872804\n",
      "train loss:1.5132011383465653\n",
      "train loss:1.4800886471569012\n",
      "=== epoch:264, train acc:0.6466666666666666, test acc:0.5048 ===\n",
      "train loss:1.5223649325545756\n",
      "train loss:1.3002384509686384\n",
      "train loss:1.430872855502298\n",
      "=== epoch:265, train acc:0.6466666666666666, test acc:0.5044 ===\n",
      "train loss:1.5252141063168037\n",
      "train loss:1.3729904897083773\n",
      "train loss:1.4030414726003801\n",
      "=== epoch:266, train acc:0.6466666666666666, test acc:0.5045 ===\n",
      "train loss:1.5612909771953394\n",
      "train loss:1.4773415392547162\n",
      "train loss:1.4924414105931998\n",
      "=== epoch:267, train acc:0.65, test acc:0.5101 ===\n",
      "train loss:1.3371312091929155\n",
      "train loss:1.4909960883698892\n",
      "train loss:1.4355712137900374\n",
      "=== epoch:268, train acc:0.65, test acc:0.5112 ===\n",
      "train loss:1.3043206567663097\n",
      "train loss:1.479896195298916\n",
      "train loss:1.335004776712311\n",
      "=== epoch:269, train acc:0.65, test acc:0.5081 ===\n",
      "train loss:1.3657119730163472\n",
      "train loss:1.3871840509828501\n",
      "train loss:1.3985044500639798\n",
      "=== epoch:270, train acc:0.6533333333333333, test acc:0.5081 ===\n",
      "train loss:1.480257866055245\n",
      "train loss:1.419795976517421\n",
      "train loss:1.5568281612393937\n",
      "=== epoch:271, train acc:0.65, test acc:0.5106 ===\n",
      "train loss:1.379561680513456\n",
      "train loss:1.4039790505317398\n",
      "train loss:1.469817698877998\n",
      "=== epoch:272, train acc:0.6566666666666666, test acc:0.5121 ===\n",
      "train loss:1.393940651255641\n",
      "train loss:1.4913616088970532\n",
      "train loss:1.4131118137034815\n",
      "=== epoch:273, train acc:0.6566666666666666, test acc:0.5161 ===\n",
      "train loss:1.3935322989493073\n",
      "train loss:1.4729849065198306\n",
      "train loss:1.319746401030374\n",
      "=== epoch:274, train acc:0.6633333333333333, test acc:0.5168 ===\n",
      "train loss:1.2610788680248026\n",
      "train loss:1.4018036391718725\n",
      "train loss:1.4079013600451842\n",
      "=== epoch:275, train acc:0.6633333333333333, test acc:0.5175 ===\n",
      "train loss:1.3750253778870902\n",
      "train loss:1.399234196582923\n",
      "train loss:1.254623548240226\n",
      "=== epoch:276, train acc:0.6666666666666666, test acc:0.5219 ===\n",
      "train loss:1.3899734504267516\n",
      "train loss:1.201600882870079\n",
      "train loss:1.4474894043043816\n",
      "=== epoch:277, train acc:0.6633333333333333, test acc:0.5226 ===\n",
      "train loss:1.3150299298868282\n",
      "train loss:1.3637474791982003\n",
      "train loss:1.3787862425099462\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "=== epoch:278, train acc:0.67, test acc:0.525 ===\n",
      "train loss:1.4752879261815288\n",
      "train loss:1.3670785942514967\n",
      "train loss:1.393251224833496\n",
      "=== epoch:279, train acc:0.6666666666666666, test acc:0.5301 ===\n",
      "train loss:1.3153082228701591\n",
      "train loss:1.4051330071163313\n",
      "train loss:1.3778609502917942\n",
      "=== epoch:280, train acc:0.67, test acc:0.531 ===\n",
      "train loss:1.4778113460365967\n",
      "train loss:1.304002919961053\n",
