ch01-复习神经网络.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 前馈神经网络\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch-2/blob/master/ch01/forward_net.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[-0.30834003 -0.06579832]\n",
      " [-0.6295676  -0.80444605]\n",
      " [-0.01685085 -0.38525012]\n",
      " [-0.70253815  0.14422036]\n",
      " [ 0.15505879 -1.05501557]\n",
      " [-0.17494723  0.51660855]\n",
      " [ 0.20985684  0.64778033]\n",
      " [ 0.69458196 -0.70942061]\n",
      " [-0.09277879 -1.57050443]\n",
      " [ 0.25062051 -1.39298815]]\n",
      "[[-3.05772457e+00 -2.60360790e-02 -9.42533224e-01]\n",
      " [-3.01628072e+00 -1.13672094e-01 -8.87446064e-01]\n",
      " [-2.91303378e+00 -1.38681717e-01 -5.73889828e-01]\n",
      " [-3.22190525e+00  9.24841199e-02 -1.35911930e+00]\n",
      " [-2.75257618e+00 -2.79126287e-01 -1.96587330e-01]\n",
      " [-3.11803667e+00  6.42211766e-02 -1.08266520e+00]\n",
      " [-3.03201096e+00  3.75781013e-04 -8.52611145e-01]\n",
      " [-2.70540075e+00 -2.85002246e-01  6.69016770e-03]\n",
      " [-2.72780795e+00 -3.44036463e-01 -2.00694975e-01]\n",
      " [-2.67611220e+00 -3.40577442e-01 -2.99368108e-02]]\n"
     ]
    }
   ],
   "source": [
    "# coding: utf-8\n",
    "import numpy as np\n",
    "\n",
    "class Sigmoid:\n",
    "    def __init__(self):\n",
    "        self.params = []\n",
    "\n",
    "    def forward(self, x):\n",
    "        return 1 / (1 + np.exp(-x))\n",
    "\n",
    "\n",
    "class Affine:\n",
    "    def __init__(self, W, b):\n",
    "        self.params = [W, b]\n",
    "\n",
    "    def forward(self, x):\n",
    "        W, b = self.params\n",
    "        out = np.dot(x, W) + b\n",
    "        return out\n",
    "\n",
    "\n",
    "class TwoLayerNet:\n",
    "    def __init__(self, input_size, hidden_size, output_size):\n",
    "        I, H, O = input_size, hidden_size, output_size\n",
    "\n",
    "        # 初始化权重和偏置\n",
    "        W1 = np.random.randn(I, H)\n",
    "        b1 = np.random.randn(H)\n",
    "        W2 = np.random.randn(H, O)\n",
    "        b2 = np.random.randn(O)\n",
    "\n",
    "        # 生成层\n",
    "        self.layers = [\n",
    "            Affine(W1, b1),\n",
    "            Sigmoid(),\n",
    "            Affine(W2, b2)\n",
    "        ]\n",
    "\n",
    "        # 将所有的权重汇总到列表\n",
    "        self.params = []\n",
    "        for layer in self.layers:\n",
    "            self.params += layer.params\n",
    "\n",
    "    def predict(self, x):\n",
    "        for layer in self.layers:\n",
    "            x = layer.forward(x)\n",
    "        return x\n",
    "\n",
    "\n",
    "x = np.random.randn(10, 2)\n",
    "\n",
    "print(x)\n",
    "\n",
    "model = TwoLayerNet(2, 4, 3)\n",
    "s = model.predict(x)\n",
    "print(s)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 双层网络\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch-2/blob/master/ch01/two_layer_net.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# coding: utf-8\n",
    "import sys\n",
    "sys.path.append('..')  # 为导入父目录中的文件而进行的设置\n",
    "import numpy as np\n",
    "from common.layers import Affine, Sigmoid, SoftmaxWithLoss\n",
    "\n",
    "\n",
    "class TwoLayerNet:\n",
    "    def __init__(self, input_size, hidden_size, output_size):\n",
    "        I, H, O = input_size, hidden_size, output_size\n",
    "\n",
    "        # 初始化权重和偏置\n",
    "        W1 = 0.01 * np.random.randn(I, H)\n",
    "        b1 = np.zeros(H)\n",
    "        W2 = 0.01 * np.random.randn(H, O)\n",
    "        b2 = np.zeros(O)\n",
    "\n",
    "        # 生成层\n",
    "        self.layers = [\n",
    "            Affine(W1, b1),\n",
    "            Sigmoid(),\n",
    "            Affine(W2, b2)\n",
    "        ]\n",
    "        self.loss_layer = SoftmaxWithLoss()\n",
    "\n",
    "        # 将所有的权重和梯度汇总到列表\n",
    "        self.params, self.grads = [], []\n",
    "        for layer in self.layers:\n",
    "            self.params += layer.params\n",
    "            self.grads += layer.grads\n",
    "\n",
    "    def predict(self, x):\n",
    "        for layer in self.layers:\n",
    "            x = layer.forward(x)\n",
    "        return x\n",
    "\n",
    "    def forward(self, x, t):\n",
    "        score = self.predict(x)\n",
    "        loss = self.loss_layer.forward(score, t)\n",
    "        return loss\n",
    "\n",
    "    def backward(self, dout=1):\n",
    "        dout = self.loss_layer.backward(dout)\n",
    "        for layer in reversed(self.layers):\n",
    "            dout = layer.backward(dout)\n",
    "        return dout"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 训练网络\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch-2/blob/master/ch01/train.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "| epoch 1 |  iter 1 / 10 | time 0[s] | loss 1.10\n",
      "| epoch 2 |  iter 1 / 10 | time 0[s] | loss 1.12\n",
      "| epoch 3 |  iter 1 / 10 | time 0[s] | loss 1.13\n",
      "| epoch 4 |  iter 1 / 10 | time 0[s] | loss 1.12\n",
      "| epoch 5 |  iter 1 / 10 | time 0[s] | loss 1.12\n",
      "| epoch 6 |  iter 1 / 10 | time 0[s] | loss 1.10\n",
      "| epoch 7 |  iter 1 / 10 | time 0[s] | loss 1.14\n",
      "| epoch 8 |  iter 1 / 10 | time 0[s] | loss 1.16\n",
      "| epoch 9 |  iter 1 / 10 | time 0[s] | loss 1.11\n",
      "| epoch 10 |  iter 1 / 10 | time 0[s] | loss 1.12\n",
      "| epoch 11 |  iter 1 / 10 | time 0[s] | loss 1.12\n",
      "| epoch 12 |  iter 1 / 10 | time 0[s] | loss 1.12\n",
      "| epoch 13 |  iter 1 / 10 | time 0[s] | loss 1.10\n",
      "| epoch 14 |  iter 1 / 10 | time 0[s] | loss 1.09\n",
      "| epoch 15 |  iter 1 / 10 | time 0[s] | loss 1.08\n",
      "| epoch 16 |  iter 1 / 10 | time 0[s] | loss 1.04\n",
      "| epoch 17 |  iter 1 / 10 | time 0[s] | loss 1.03\n",