      "train loss:1.2736098563074014\n",
      "=== epoch:281, train acc:0.6733333333333333, test acc:0.5319 ===\n",
      "train loss:1.4145567512280448\n",
      "train loss:1.3020489910329713\n",
      "train loss:1.3555116670481708\n",
      "=== epoch:282, train acc:0.6766666666666666, test acc:0.5355 ===\n",
      "train loss:1.2794324195827713\n",
      "train loss:1.207383460367849\n",
      "train loss:1.3150958785866038\n",
      "=== epoch:283, train acc:0.6766666666666666, test acc:0.5336 ===\n",
      "train loss:1.4318969659515934\n",
      "train loss:1.2771091416937113\n",
      "train loss:1.3685135908478951\n",
      "=== epoch:284, train acc:0.6766666666666666, test acc:0.5373 ===\n",
      "train loss:1.3067810569734717\n",
      "train loss:1.2784117957959718\n",
      "train loss:1.38491786665132\n",
      "=== epoch:285, train acc:0.6733333333333333, test acc:0.5408 ===\n",
      "train loss:1.317990215006794\n",
      "train loss:1.3248650493097314\n",
      "train loss:1.336070835139298\n",
      "=== epoch:286, train acc:0.6766666666666666, test acc:0.5386 ===\n",
      "train loss:1.375196507617037\n",
      "train loss:1.4651239393520274\n",
      "train loss:1.3407845556920817\n",
      "=== epoch:287, train acc:0.6733333333333333, test acc:0.5435 ===\n",
      "train loss:1.2835201806173961\n",
      "train loss:1.374604598967195\n",
      "train loss:1.3524711703987753\n",
      "=== epoch:288, train acc:0.6733333333333333, test acc:0.5449 ===\n",
      "train loss:1.3440251678681272\n",
      "train loss:1.381675246522187\n",
      "train loss:1.3695227803887986\n",
      "=== epoch:289, train acc:0.6733333333333333, test acc:0.5443 ===\n",
      "train loss:1.4327356927120465\n",
      "train loss:1.201123203840719\n",
      "train loss:1.280258530906722\n",
      "=== epoch:290, train acc:0.6833333333333333, test acc:0.5468 ===\n",
      "train loss:1.1717164747865958\n",
      "train loss:1.2165041510129384\n",
      "train loss:1.374628712891876\n",
      "=== epoch:291, train acc:0.6733333333333333, test acc:0.5511 ===\n",
      "train loss:1.3090326306949618\n",
      "train loss:1.3437321812496825\n",
      "train loss:1.177740480052209\n",
      "=== epoch:292, train acc:0.67, test acc:0.5495 ===\n",
      "train loss:1.2808611242685313\n",
      "train loss:1.355620265213523\n",
      "train loss:1.2529481687738289\n",
      "=== epoch:293, train acc:0.6733333333333333, test acc:0.5522 ===\n",
      "train loss:1.3795662422821593\n",
      "train loss:1.1612999866077134\n",
      "train loss:1.4017618595353258\n",
      "=== epoch:294, train acc:0.68, test acc:0.5546 ===\n",
      "train loss:1.3536916074031224\n",
      "train loss:1.1837651932191995\n",
      "train loss:1.192744461627401\n",
      "=== epoch:295, train acc:0.68, test acc:0.5517 ===\n",
      "train loss:1.2764109417827054\n",
      "train loss:1.2480631935565412\n",
      "train loss:1.3673043637391644\n",
      "=== epoch:296, train acc:0.6833333333333333, test acc:0.5575 ===\n",
      "train loss:1.3947611804368585\n",
      "train loss:1.3193387313520055\n",
      "train loss:1.1407847206752328\n",
      "=== epoch:297, train acc:0.6766666666666666, test acc:0.5554 ===\n",
      "train loss:1.3104721629741507\n",
      "train loss:1.2942833803933622\n",
      "train loss:1.2754266252885111\n",
      "=== epoch:298, train acc:0.69, test acc:0.5551 ===\n",
      "train loss:1.3632591498504198\n",
      "train loss:1.135861946150914\n",
      "train loss:1.3229733102649788\n",
      "=== epoch:299, train acc:0.6966666666666667, test acc:0.5592 ===\n",
      "train loss:1.2270092066861267\n",
      "train loss:1.2269765012519587\n",
      "train loss:1.1564445610726357\n",
      "=== epoch:300, train acc:0.69, test acc:0.5609 ===\n",
      "train loss:1.331116893579914\n",
      "train loss:1.1639781938376432\n",
      "train loss:1.192812228407215\n",
      "=== epoch:301, train acc:0.69, test acc:0.5633 ===\n",
      "train loss:1.1916785747019214\n",
      "train loss:1.240374019433102\n",
      "=============== Final Test Accuracy ===============\n",
      "test acc:0.5647\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"
    }
   ],
   "source": [
    "# coding: utf-8\n",
    "import os\n",
    "import sys\n",
    "sys.path.append(os.pardir)  # 为导入父目录中的文件而进行的设置\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from dataset.mnist import load_mnist\n",
    "from common.multi_layer_net_extend import MultiLayerNetExtend\n",
    "from common.trainer import Trainer\n",
    "\n",
    "(x_train, t_train), (x_test, t_test) = load_mnist(normalize=True)\n",
    "\n",
    "# 为了重现过拟合，减少学习数据\n",
    "x_train = x_train[:300]\n",
    "t_train = t_train[:300]\n",
    "\n",
    "# Dropuout有无、比例的设置 ========================\n",
    "use_dropout = True  # 不使用Dropout时修改为False\n",
    "dropout_ratio = 0.2\n",
    "# ====================================================\n",
    "\n",
    "network = MultiLayerNetExtend(input_size=784, hidden_size_list=[100, 100, 100, 100, 100, 100],\n",
    "                              output_size=10, use_dropout=use_dropout, dropout_ration=dropout_ratio)\n",
    "trainer = Trainer(network, x_train, t_train, x_test, t_test,\n",
    "                  epochs=301, mini_batch_size=100,\n",
    "                  optimizer='sgd', optimizer_param={'lr': 0.01}, verbose=True)\n",
    "trainer.train()\n",
    "\n",
    "train_acc_list, test_acc_list = trainer.train_acc_list, trainer.test_acc_list\n",
    "\n",
    "# 绘制图形==========\n",
    "markers = {'train': 'o', 'test': 's'}\n",
    "x = np.arange(len(train_acc_list))\n",
    "plt.plot(x, train_acc_list, marker='o', label='train', markevery=10)\n",
    "plt.plot(x, test_acc_list, marker='s', label='test', markevery=10)\n",
    "plt.xlabel(\"epochs\")\n",
    "plt.ylabel(\"accuracy\")\n",
    "plt.ylim(0, 1.0)\n",
    "plt.legend(loc='lower right')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 超参数优化\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch/blob/master/ch06/hyperparameter_optimization.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "val acc:0.2 | lr:0.0007769588213494039, weight decay:5.57176176211697e-07\n",
      "val acc:0.09 | lr:1.3267256882921564e-06, weight decay:5.5752601203992484e-05\n",
      "val acc:0.38 | lr:0.002504605960206648, weight decay:2.0202842319040335e-08\n",
      "val acc:0.23 | lr:0.0008027405772779653, weight decay:4.0924190556214115e-08\n",
      "val acc:0.14 | lr:0.0007483691319670275, weight decay:3.756941308098675e-08\n",
      "val acc:0.1 | lr:7.536899240905146e-06, weight decay:6.030129778228275e-08\n",
      "val acc:0.12 | lr:1.1912606390468455e-05, weight decay:5.118911336172855e-06\n",