      "| epoch 18 |  iter 1 / 10 | time 0[s] | loss 0.94\n",
      "| epoch 19 |  iter 1 / 10 | time 0[s] | loss 0.92\n",
      "| epoch 20 |  iter 1 / 10 | time 0[s] | loss 0.92\n",
      "| epoch 21 |  iter 1 / 10 | time 0[s] | loss 0.87\n",
      "| epoch 22 |  iter 1 / 10 | time 0[s] | loss 0.85\n",
      "| epoch 23 |  iter 1 / 10 | time 0[s] | loss 0.80\n",
      "| epoch 24 |  iter 1 / 10 | time 0[s] | loss 0.79\n",
      "| epoch 25 |  iter 1 / 10 | time 0[s] | loss 0.78\n",
      "| epoch 26 |  iter 1 / 10 | time 0[s] | loss 0.83\n",
      "| epoch 27 |  iter 1 / 10 | time 0[s] | loss 0.77\n",
      "| epoch 28 |  iter 1 / 10 | time 0[s] | loss 0.76\n",
      "| epoch 29 |  iter 1 / 10 | time 0[s] | loss 0.77\n",
      "| epoch 30 |  iter 1 / 10 | time 0[s] | loss 0.76\n",
      "| epoch 31 |  iter 1 / 10 | time 0[s] | loss 0.77\n",
      "| epoch 32 |  iter 1 / 10 | time 0[s] | loss 0.75\n",
      "| epoch 33 |  iter 1 / 10 | time 0[s] | loss 0.78\n",
      "| epoch 34 |  iter 1 / 10 | time 0[s] | loss 0.77\n",
      "| epoch 35 |  iter 1 / 10 | time 0[s] | loss 0.78\n",
      "| epoch 36 |  iter 1 / 10 | time 0[s] | loss 0.74\n",
      "| epoch 37 |  iter 1 / 10 | time 0[s] | loss 0.75\n",
      "| epoch 38 |  iter 1 / 10 | time 0[s] | loss 0.77\n",
      "| epoch 39 |  iter 1 / 10 | time 0[s] | loss 0.75\n",
      "| epoch 40 |  iter 1 / 10 | time 0[s] | loss 0.73\n",
      "| epoch 41 |  iter 1 / 10 | time 0[s] | loss 0.75\n",
      "| epoch 42 |  iter 1 / 10 | time 0[s] | loss 0.76\n",
      "| epoch 43 |  iter 1 / 10 | time 0[s] | loss 0.79\n",
      "| epoch 44 |  iter 1 / 10 | time 0[s] | loss 0.74\n",
      "| epoch 45 |  iter 1 / 10 | time 0[s] | loss 0.75\n",
      "| epoch 46 |  iter 1 / 10 | time 0[s] | loss 0.73\n",
      "| epoch 47 |  iter 1 / 10 | time 0[s] | loss 0.73\n",
      "| epoch 48 |  iter 1 / 10 | time 0[s] | loss 0.73\n",
      "| epoch 49 |  iter 1 / 10 | time 0[s] | loss 0.73\n",
      "| epoch 50 |  iter 1 / 10 | time 0[s] | loss 0.72\n",
      "| epoch 51 |  iter 1 / 10 | time 0[s] | loss 0.72\n",
      "| epoch 52 |  iter 1 / 10 | time 0[s] | loss 0.72\n",
      "| epoch 53 |  iter 1 / 10 | time 0[s] | loss 0.72\n",
      "| epoch 54 |  iter 1 / 10 | time 0[s] | loss 0.74\n",
      "| epoch 55 |  iter 1 / 10 | time 0[s] | loss 0.74\n",
      "| epoch 56 |  iter 1 / 10 | time 0[s] | loss 0.73\n",
      "| epoch 57 |  iter 1 / 10 | time 0[s] | loss 0.72\n",
      "| epoch 58 |  iter 1 / 10 | time 0[s] | loss 0.69\n",
      "| epoch 59 |  iter 1 / 10 | time 0[s] | loss 0.72\n",
      "| epoch 60 |  iter 1 / 10 | time 0[s] | loss 0.70\n",
      "| epoch 61 |  iter 1 / 10 | time 0[s] | loss 0.69\n",
      "| epoch 62 |  iter 1 / 10 | time 0[s] | loss 0.71\n",
      "| epoch 63 |  iter 1 / 10 | time 0[s] | loss 0.70\n",
      "| epoch 64 |  iter 1 / 10 | time 0[s] | loss 0.71\n",
      "| epoch 65 |  iter 1 / 10 | time 0[s] | loss 0.72\n",
      "| epoch 66 |  iter 1 / 10 | time 0[s] | loss 0.71\n",
      "| epoch 67 |  iter 1 / 10 | time 0[s] | loss 0.71\n",
      "| epoch 68 |  iter 1 / 10 | time 0[s] | loss 0.71\n",
      "| epoch 69 |  iter 1 / 10 | time 0[s] | loss 0.70\n",
      "| epoch 70 |  iter 1 / 10 | time 0[s] | loss 0.68\n",
      "| epoch 71 |  iter 1 / 10 | time 0[s] | loss 0.73\n",
      "| epoch 72 |  iter 1 / 10 | time 0[s] | loss 0.66\n",
      "| epoch 73 |  iter 1 / 10 | time 0[s] | loss 0.69\n",
      "| epoch 74 |  iter 1 / 10 | time 0[s] | loss 0.66\n",
      "| epoch 75 |  iter 1 / 10 | time 0[s] | loss 0.70\n",
      "| epoch 76 |  iter 1 / 10 | time 0[s] | loss 0.65\n",
      "| epoch 77 |  iter 1 / 10 | time 0[s] | loss 0.67\n",
      "| epoch 78 |  iter 1 / 10 | time 0[s] | loss 0.70\n",
      "| epoch 79 |  iter 1 / 10 | time 0[s] | loss 0.63\n",
      "| epoch 80 |  iter 1 / 10 | time 0[s] | loss 0.66\n",
      "| epoch 81 |  iter 1 / 10 | time 0[s] | loss 0.65\n",
      "| epoch 82 |  iter 1 / 10 | time 0[s] | loss 0.66\n",
      "| epoch 83 |  iter 1 / 10 | time 0[s] | loss 0.64\n",
      "| epoch 84 |  iter 1 / 10 | time 0[s] | loss 0.62\n",
      "| epoch 85 |  iter 1 / 10 | time 0[s] | loss 0.62\n",
      "| epoch 86 |  iter 1 / 10 | time 0[s] | loss 0.63\n",
      "| epoch 87 |  iter 1 / 10 | time 0[s] | loss 0.59\n",
      "| epoch 88 |  iter 1 / 10 | time 0[s] | loss 0.58\n",
      "| epoch 89 |  iter 1 / 10 | time 0[s] | loss 0.61\n",
      "| epoch 90 |  iter 1 / 10 | time 0[s] | loss 0.59\n",
      "| epoch 91 |  iter 1 / 10 | time 0[s] | loss 0.58\n",
      "| epoch 92 |  iter 1 / 10 | time 0[s] | loss 0.57\n",
      "| epoch 93 |  iter 1 / 10 | time 0[s] | loss 0.55\n",
      "| epoch 94 |  iter 1 / 10 | time 0[s] | loss 0.54\n",
      "| epoch 95 |  iter 1 / 10 | time 0[s] | loss 0.53\n",
      "| epoch 96 |  iter 1 / 10 | time 0[s] | loss 0.54\n",
      "| epoch 97 |  iter 1 / 10 | time 0[s] | loss 0.51\n",
      "| epoch 98 |  iter 1 / 10 | time 0[s] | loss 0.51\n",
      "| epoch 99 |  iter 1 / 10 | time 0[s] | loss 0.50\n",
      "| epoch 100 |  iter 1 / 10 | time 0[s] | loss 0.47\n",
      "| epoch 101 |  iter 1 / 10 | time 0[s] | loss 0.49\n",
      "| epoch 102 |  iter 1 / 10 | time 0[s] | loss 0.46\n",
      "| epoch 103 |  iter 1 / 10 | time 0[s] | loss 0.44\n",
      "| epoch 104 |  iter 1 / 10 | time 0[s] | loss 0.47\n",
      "| epoch 105 |  iter 1 / 10 | time 0[s] | loss 0.44\n",
      "| epoch 106 |  iter 1 / 10 | time 0[s] | loss 0.43\n",
      "| epoch 107 |  iter 1 / 10 | time 0[s] | loss 0.43\n",
      "| epoch 108 |  iter 1 / 10 | time 0[s] | loss 0.39\n",
      "| epoch 109 |  iter 1 / 10 | time 0[s] | loss 0.40\n",
      "| epoch 110 |  iter 1 / 10 | time 0[s] | loss 0.41\n",
      "| epoch 111 |  iter 1 / 10 | time 0[s] | loss 0.38\n",
      "| epoch 112 |  iter 1 / 10 | time 0[s] | loss 0.38\n",
      "| epoch 113 |  iter 1 / 10 | time 0[s] | loss 0.38\n",
      "| epoch 114 |  iter 1 / 10 | time 0[s] | loss 0.37\n",