      "val acc:0.12 | lr:4.607219519807864e-06, weight decay:8.072011260050135e-07\n",
      "val acc:0.24 | lr:0.0017525745882844492, weight decay:4.1116141319315954e-08\n",
      "val acc:0.16 | lr:2.329540878455772e-05, weight decay:2.513560161478032e-07\n",
      "val acc:0.22 | lr:0.0014971366110986379, weight decay:3.384077905156224e-07\n",
      "val acc:0.42 | lr:0.003537235885586346, weight decay:1.9771329007723216e-07\n",
      "val acc:0.09 | lr:0.0001163661382149746, weight decay:2.0834277866463343e-08\n",
      "val acc:0.09 | lr:0.00012298579192778668, weight decay:1.9446865367562307e-05\n",
      "val acc:0.13 | lr:0.0007943811289274845, weight decay:1.659346102995125e-05\n",
      "val acc:0.1 | lr:1.4078693434373915e-05, weight decay:2.3372015981995783e-07\n",
      "val acc:0.13 | lr:0.0010382257945057071, weight decay:1.0732135726312195e-08\n",
      "val acc:0.12 | lr:1.2592544448426485e-05, weight decay:2.7062564962433992e-05\n",
      "val acc:0.13 | lr:7.315764534308275e-06, weight decay:1.8188746423836085e-07\n",
      "val acc:0.1 | lr:0.0003138485002631171, weight decay:5.788434404648824e-06\n",
      "val acc:0.1 | lr:2.0139899157270438e-05, weight decay:4.980652853856161e-05\n",
      "val acc:0.19 | lr:3.394629494696132e-05, weight decay:5.797016204509946e-07\n",
      "val acc:0.12 | lr:0.0012555808769206095, weight decay:1.312457574305423e-07\n",
      "val acc:0.1 | lr:0.0006114165627285994, weight decay:3.92880543294753e-08\n",
      "val acc:0.09 | lr:0.0007488501503775594, weight decay:5.638582612171145e-08\n",
      "val acc:0.27 | lr:0.0012618198190772704, weight decay:9.670098378187491e-05\n",
      "val acc:0.26 | lr:0.0017224459141046927, weight decay:3.342865122530131e-05\n",
      "val acc:0.16 | lr:0.0005499207136182258, weight decay:2.0625510795073888e-07\n",
      "val acc:0.04 | lr:4.768902503674959e-05, weight decay:8.048479435560561e-05\n",
      "val acc:0.13 | lr:8.307630289273723e-05, weight decay:4.2437830283210317e-07\n",
      "val acc:0.1 | lr:1.0099901382080252e-06, weight decay:3.119502823204179e-07\n",
      "val acc:0.08 | lr:0.0002161022578132491, weight decay:1.8159286533663447e-07\n",
      "val acc:0.24 | lr:0.0010360213469005537, weight decay:1.0790196997224994e-06\n",
      "val acc:0.69 | lr:0.006583740934811419, weight decay:2.4510878359141775e-08\n",
      "val acc:0.15 | lr:1.3315489327297027e-06, weight decay:3.204712314595411e-08\n",
      "val acc:0.31 | lr:0.0007123541073071506, weight decay:2.0479421066247043e-05\n",
      "val acc:0.07 | lr:1.2466524538860027e-06, weight decay:1.7260301418093922e-05\n",
      "val acc:0.12 | lr:5.713857136253927e-05, weight decay:2.849425581223828e-08\n",
      "val acc:0.08 | lr:3.272747422717598e-06, weight decay:1.0204875649732498e-05\n",
      "val acc:0.08 | lr:0.00013918896438175655, weight decay:3.503717373558697e-08\n",
      "val acc:0.09 | lr:1.5810131642563531e-06, weight decay:1.59027098272125e-08\n",
      "val acc:0.1 | lr:0.00014978847160211322, weight decay:3.672248589341479e-07\n",
      "val acc:0.48 | lr:0.003401802580016628, weight decay:2.3978430224318266e-07\n",
      "val acc:0.74 | lr:0.009564022889235186, weight decay:1.9174147145893935e-07\n",
      "val acc:0.06 | lr:9.367140833934535e-06, weight decay:3.2480735862966805e-08\n",
      "val acc:0.1 | lr:1.3306006217877727e-05, weight decay:1.2865700584012741e-08\n",
      "val acc:0.2 | lr:0.0004887201225120626, weight decay:8.109996237837384e-06\n",
      "val acc:0.16 | lr:0.00017514747042300007, weight decay:6.507658399814367e-05\n",
      "val acc:0.05 | lr:3.856699929844645e-06, weight decay:1.6568123189667015e-08\n",
      "val acc:0.15 | lr:1.4991918066661082e-05, weight decay:5.23067932727341e-07\n",
      "val acc:0.22 | lr:0.00020925236488394526, weight decay:1.391913628468251e-06\n",
      "val acc:0.11 | lr:5.300399961299632e-05, weight decay:4.165535337753011e-08\n",
      "val acc:0.13 | lr:3.3858697638718774e-06, weight decay:4.764509793916045e-08\n",
      "val acc:0.12 | lr:2.900625016284411e-05, weight decay:1.9751068400418394e-07\n",
      "val acc:0.17 | lr:1.7840456188857138e-06, weight decay:1.1369895984269751e-05\n",
      "val acc:0.17 | lr:0.0005495163568864981, weight decay:2.592796581541801e-05\n",
      "val acc:0.59 | lr:0.004053312887321718, weight decay:6.168489257591741e-07\n",
      "val acc:0.1 | lr:1.814143712053503e-05, weight decay:5.307475885369047e-08\n",
      "val acc:0.14 | lr:0.00031759290120168614, weight decay:2.7804018838303672e-08\n",
      "val acc:0.1 | lr:2.193931798257391e-06, weight decay:4.6264230211388114e-08\n",
      "val acc:0.11 | lr:5.8671608342714224e-05, weight decay:7.343963377530732e-07\n",
      "val acc:0.55 | lr:0.003987379379558735, weight decay:3.294575887847477e-07\n",
      "val acc:0.08 | lr:3.570366628832604e-06, weight decay:7.013831647247545e-05\n",
      "val acc:0.06 | lr:5.880270516568243e-06, weight decay:4.037982814274879e-07\n",
      "val acc:0.73 | lr:0.008213942470539732, weight decay:1.6817057159020234e-06\n",
      "val acc:0.04 | lr:7.864020481418103e-05, weight decay:1.070757718847221e-07\n",
      "val acc:0.08 | lr:3.309258969912336e-05, weight decay:1.0352176122469794e-08\n",
      "val acc:0.22 | lr:0.0013793383074247685, weight decay:5.0763754503523075e-08\n",
      "val acc:0.62 | lr:0.004128881610091572, weight decay:6.557661560570525e-07\n",
      "val acc:0.09 | lr:0.0008089932066406872, weight decay:2.3372715159720323e-07\n",
      "val acc:0.14 | lr:3.243138442140921e-06, weight decay:4.76015755601967e-06\n",
      "val acc:0.08 | lr:7.933742519865885e-06, weight decay:2.8539702395826546e-05\n",
      "val acc:0.67 | lr:0.006640232657250519, weight decay:1.0069085366260983e-06\n",
      "val acc:0.75 | lr:0.007490756649803174, weight decay:1.702180778373477e-06\n",
      "val acc:0.2 | lr:4.712692697137205e-06, weight decay:4.418104250556227e-07\n",
      "val acc:0.1 | lr:9.078005415040448e-06, weight decay:3.0598003927489806e-08\n",
      "val acc:0.11 | lr:1.1389276929769857e-06, weight decay:1.424102400518378e-07\n",
      "val acc:0.16 | lr:0.00011177760626055852, weight decay:4.9208364266583125e-05\n",
      "val acc:0.11 | lr:1.626256132277201e-05, weight decay:8.911727718598043e-08\n",
      "val acc:0.18 | lr:1.3939229554873648e-05, weight decay:8.601469036364961e-05\n",
      "val acc:0.12 | lr:1.8970728879082828e-06, weight decay:1.2252914221652334e-07\n",
      "val acc:0.33 | lr:0.0019654681016951286, weight decay:3.9799475911747793e-08\n",
      "val acc:0.15 | lr:2.144650748003802e-06, weight decay:3.582539932182336e-07\n",
      "val acc:0.45 | lr:0.0014593436078593125, weight decay:1.549177513799557e-08\n",
      "val acc:0.04 | lr:6.936323751575047e-05, weight decay:4.036200786133129e-06\n",