      "| epoch 115 |  iter 1 / 10 | time 0[s] | loss 0.36\n",
      "| epoch 116 |  iter 1 / 10 | time 0[s] | loss 0.34\n",
      "| epoch 117 |  iter 1 / 10 | time 0[s] | loss 0.35\n",
      "| epoch 118 |  iter 1 / 10 | time 0[s] | loss 0.33\n",
      "| epoch 119 |  iter 1 / 10 | time 0[s] | loss 0.35\n",
      "| epoch 120 |  iter 1 / 10 | time 0[s] | loss 0.33\n",
      "| epoch 121 |  iter 1 / 10 | time 0[s] | loss 0.33\n",
      "| epoch 122 |  iter 1 / 10 | time 0[s] | loss 0.32\n",
      "| epoch 123 |  iter 1 / 10 | time 0[s] | loss 0.31\n",
      "| epoch 124 |  iter 1 / 10 | time 0[s] | loss 0.31\n",
      "| epoch 125 |  iter 1 / 10 | time 0[s] | loss 0.31\n",
      "| epoch 126 |  iter 1 / 10 | time 0[s] | loss 0.30\n",
      "| epoch 127 |  iter 1 / 10 | time 0[s] | loss 0.30\n",
      "| epoch 128 |  iter 1 / 10 | time 0[s] | loss 0.27\n",
      "| epoch 129 |  iter 1 / 10 | time 0[s] | loss 0.30\n",
      "| epoch 130 |  iter 1 / 10 | time 0[s] | loss 0.28\n",
      "| epoch 131 |  iter 1 / 10 | time 0[s] | loss 0.26\n",
      "| epoch 132 |  iter 1 / 10 | time 0[s] | loss 0.27\n",
      "| epoch 133 |  iter 1 / 10 | time 0[s] | loss 0.27\n",
      "| epoch 134 |  iter 1 / 10 | time 0[s] | loss 0.28\n",
      "| epoch 135 |  iter 1 / 10 | time 0[s] | loss 0.26\n",
      "| epoch 136 |  iter 1 / 10 | time 0[s] | loss 0.28\n",
      "| epoch 137 |  iter 1 / 10 | time 0[s] | loss 0.25\n",
      "| epoch 138 |  iter 1 / 10 | time 0[s] | loss 0.26\n",
      "| epoch 139 |  iter 1 / 10 | time 0[s] | loss 0.26\n",
      "| epoch 140 |  iter 1 / 10 | time 0[s] | loss 0.26\n",
      "| epoch 141 |  iter 1 / 10 | time 0[s] | loss 0.23\n",
      "| epoch 142 |  iter 1 / 10 | time 0[s] | loss 0.23\n",
      "| epoch 143 |  iter 1 / 10 | time 0[s] | loss 0.26\n",
      "| epoch 144 |  iter 1 / 10 | time 0[s] | loss 0.23\n",
      "| epoch 145 |  iter 1 / 10 | time 0[s] | loss 0.24\n",
      "| epoch 146 |  iter 1 / 10 | time 0[s] | loss 0.24\n",
      "| epoch 147 |  iter 1 / 10 | time 0[s] | loss 0.25\n",
      "| epoch 148 |  iter 1 / 10 | time 0[s] | loss 0.21\n",
      "| epoch 149 |  iter 1 / 10 | time 0[s] | loss 0.23\n",
      "| epoch 150 |  iter 1 / 10 | time 0[s] | loss 0.22\n",
      "| epoch 151 |  iter 1 / 10 | time 0[s] | loss 0.22\n",
      "| epoch 152 |  iter 1 / 10 | time 0[s] | loss 0.23\n",
      "| epoch 153 |  iter 1 / 10 | time 0[s] | loss 0.23\n",
      "| epoch 154 |  iter 1 / 10 | time 0[s] | loss 0.20\n",
      "| epoch 155 |  iter 1 / 10 | time 0[s] | loss 0.22\n",
      "| epoch 156 |  iter 1 / 10 | time 0[s] | loss 0.21\n",
      "| epoch 157 |  iter 1 / 10 | time 0[s] | loss 0.21\n",
      "| epoch 158 |  iter 1 / 10 | time 0[s] | loss 0.20\n",
      "| epoch 159 |  iter 1 / 10 | time 0[s] | loss 0.21\n",
      "| epoch 160 |  iter 1 / 10 | time 0[s] | loss 0.20\n",
      "| epoch 161 |  iter 1 / 10 | time 0[s] | loss 0.19\n",
      "| epoch 162 |  iter 1 / 10 | time 0[s] | loss 0.22\n",
      "| epoch 163 |  iter 1 / 10 | time 0[s] | loss 0.19\n",
      "| epoch 164 |  iter 1 / 10 | time 0[s] | loss 0.21\n",
      "| epoch 165 |  iter 1 / 10 | time 0[s] | loss 0.20\n",
      "| epoch 166 |  iter 1 / 10 | time 0[s] | loss 0.20\n",
      "| epoch 167 |  iter 1 / 10 | time 0[s] | loss 0.20\n",
      "| epoch 168 |  iter 1 / 10 | time 0[s] | loss 0.19\n",
      "| epoch 169 |  iter 1 / 10 | time 0[s] | loss 0.18\n",
      "| epoch 170 |  iter 1 / 10 | time 0[s] | loss 0.19\n",
      "| epoch 171 |  iter 1 / 10 | time 0[s] | loss 0.19\n",
      "| epoch 172 |  iter 1 / 10 | time 0[s] | loss 0.20\n",
      "| epoch 173 |  iter 1 / 10 | time 0[s] | loss 0.16\n",
      "| epoch 174 |  iter 1 / 10 | time 0[s] | loss 0.20\n",
      "| epoch 175 |  iter 1 / 10 | time 0[s] | loss 0.18\n",
      "| epoch 176 |  iter 1 / 10 | time 0[s] | loss 0.17\n",
      "| epoch 177 |  iter 1 / 10 | time 0[s] | loss 0.17\n",
      "| epoch 178 |  iter 1 / 10 | time 0[s] | loss 0.17\n",
      "| epoch 179 |  iter 1 / 10 | time 0[s] | loss 0.18\n",
      "| epoch 180 |  iter 1 / 10 | time 0[s] | loss 0.19\n",
      "| epoch 181 |  iter 1 / 10 | time 0[s] | loss 0.17\n",
      "| epoch 182 |  iter 1 / 10 | time 0[s] | loss 0.18\n",
      "| epoch 183 |  iter 1 / 10 | time 0[s] | loss 0.16\n",
      "| epoch 184 |  iter 1 / 10 | time 0[s] | loss 0.18\n",
      "| epoch 185 |  iter 1 / 10 | time 0[s] | loss 0.18\n",
      "| epoch 186 |  iter 1 / 10 | time 0[s] | loss 0.17\n",
      "| epoch 187 |  iter 1 / 10 | time 0[s] | loss 0.17\n",
      "| epoch 188 |  iter 1 / 10 | time 0[s] | loss 0.18\n",
      "| epoch 189 |  iter 1 / 10 | time 0[s] | loss 0.16\n",
      "| epoch 190 |  iter 1 / 10 | time 0[s] | loss 0.16\n",
      "| epoch 191 |  iter 1 / 10 | time 0[s] | loss 0.17\n",
      "| epoch 192 |  iter 1 / 10 | time 0[s] | loss 0.17\n",
      "| epoch 193 |  iter 1 / 10 | time 0[s] | loss 0.16\n",
      "| epoch 194 |  iter 1 / 10 | time 0[s] | loss 0.16\n",
      "| epoch 195 |  iter 1 / 10 | time 0[s] | loss 0.16\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "| epoch 196 |  iter 1 / 10 | time 0[s] | loss 0.17\n",
      "| epoch 197 |  iter 1 / 10 | time 0[s] | loss 0.16\n",
      "| epoch 198 |  iter 1 / 10 | time 0[s] | loss 0.17\n",
      "| epoch 199 |  iter 1 / 10 | time 0[s] | loss 0.16\n",
      "| epoch 200 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 201 |  iter 1 / 10 | time 0[s] | loss 0.16\n",
      "| epoch 202 |  iter 1 / 10 | time 0[s] | loss 0.16\n",
      "| epoch 203 |  iter 1 / 10 | time 0[s] | loss 0.15\n",
      "| epoch 204 |  iter 1 / 10 | time 0[s] | loss 0.16\n",
      "| epoch 205 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 206 |  iter 1 / 10 | time 0[s] | loss 0.16\n",
      "| epoch 207 |  iter 1 / 10 | time 0[s] | loss 0.16\n",
      "| epoch 208 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 209 |  iter 1 / 10 | time 0[s] | loss 0.15\n",