      "val acc:0.13 | lr:4.201950600462706e-06, weight decay:2.3644804644734146e-06\n",
      "val acc:0.14 | lr:0.00035005761691108333, weight decay:3.125685897115486e-08\n",
      "val acc:0.09 | lr:1.129211914566036e-05, weight decay:2.8026240999239493e-08\n",
      "val acc:0.07 | lr:0.0005968323278638506, weight decay:6.070858152602163e-08\n",
      "val acc:0.34 | lr:0.0004344458602276125, weight decay:2.2830959152957767e-08\n",
      "val acc:0.16 | lr:0.0008889720464393857, weight decay:2.1290769881213858e-06\n",
      "val acc:0.06 | lr:7.0124443963030376e-06, weight decay:1.1890242303756916e-08\n",
      "val acc:0.1 | lr:0.0007829260167752746, weight decay:7.1866787497968116e-06\n",
      "val acc:0.28 | lr:0.0010649941529204936, weight decay:5.365355335264311e-05\n",
      "val acc:0.08 | lr:0.00015099534969899403, weight decay:7.043624929700249e-05\n",
      "val acc:0.1 | lr:4.846461255236942e-06, weight decay:1.2567668205617925e-07\n",
      "val acc:0.23 | lr:0.00241167843860275, weight decay:4.482291841142313e-06\n",
      "val acc:0.17 | lr:0.000714298548542353, weight decay:4.972406196165171e-08\n",
      "val acc:0.1 | lr:9.863468268563811e-05, weight decay:1.4806632709973454e-08\n",
      "val acc:0.46 | lr:0.002641957953354774, weight decay:9.16602722876144e-05\n",
      "=========== Hyper-Parameter Optimization Result ===========\n",
      "Best-1(val acc:0.75) | lr:0.007490756649803174, weight decay:1.702180778373477e-06\n",
      "Best-2(val acc:0.74) | lr:0.009564022889235186, weight decay:1.9174147145893935e-07\n",
      "Best-3(val acc:0.73) | lr:0.008213942470539732, weight decay:1.6817057159020234e-06\n",
      "Best-4(val acc:0.69) | lr:0.006583740934811419, weight decay:2.4510878359141775e-08\n",
      "Best-5(val acc:0.67) | lr:0.006640232657250519, weight decay:1.0069085366260983e-06\n",
      "Best-6(val acc:0.62) | lr:0.004128881610091572, weight decay:6.557661560570525e-07\n",
      "Best-7(val acc:0.59) | lr:0.004053312887321718, weight decay:6.168489257591741e-07\n",
      "Best-8(val acc:0.55) | lr:0.003987379379558735, weight decay:3.294575887847477e-07\n",
      "Best-9(val acc:0.48) | lr:0.003401802580016628, weight decay:2.3978430224318266e-07\n",
      "Best-10(val acc:0.46) | lr:0.002641957953354774, weight decay:9.16602722876144e-05\n",
      "Best-11(val acc:0.45) | lr:0.0014593436078593125, weight decay:1.549177513799557e-08\n",
      "Best-12(val acc:0.42) | lr:0.003537235885586346, weight decay:1.9771329007723216e-07\n",
      "Best-13(val acc:0.38) | lr:0.002504605960206648, weight decay:2.0202842319040335e-08\n",
      "Best-14(val acc:0.34) | lr:0.0004344458602276125, weight decay:2.2830959152957767e-08\n",
      "Best-15(val acc:0.33) | lr:0.0019654681016951286, weight decay:3.9799475911747793e-08\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best-16(val acc:0.31) | lr:0.0007123541073071506, weight decay:2.0479421066247043e-05\n",
      "Best-17(val acc:0.28) | lr:0.0010649941529204936, weight decay:5.365355335264311e-05\n",
      "Best-18(val acc:0.27) | lr:0.0012618198190772704, weight decay:9.670098378187491e-05\n",
      "Best-19(val acc:0.26) | lr:0.0017224459141046927, weight decay:3.342865122530131e-05\n",
      "Best-20(val acc:0.24) | lr:0.0017525745882844492, weight decay:4.1116141319315954e-08\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 20 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# coding: utf-8\n",