      "| epoch 210 |  iter 1 / 10 | time 0[s] | loss 0.16\n",
      "| epoch 211 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 212 |  iter 1 / 10 | time 0[s] | loss 0.15\n",
      "| epoch 213 |  iter 1 / 10 | time 0[s] | loss 0.15\n",
      "| epoch 214 |  iter 1 / 10 | time 0[s] | loss 0.15\n",
      "| epoch 215 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 216 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 217 |  iter 1 / 10 | time 0[s] | loss 0.15\n",
      "| epoch 218 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 219 |  iter 1 / 10 | time 0[s] | loss 0.15\n",
      "| epoch 220 |  iter 1 / 10 | time 0[s] | loss 0.15\n",
      "| epoch 221 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 222 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 223 |  iter 1 / 10 | time 0[s] | loss 0.15\n",
      "| epoch 224 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 225 |  iter 1 / 10 | time 0[s] | loss 0.16\n",
      "| epoch 226 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 227 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 228 |  iter 1 / 10 | time 0[s] | loss 0.15\n",
      "| epoch 229 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 230 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 231 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 232 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 233 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 234 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 235 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 236 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 237 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 238 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 239 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 240 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 241 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 242 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 243 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 244 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 245 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 246 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 247 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 248 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 249 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 250 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 251 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 252 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 253 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 254 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 255 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 256 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 257 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 258 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 259 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 260 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 261 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 262 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 263 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 264 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 265 |  iter 1 / 10 | time 0[s] | loss 0.14\n",
      "| epoch 266 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 267 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 268 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 269 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 270 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 271 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 272 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 273 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 274 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 275 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 276 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 277 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 278 |  iter 1 / 10 | time 0[s] | loss 0.13\n",
      "| epoch 279 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 280 |  iter 1 / 10 | time 0[s] | loss 0.10\n",
      "| epoch 281 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 282 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 283 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 284 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 285 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 286 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 287 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 288 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 289 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 290 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 291 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 292 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 293 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 294 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 295 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 296 |  iter 1 / 10 | time 0[s] | loss 0.12\n",
      "| epoch 297 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 298 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 299 |  iter 1 / 10 | time 0[s] | loss 0.11\n",
      "| epoch 300 |  iter 1 / 10 | time 0[s] | loss 0.11\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# coding: utf-8\n",
    "import sys\n",
    "sys.path.append('..')  # 为导入父目录中的文件而进行的设置\n",
    "from common.optimizer import SGD\n",
    "from common.trainer import Trainer\n",
    "from dataset import spiral\n",
    "# from two_layer_net import TwoLayerNet\n",
    "\n",
    "# 设置超参数\n",
    "max_epoch = 300\n",
    "batch_size = 30\n",
    "hidden_size = 10\n",
    "learning_rate = 1.0\n",
    "\n",
    "x, t = spiral.load_data()\n",
    "model = TwoLayerNet(input_size=2, hidden_size=hidden_size, output_size=3)\n",
    "optimizer = SGD(lr=learning_rate)\n",
    "\n",
    "trainer = Trainer(model, optimizer)\n",
    "trainer.fit(x, t, max_epoch, batch_size, eval_interval=10)\n",
    "trainer.plot()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 绘制螺旋数据集\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch-2/blob/master/ch01/show_spiral_dataset.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "x (300, 2)\n",
      "t (300, 3)\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 sys\n",
    "sys.path.append('..')  # 为导入父目录中的文件而进行的设置\n",
    "from dataset import spiral\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "\n",
    "x, t = spiral.load_data()\n",
    "print('x', x.shape)  # (300, 2)\n",
    "print('t', t.shape)  # (300, 3)\n",
    "\n",
    "# 绘制数据点\n",
    "N = 100\n",
    "CLS_NUM = 3\n",
    "markers = ['o', 'x', '^']\n",
    "for i in range(CLS_NUM):\n",
    "    plt.scatter(x[i*N:(i+1)*N, 0], x[i*N:(i+1)*N, 1], s=40, marker=markers[i])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 训练自定义循环\n",
    "\n",
    "https://github.com/oreilly-japan/deep-learning-from-scratch-2/blob/master/ch01/train_custom_loop.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "| epoch 1 |  iter 10 / 10 | loss 1.13\n",
      "| epoch 2 |  iter 10 / 10 | loss 1.13\n",
      "| epoch 3 |  iter 10 / 10 | loss 1.12\n",
      "| epoch 4 |  iter 10 / 10 | loss 1.12\n",
      "| epoch 5 |  iter 10 / 10 | loss 1.11\n",
      "| epoch 6 |  iter 10 / 10 | loss 1.14\n",
      "| epoch 7 |  iter 10 / 10 | loss 1.16\n",
      "| epoch 8 |  iter 10 / 10 | loss 1.11\n",
      "| epoch 9 |  iter 10 / 10 | loss 1.12\n",
      "| epoch 10 |  iter 10 / 10 | loss 1.13\n",
      "| epoch 11 |  iter 10 / 10 | loss 1.12\n",
      "| epoch 12 |  iter 10 / 10 | loss 1.11\n",
      "| epoch 13 |  iter 10 / 10 | loss 1.09\n",
      "| epoch 14 |  iter 10 / 10 | loss 1.08\n",
      "| epoch 15 |  iter 10 / 10 | loss 1.04\n",
      "| epoch 16 |  iter 10 / 10 | loss 1.03\n",
      "| epoch 17 |  iter 10 / 10 | loss 0.96\n",
      "| epoch 18 |  iter 10 / 10 | loss 0.92\n",
      "| epoch 19 |  iter 10 / 10 | loss 0.92\n",
      "| epoch 20 |  iter 10 / 10 | loss 0.87\n",
      "| epoch 21 |  iter 10 / 10 | loss 0.85\n",
      "| epoch 22 |  iter 10 / 10 | loss 0.82\n",
      "| epoch 23 |  iter 10 / 10 | loss 0.79\n",
      "| epoch 24 |  iter 10 / 10 | loss 0.78\n",
      "| epoch 25 |  iter 10 / 10 | loss 0.82\n",
      "| epoch 26 |  iter 10 / 10 | loss 0.78\n",
      "| epoch 27 |  iter 10 / 10 | loss 0.76\n",
      "| epoch 28 |  iter 10 / 10 | loss 0.76\n",
      "| epoch 29 |  iter 10 / 10 | loss 0.78\n",
      "| epoch 30 |  iter 10 / 10 | loss 0.75\n",
      "| epoch 31 |  iter 10 / 10 | loss 0.78\n",
      "| epoch 32 |  iter 10 / 10 | loss 0.77\n",
      "| epoch 33 |  iter 10 / 10 | loss 0.77\n",
      "| epoch 34 |  iter 10 / 10 | loss 0.78\n",
      "| epoch 35 |  iter 10 / 10 | loss 0.75\n",
      "| epoch 36 |  iter 10 / 10 | loss 0.74\n",
      "| epoch 37 |  iter 10 / 10 | loss 0.76\n",
      "| epoch 38 |  iter 10 / 10 | loss 0.76\n",
      "| epoch 39 |  iter 10 / 10 | loss 0.73\n",
      "| epoch 40 |  iter 10 / 10 | loss 0.75\n",
      "| epoch 41 |  iter 10 / 10 | loss 0.76\n",
      "| epoch 42 |  iter 10 / 10 | loss 0.76\n",
      "| epoch 43 |  iter 10 / 10 | loss 0.76\n",
      "| epoch 44 |  iter 10 / 10 | loss 0.74\n",
      "| epoch 45 |  iter 10 / 10 | loss 0.75\n",
      "| epoch 46 |  iter 10 / 10 | loss 0.73\n",
      "| epoch 47 |  iter 10 / 10 | loss 0.72\n",
      "| epoch 48 |  iter 10 / 10 | loss 0.73\n",
      "| epoch 49 |  iter 10 / 10 | loss 0.72\n",
      "| epoch 50 |  iter 10 / 10 | loss 0.72\n",
      "| epoch 51 |  iter 10 / 10 | loss 0.72\n",
      "| epoch 52 |  iter 10 / 10 | loss 0.72\n",
      "| epoch 53 |  iter 10 / 10 | loss 0.74\n",
      "| epoch 54 |  iter 10 / 10 | loss 0.74\n",
      "| epoch 55 |  iter 10 / 10 | loss 0.72\n",
      "| epoch 56 |  iter 10 / 10 | loss 0.72\n",
      "| epoch 57 |  iter 10 / 10 | loss 0.71\n",
      "| epoch 58 |  iter 10 / 10 | loss 0.70\n",
      "| epoch 59 |  iter 10 / 10 | loss 0.72\n",
      "| epoch 60 |  iter 10 / 10 | loss 0.70\n",
      "| epoch 61 |  iter 10 / 10 | loss 0.71\n",
      "| epoch 62 |  iter 10 / 10 | loss 0.72\n",
      "| epoch 63 |  iter 10 / 10 | loss 0.70\n",
      "| epoch 64 |  iter 10 / 10 | loss 0.71\n",
      "| epoch 65 |  iter 10 / 10 | loss 0.73\n",
      "| epoch 66 |  iter 10 / 10 | loss 0.70\n",
      "| epoch 67 |  iter 10 / 10 | loss 0.71\n",
      "| epoch 68 |  iter 10 / 10 | loss 0.69\n",
      "| epoch 69 |  iter 10 / 10 | loss 0.70\n",
      "| epoch 70 |  iter 10 / 10 | loss 0.71\n",
      "| epoch 71 |  iter 10 / 10 | loss 0.68\n",
      "| epoch 72 |  iter 10 / 10 | loss 0.69\n",
      "| epoch 73 |  iter 10 / 10 | loss 0.67\n",
      "| epoch 74 |  iter 10 / 10 | loss 0.68\n",
      "| epoch 75 |  iter 10 / 10 | loss 0.67\n",
      "| epoch 76 |  iter 10 / 10 | loss 0.66\n",
      "| epoch 77 |  iter 10 / 10 | loss 0.69\n",
      "| epoch 78 |  iter 10 / 10 | loss 0.64\n",
      "| epoch 79 |  iter 10 / 10 | loss 0.68\n",
      "| epoch 80 |  iter 10 / 10 | loss 0.64\n",
      "| epoch 81 |  iter 10 / 10 | loss 0.64\n",
      "| epoch 82 |  iter 10 / 10 | loss 0.66\n",
      "| epoch 83 |  iter 10 / 10 | loss 0.62\n",
      "| epoch 84 |  iter 10 / 10 | loss 0.62\n",
      "| epoch 85 |  iter 10 / 10 | loss 0.61\n",
      "| epoch 86 |  iter 10 / 10 | loss 0.60\n",
      "| epoch 87 |  iter 10 / 10 | loss 0.60\n",
      "| epoch 88 |  iter 10 / 10 | loss 0.61\n",
      "| epoch 89 |  iter 10 / 10 | loss 0.59\n",
      "| epoch 90 |  iter 10 / 10 | loss 0.58\n",
      "| epoch 91 |  iter 10 / 10 | loss 0.56\n",
      "| epoch 92 |  iter 10 / 10 | loss 0.56\n",
      "| epoch 93 |  iter 10 / 10 | loss 0.54\n",