    "import sys, os\n",
    "sys.path.append(os.pardir)   # 为导入父目录中的文件而进行的设置\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from dataset.mnist import load_mnist\n",
    "from common.multi_layer_net import MultiLayerNet\n",
    "from common.util import shuffle_dataset\n",
    "from common.trainer import Trainer\n",
    "\n",
    "(x_train, t_train), (x_test, t_test) = load_mnist(normalize=True)\n",
    "\n",
    "# 为了加快速度，减少训练数据\n",
    "x_train = x_train[:500]\n",
    "t_train = t_train[:500]\n",
    "\n",
    "# 分离验证数据\n",
    "validation_rate = 0.20\n",
    "validation_num = int(x_train.shape[0] * validation_rate)\n",
    "x_train, t_train = shuffle_dataset(x_train, t_train)\n",
    "x_val = x_train[:validation_num]\n",
    "t_val = t_train[:validation_num]\n",
    "x_train = x_train[validation_num:]\n",
    "t_train = t_train[validation_num:]\n",
    "\n",
    "\n",
    "def __train(lr, weight_decay, epocs=50):\n",
    "    network = MultiLayerNet(input_size=784, hidden_size_list=[100, 100, 100, 100, 100, 100],\n",
    "                            output_size=10, weight_decay_lambda=weight_decay)\n",
    "    trainer = Trainer(network, x_train, t_train, x_val, t_val,\n",
    "                      epochs=epocs, mini_batch_size=100,\n",
    "                      optimizer='sgd', optimizer_param={'lr': lr}, verbose=False)\n",
    "    trainer.train()\n",
    "\n",
    "    return trainer.test_acc_list, trainer.train_acc_list\n",
    "\n",
    "\n",
    "# 随机探索超参数======================================\n",
    "optimization_trial = 100\n",
    "results_val = {}\n",
    "results_train = {}\n",
    "for _ in range(optimization_trial):\n",
    "    # 指定探索的超参数的范围===============\n",
    "    weight_decay = 10 ** np.random.uniform(-8, -4)\n",
    "    lr = 10 ** np.random.uniform(-6, -2)\n",
    "    # ================================================\n",
    "\n",
    "    val_acc_list, train_acc_list = __train(lr, weight_decay)\n",
    "    print(\"val acc:\" + str(val_acc_list[-1]) + \" | lr:\" + str(lr) + \", weight decay:\" + str(weight_decay))\n",
    "    key = \"lr:\" + str(lr) + \", weight decay:\" + str(weight_decay)\n",
    "    results_val[key] = val_acc_list\n",
    "    results_train[key] = train_acc_list\n",
    "\n",
    "# 绘制图表========================================================\n",
    "print(\"=========== Hyper-Parameter Optimization Result ===========\")\n",
    "graph_draw_num = 20\n",
    "col_num = 5\n",
    "row_num = int(np.ceil(graph_draw_num / col_num))\n",
    "i = 0\n",
    "\n",
    "for key, val_acc_list in sorted(results_val.items(), key=lambda x:x[1][-1], reverse=True):\n",
    "    print(\"Best-\" + str(i+1) + \"(val acc:\" + str(val_acc_list[-1]) + \") | \" + key)\n",
    "\n",
    "    plt.subplot(row_num, col_num, i+1)\n",
    "    plt.title(\"Best-\" + str(i+1))\n",
    "    plt.ylim(0.0, 1.0)\n",
    "    if i % 5: plt.yticks([])\n",
    "    plt.xticks([])\n",
    "    x = np.arange(len(val_acc_list))\n",
    "    plt.plot(x, val_acc_list)\n",
    "    plt.plot(x, results_train[key], \"--\")\n",
    "    i += 1\n",
    "\n",
    "    if i >= graph_draw_num:\n",
    "        break\n",
    "\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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}