      "| epoch 94 |  iter 10 / 10 | loss 0.53\n",
      "| epoch 95 |  iter 10 / 10 | loss 0.53\n",
      "| epoch 96 |  iter 10 / 10 | loss 0.52\n",
      "| epoch 97 |  iter 10 / 10 | loss 0.51\n",
      "| epoch 98 |  iter 10 / 10 | loss 0.50\n",
      "| epoch 99 |  iter 10 / 10 | loss 0.48\n",
      "| epoch 100 |  iter 10 / 10 | loss 0.48\n",
      "| epoch 101 |  iter 10 / 10 | loss 0.46\n",
      "| epoch 102 |  iter 10 / 10 | loss 0.45\n",
      "| epoch 103 |  iter 10 / 10 | loss 0.45\n",
      "| epoch 104 |  iter 10 / 10 | loss 0.44\n",
      "| epoch 105 |  iter 10 / 10 | loss 0.44\n",
      "| epoch 106 |  iter 10 / 10 | loss 0.41\n",
      "| epoch 107 |  iter 10 / 10 | loss 0.40\n",
      "| epoch 108 |  iter 10 / 10 | loss 0.41\n",
      "| epoch 109 |  iter 10 / 10 | loss 0.40\n",
      "| epoch 110 |  iter 10 / 10 | loss 0.40\n",
      "| epoch 111 |  iter 10 / 10 | loss 0.38\n",
      "| epoch 112 |  iter 10 / 10 | loss 0.38\n",
      "| epoch 113 |  iter 10 / 10 | loss 0.36\n",
      "| epoch 114 |  iter 10 / 10 | loss 0.37\n",
      "| epoch 115 |  iter 10 / 10 | loss 0.35\n",
      "| epoch 116 |  iter 10 / 10 | loss 0.34\n",
      "| epoch 117 |  iter 10 / 10 | loss 0.34\n",
      "| epoch 118 |  iter 10 / 10 | loss 0.34\n",
      "| epoch 119 |  iter 10 / 10 | loss 0.33\n",
      "| epoch 120 |  iter 10 / 10 | loss 0.34\n",
      "| epoch 121 |  iter 10 / 10 | loss 0.32\n",
      "| epoch 122 |  iter 10 / 10 | loss 0.32\n",
      "| epoch 123 |  iter 10 / 10 | loss 0.31\n",
      "| epoch 124 |  iter 10 / 10 | loss 0.31\n",
      "| epoch 125 |  iter 10 / 10 | loss 0.30\n",
      "| epoch 126 |  iter 10 / 10 | loss 0.30\n",
      "| epoch 127 |  iter 10 / 10 | loss 0.28\n",
      "| epoch 128 |  iter 10 / 10 | loss 0.28\n",
      "| epoch 129 |  iter 10 / 10 | loss 0.28\n",
      "| epoch 130 |  iter 10 / 10 | loss 0.28\n",
      "| epoch 131 |  iter 10 / 10 | loss 0.27\n",
      "| epoch 132 |  iter 10 / 10 | loss 0.27\n",
      "| epoch 133 |  iter 10 / 10 | loss 0.27\n",
      "| epoch 134 |  iter 10 / 10 | loss 0.27\n",
      "| epoch 135 |  iter 10 / 10 | loss 0.27\n",
      "| epoch 136 |  iter 10 / 10 | loss 0.26\n",
      "| epoch 137 |  iter 10 / 10 | loss 0.26\n",
      "| epoch 138 |  iter 10 / 10 | loss 0.26\n",
      "| epoch 139 |  iter 10 / 10 | loss 0.25\n",
      "| epoch 140 |  iter 10 / 10 | loss 0.24\n",
      "| epoch 141 |  iter 10 / 10 | loss 0.24\n",
      "| epoch 142 |  iter 10 / 10 | loss 0.25\n",
      "| epoch 143 |  iter 10 / 10 | loss 0.24\n",
      "| epoch 144 |  iter 10 / 10 | loss 0.24\n",
      "| epoch 145 |  iter 10 / 10 | loss 0.23\n",
      "| epoch 146 |  iter 10 / 10 | loss 0.24\n",
      "| epoch 147 |  iter 10 / 10 | loss 0.23\n",
      "| epoch 148 |  iter 10 / 10 | loss 0.23\n",
      "| epoch 149 |  iter 10 / 10 | loss 0.22\n",
      "| epoch 150 |  iter 10 / 10 | loss 0.22\n",
      "| epoch 151 |  iter 10 / 10 | loss 0.22\n",
      "| epoch 152 |  iter 10 / 10 | loss 0.22\n",
      "| epoch 153 |  iter 10 / 10 | loss 0.22\n",
      "| epoch 154 |  iter 10 / 10 | loss 0.22\n",
      "| epoch 155 |  iter 10 / 10 | loss 0.22\n",
      "| epoch 156 |  iter 10 / 10 | loss 0.21\n",
      "| epoch 157 |  iter 10 / 10 | loss 0.21\n",
      "| epoch 158 |  iter 10 / 10 | loss 0.20\n",
      "| epoch 159 |  iter 10 / 10 | loss 0.21\n",
      "| epoch 160 |  iter 10 / 10 | loss 0.20\n",
      "| epoch 161 |  iter 10 / 10 | loss 0.20\n",
      "| epoch 162 |  iter 10 / 10 | loss 0.20\n",
      "| epoch 163 |  iter 10 / 10 | loss 0.21\n",
      "| epoch 164 |  iter 10 / 10 | loss 0.20\n",
      "| epoch 165 |  iter 10 / 10 | loss 0.20\n",
      "| epoch 166 |  iter 10 / 10 | loss 0.19\n",
      "| epoch 167 |  iter 10 / 10 | loss 0.19\n",
      "| epoch 168 |  iter 10 / 10 | loss 0.19\n",
      "| epoch 169 |  iter 10 / 10 | loss 0.19\n",
      "| epoch 170 |  iter 10 / 10 | loss 0.19\n",
      "| epoch 171 |  iter 10 / 10 | loss 0.19\n",
      "| epoch 172 |  iter 10 / 10 | loss 0.18\n",
      "| epoch 173 |  iter 10 / 10 | loss 0.18\n",
      "| epoch 174 |  iter 10 / 10 | loss 0.18\n",
      "| epoch 175 |  iter 10 / 10 | loss 0.18\n",
      "| epoch 176 |  iter 10 / 10 | loss 0.18\n",
      "| epoch 177 |  iter 10 / 10 | loss 0.18\n",
      "| epoch 178 |  iter 10 / 10 | loss 0.18\n",
      "| epoch 179 |  iter 10 / 10 | loss 0.17\n",
      "| epoch 180 |  iter 10 / 10 | loss 0.17\n",
      "| epoch 181 |  iter 10 / 10 | loss 0.18\n",
      "| epoch 182 |  iter 10 / 10 | loss 0.17\n",
      "| epoch 183 |  iter 10 / 10 | loss 0.18\n",
      "| epoch 184 |  iter 10 / 10 | loss 0.17\n",
      "| epoch 185 |  iter 10 / 10 | loss 0.17\n",
      "| epoch 186 |  iter 10 / 10 | loss 0.18\n",
      "| epoch 187 |  iter 10 / 10 | loss 0.17\n",
      "| epoch 188 |  iter 10 / 10 | loss 0.17\n",
      "| epoch 189 |  iter 10 / 10 | loss 0.17\n",
      "| epoch 190 |  iter 10 / 10 | loss 0.17\n",
      "| epoch 191 |  iter 10 / 10 | loss 0.16\n",
      "| epoch 192 |  iter 10 / 10 | loss 0.17\n",
      "| epoch 193 |  iter 10 / 10 | loss 0.16\n",
      "| epoch 194 |  iter 10 / 10 | loss 0.16\n",
      "| epoch 195 |  iter 10 / 10 | loss 0.16\n",
      "| epoch 196 |  iter 10 / 10 | loss 0.16\n",
      "| epoch 197 |  iter 10 / 10 | loss 0.16\n",
      "| epoch 198 |  iter 10 / 10 | loss 0.15\n",
      "| epoch 199 |  iter 10 / 10 | loss 0.16\n",
      "| epoch 200 |  iter 10 / 10 | loss 0.16\n",
      "| epoch 201 |  iter 10 / 10 | loss 0.15\n",
      "| epoch 202 |  iter 10 / 10 | loss 0.16\n",
      "| epoch 203 |  iter 10 / 10 | loss 0.16\n",
      "| epoch 204 |  iter 10 / 10 | loss 0.15\n",
      "| epoch 205 |  iter 10 / 10 | loss 0.16\n",
      "| epoch 206 |  iter 10 / 10 | loss 0.15\n",
      "| epoch 207 |  iter 10 / 10 | loss 0.15\n",
      "| epoch 208 |  iter 10 / 10 | loss 0.15\n",
      "| epoch 209 |  iter 10 / 10 | loss 0.15\n",
      "| epoch 210 |  iter 10 / 10 | loss 0.15\n",
      "| epoch 211 |  iter 10 / 10 | loss 0.15\n",
      "| epoch 212 |  iter 10 / 10 | loss 0.15\n",
      "| epoch 213 |  iter 10 / 10 | loss 0.15\n",
      "| epoch 214 |  iter 10 / 10 | loss 0.15\n",
      "| epoch 215 |  iter 10 / 10 | loss 0.15\n",
      "| epoch 216 |  iter 10 / 10 | loss 0.14\n",
      "| epoch 217 |  iter 10 / 10 | loss 0.14\n",
      "| epoch 218 |  iter 10 / 10 | loss 0.15\n",
      "| epoch 219 |  iter 10 / 10 | loss 0.14\n",
      "| epoch 220 |  iter 10 / 10 | loss 0.14\n",
      "| epoch 221 |  iter 10 / 10 | loss 0.14\n",
      "| epoch 222 |  iter 10 / 10 | loss 0.14\n",
      "| epoch 223 |  iter 10 / 10 | loss 0.14\n",
      "| epoch 224 |  iter 10 / 10 | loss 0.14\n",
      "| epoch 225 |  iter 10 / 10 | loss 0.14\n",
      "| epoch 226 |  iter 10 / 10 | loss 0.14\n",
      "| epoch 227 |  iter 10 / 10 | loss 0.14\n",
      "| epoch 228 |  iter 10 / 10 | loss 0.14\n",
      "| epoch 229 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 230 |  iter 10 / 10 | loss 0.14\n",
      "| epoch 231 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 232 |  iter 10 / 10 | loss 0.14\n",
      "| epoch 233 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 234 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 235 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 236 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 237 |  iter 10 / 10 | loss 0.14\n",
      "| epoch 238 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 239 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 240 |  iter 10 / 10 | loss 0.14\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "| epoch 241 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 242 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 243 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 244 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 245 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 246 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 247 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 248 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 249 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 250 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 251 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 252 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 253 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 254 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 255 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 256 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 257 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 258 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 259 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 260 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 261 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 262 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 263 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 264 |  iter 10 / 10 | loss 0.13\n",
      "| epoch 265 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 266 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 267 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 268 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 269 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 270 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 271 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 272 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 273 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 274 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 275 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 276 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 277 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 278 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 279 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 280 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 281 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 282 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 283 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 284 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 285 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 286 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 287 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 288 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 289 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 290 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 291 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 292 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 293 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 294 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 295 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 296 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 297 |  iter 10 / 10 | loss 0.12\n",
      "| epoch 298 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 299 |  iter 10 / 10 | loss 0.11\n",
      "| epoch 300 |  iter 10 / 10 | loss 0.11\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYUAAAEKCAYAAAD9xUlFAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4zLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvnQurowAAIABJREFUeJzt3Xd4HOW5/vHvs6tuq1rFRbLlXjA2NjYGm2JCCRgCoQcOCTUcCAmhJPwoCUnIKTmH5JyTBAgxAQIJBAgtFAcI1QRwx73KVbJlq1mS1aXV+/tj10IYWZJtrUYr3Z/r0uWdsrPPaGTdmvedececc4iIiAD4vC5ARER6DoWCiIi0UCiIiEgLhYKIiLRQKIiISAuFgoiItFAoiIhIC4WCiIi0CFsomNnjZlZkZqsPsvxfzGxl6OsTM5scrlpERKRzLFx3NJvZyUAV8JRzbmIby2cC65xze83sbOCnzrkZHW03PT3d5ebmdnm9IiK92dKlS0uccxkdrRcVrgKcc/PNLLed5Z+0mlwAZHdmu7m5uSxZsuTIihMR6WPMbHtn1uspfQrXAX/3uggRkb4ubGcKnWVmpxIMhRPbWecG4AaAoUOHdlNlIiJ9j6dnCmY2CfgDcL5zrvRg6znn5jrnpjnnpmVkdNgkJiIih8mzUDCzocBLwDedcxu9qkNERD4XtuYjM/sLMBtIN7MC4CdANIBz7hHgPmAA8LCZATQ556aFqx4REelYOK8+uryD5dcD14fr80VE5ND1lKuPRESkB+jzodAUaObZRTtoaGr2uhQREc/1+VD44yfbuOulVfx1ab7XpYiIeK7Ph8KSbXsBqG/UmYKISJ8PhfW7KwEoqar3uBIREe/16VAorKhlW2kNALsr6zyuRkTEe306FPY3HcX4feyuUCiIiPSZUNhb3cBnO/aSX1bTMm9lQTkxUT5OGZvB7so6yqobuPqJRWzas8/DSkVEvNNnQuGTzaVc8PAnnPLA+8zfWAzAioIKjhqcRE5qArsr6njsn1v4YEMxv353k8fVioh4o8+EwvTcVJ64ZjpD0xK472+reeLjrSzaWsbk7BQGJsdS0xDgofc3ExvlY96qQnaU1nS8URGRXqbPhEJmUhynjs3k/vMnsqeynp+9thaASdnJDEyOb1nvkSuPJSbKx/2vryVcT6UTEemp+kwo7HfymAxW/ORMnrl+BudMGsSpYzMZmBTXsuzUcZncdvoY3lm3h482lXhcrYhI9+pzoQAQE+Vj5qh0HrpiKqn9Ypg2LJVfXTKZ3195LADXzBpOQoyfd9bt8bhSEZHu5fmT13oCn8+46NjPHxEdE+Vjem4aH+fpTEFE+pY+eabQGbNGDWBzcbXuXxCRPkWhcBAzR6YDqAlJRPoUhcJBHDU4iUnZyfx+/mYaAxosT0T6BoXCQZgZt3xlNPlltby1ZrfX5YiIdAuFQjtmj80gxu9jVUGF16WIiHQLhUI7ovw+RmT0Y6PGQhKRPkKh0IExWYls3FPldRkiIt1CodCBMVn92VleS3V9k9eliIiEnUKhA6MyEwHIK9LZgoj0fgqFDozJ6g8En70gItLbKRQ6MDy9H0cNTuKRD7dQ1xjwuhwRkbBSKHTAzLh3znh2ltfy6vJdXpcjIhJWCoVOOG54GgC7Kmo9rkREJLwUCp0Q5feRGBdFeU2j16WIiIRV2ELBzB43syIzW32Q5WZmvzGzPDNbaWZTw1VLV0hJiKaiVqEgIr1bOM8U/gic1c7ys4HRoa8bgN+FsZYjlhIfQ3lNg9dliIiEVdhCwTk3HyhrZ5Xzgadc0AIgxcwGhaueI5WSEE25zhREpJfzsk9hCJDfarogNK9HSo6PpkJ9CiLSy3kZCtbGPNfmimY3mNkSM1tSXFwc5rLapjMFEekLvAyFAiCn1XQ20OaNAM65uc65ac65aRkZGd1S3IH29yk0N7eZWyIivYKXofAq8K3QVUjHAxXOuUIP62lXSkI0zQ6qGjQwnoj0XlHh2rCZ/QWYDaSbWQHwEyAawDn3CDAPmAPkATXANeGqpSskx0cDUFHTSFJctMfViIiER9hCwTl3eQfLHXBzuD6/q6UkxABQXtNITprHxYiIhInuaO6klITg2UF5re5VEJHeS6HQSfubjzTUhYj0ZgqFTkrZHwq6LFVEejGFQiel9ovBZ7Cnos7rUkREwkah0EnRfh9DUuPZXlbjdSkiImGjUDgEuQP6saO02usyRETCRqFwCIamJbCtVGcKItJ7KRQOQe6AflTUNmoIbRHptRQKh2DYgAQAXlhaoBFTRaRXUigcgtz0fgD82xvr+N2Hmz2uRkSk6ykUDsHQtISW1yVV9R5WIiISHgqFQxAX7WfRPacxbmAiZdXqVxCR3kehcIgyk+IYmBxH0T7dxCYivY9C4TBkJsZSVKnmIxHpfRQKhyEzMY6SqnoCegqbiPQyCoXDkJkUS7OD0mqdLYhI76JQOAyZibEAakISkV5HoXAYMkKhULxPoSAivYtC4TBkJsYBCgUR6X0UCodh/5lCoZ6tICK9jELhMMRF+xmT1Z/F28q8LkVEpEspFA7TKWMyWLS1jOr6Jq9LERHpMgqFw3TKmEwaAs0s2FLqdSkiIl1GoXCYpg9PJTbKp1AQkV5FoXCYYqP8jMrsz4Y9VV6XIiLSZRQKR2BMViKb9uzzugwRkS6jUDgCo7P6U1hRR2WdnsImIr2DQuEIjMlMBGBTqAnpk80l7Cit8bIkEZEjolA4AmOy9ofCPpxz3PinpfzmvU0eVyUicvjCGgpmdpaZbTCzPDO7q43lQ83sfTP7zMxWmtmccNbT1bJT40mI8bOusJLymkYq65rYubfW67JERA5b2ELBzPzAQ8DZwATgcjObcMBqPwKed85NAb4BPByuesLB5zOOHpLM8vxydpQFm40KKxQKIhK5wnmmcByQ55zb4pxrAJ4Fzj9gHQckhV4nA7vCWE9YTB2WyppdlWwMXYVUWFGHc3r4johEpnCGwhAgv9V0QWheaz8FrjSzAmAe8L0w1hMWU3JSaGp2vLl6NwD1Tc3srdHVSCISmcIZCtbGvAP/hL4c+KNzLhuYA/zJzL5Uk5ndYGZLzGxJcXFxGEo9fMcMTQHg3fVFLfN2lasJSUQiUzhDoQDIaTWdzZebh64Dngdwzn0KxAHpB27IOTfXOTfNOTctIyMjTOUenszEOI4aHGwB8/uCOaghtUUkUoUzFBYDo81suJnFEOxIfvWAdXYApwGY2XiCodCzTgU64c6zxgEQFxX8du7vbL7tueU8eMAlqq98tpPXVkRc14mI9BFR4dqwc67JzL4LvAX4gcedc2vM7H5giXPuVeAO4FEzu41g09LVLgJ7aU8Zk8HtZ4xh2rBUrnpiEbvK6yirbuCV5TuZODgZv89Hv1g/J4/O4NbnlgPwtcmDPa5aROTLwhYKAM65eQQ7kFvPu6/V67XArHDW0F1uOW00ADlpCWzcs4+PNhXjHOQVVTF3/mbqm5qZt6qwZf3GQDPRft07KCI9i34rdbFZI9NZsKWUf6zdA0BtY4C9NY3UNARYsKWMY3KCHdPbNRyGiPRACoUudtLodGoaAryxqpARGf1a5k/OTubrxwzmp+cdBcDm4iqamx13vrCC20NNSht27+O3727ip6+u4f/e2ehJ/SLSt4W1+agvOmHkAKJ8RrTfx4OXT2XObz4ixu/j+RtPIDbKT1Xo8Z15RVWsK6zk+SUFANx2xhguf3QBZdUNLdu6ZuZwkhOiPdkPEembFApdLDEumltOG82wAQlMGJxEVlIsWUlxxEb5AegfG8Wg5DgWbi1j0dbPn9r2/Wc/o74xwNu3nUxRZT1XPraQd9btYezARCYOSfZqd0Skj1EohMH+TmeAO84cS2pCzBeWT8pO5q01e/AZ/OqSydzx1xUs21HOeZMHMyYrkaFpCcT4ffzwhRVE+X0svvd0kuN1xiAi4ac+hTC7dFoOZ0zI+sK8/7xwEtfOGs4dZ47lqxMHtsyfMSINgLhoP5Nzkml20NDUzDuhTusD1TcF+GRzCRUaVkNEuojOFDyQ1i+G+772+YCxOWnx5JfVMmN4Wsu8cycNpq6xmbLqBt5YVcip4zJZtbOCU8Z8fkf3nz7dzr+9sY6kuCjevPVkBqfEd+t+iEjvozOFHmD8wCTS+8cyMqN/y7yrZuby2vdO5NzJg/hwYzHXPLGIqx5fxJJtZS3rvL1mD4OT49hX38Rzi/Pb2vSX1DQ0dXn9ItJ7KBR6gB+fO4Enrp6O2ZfHEPz2SSOIj/azoqACgHtfXs3zi/Mpq25gyfYyLj42m5NGZ/D8knzqGgPtfs6qggom/+xtPtlcEpb9EJHIp1DoAXLSEjg6u+0rjNL7x3LfuRM4dlgqv7xkMrsqarnzxZXc8NQSmh2cNj6La2blUlhRx8WPfEJVfRO1DQH+9Ok2GgPNNDQ1s2F38FkPv/swj8aAY8HmUpqbI240ERHpBupTiACXTs/h0unBAWcvnDKEa59czAcbirn42GwmZSdjZjxy5VRu/PMynvxkG4lxUdz3tzWUVTfypwXbKKlq4J454/h76JkPb63Zw2P/3Moj3zyWk0b3rFFnRcRbFmnjz02bNs0tWbLE6zI8VVXfxPId5cwaNeALTU7X/nExy3bsZdiAfqzILwcgLtpHUlw0RfvqiYnyMXPkAD7YEByI9uqZufz0vKPYUVrD4JQ4ojQWk0ivZWZLnXPTOlpPvwUiUP/YKE4cnf6lPog7zhxDdX0TK/LLSYgJ3ix3wZRs/mXGMAAumprNaeMyW9ZfuLWMlQXlzP7l+y13VotI36ZQ6EWOGpzMj86ZgM/gPy88mtwBCVx/0nCuPH4oc44eyHe/MopjclIByE6NZ11hJfe8vIpmx0E7nzcXV7UMzSEivZ+aj3qhiprGdsdM+jivhGbn+OZjiwBIjo8m2m8kx0dz19njW262awo0c8z9/+DaWbncfubYbqldRMKjs81H6mjuhToaRG/WqHQaA83cNHskU3JSKKyo4yevrqGkqoG31+xuCYUdZTVU1TexTcN8i/QZCoU+Ktrv4/+FHiO6dldly/zl+eW8u24Pk3NS2FJcDcDuijo+2lTM8PR+ZKcmeFKviHQP9SkI4wYmcvsZY7j42Gw2FVVx3ZNLuOvFVWwurgJgZ3kt335qCQ+9v9njSkUk3BQKgs9n3HLa6C88N/qddXt4cVnwiqSd5bXUNTazvbTaqxJFpJt0KhTM7PtmlmRBj5nZMjM7M9zFSfc6JjuFmCgf18zKZXByHBv3VH1h+Y6yYN/CP9bu+UKTk4j0Hp09U7jWOVcJnAlkANcAvwhbVeKJ5IRo/nHbydw7Zzw/PCt4tVF6/8+fBVFYUUdDUzO3Pbech97P86pMEQmjznY0779Lag7whHNuhbU1eptEvGEDgs+VPn/yENbuqmRwSjw/e20tAIFmx8KtpVTVN1FQXutlmSISJp09U1hqZm8TDIW3zCwRaA5fWeI1n8+495wJzDl6EAAxUcEflTdD4yft3KtQEOmNOhsK1wF3AdOdczVANMEmJOnl0vvHEuWzlgcAvbUmGAolVfUdDtUtIpGns6FwArDBOVduZlcCPwIqwleW9BR+n/HAJZP48bkTiPH7KKlqaFm2S01IIr1OZ0Phd0CNmU0G7gS2A0+FrSrpUS6Yks2YrESunpULQFZSLBC8VFVEepfOdjQ3OeecmZ0P/No595iZXRXOwqTnuWfOeM6eOBCACx7+RP0KIr1QZ0Nhn5ndDXwTOMnM/AT7FaSPmTI0lcZA8BqDu15aRcC5lqG5RSTydbb56DKgnuD9CruBIcADYatKerRov4/MxGAT0r+9vo6SqnqPKxKRrtKpUAgFwdNAspmdC9Q55zrsUzCzs8xsg5nlmdldB1nnUjNba2ZrzOyZQ6pePPP8v57AM9fPoL4pwO8/DI6J1NzsiLSh2EXkizo7zMWlwCLgEuBSYKGZXdzBe/zAQ8DZwATgcjObcMA6o4G7gVnOuaOAWw95D8QTuen9mDkqnbMmDuSlZTtpDDTzL39YyK3PLfe6NBE5Ap3tU7iX4D0KRQBmlgG8A7zQznuOA/Kcc1tC73kWOB9Y22qdbwMPOef2AuzfvkSO848ZwrxVu7nrxVV8uqWUhBg/DU3NLTe7iUhk6ez/XN8Bv7BLO/HeIUB+q+mC0LzWxgBjzOxjM1tgZme1tSEzu8HMlpjZkuLi4k6WLN1h9tgMkuKieHFZAYmxUdQ0BPhsx16vyxKRw9TZM4U3zewt4C+h6cuAeR28p62xkQ5scI4CRgOzgWzgIzOb6Jwr/8KbnJsLzIXg4zg7WbN0g9goP7+9Yio799Zy0uh0TnngfT7OK2HGiAFelyYih6FToeCc+6GZXQTMIvjLfq5z7uUO3lYA5LSazgZ2tbHOAudcI7DVzDYQDInFnalLeoZTxmS0vD4mJ4U3VhXy/dPH4PdpzESRSNPphl/n3IvOududc7d1IhAg+It9tJkNN7MY4BvAqwes8wpwKoCZpRNsTtrS2Zqk57n2xOFsLq5m3qpCr0sRkcPQbiiY2T4zq2zja5+ZtfuUFedcE/Bd4C1gHfC8c26Nmd1vZueFVnsLKDWztcD7wA+dc6VHvlvilTkTBzE6sz+P/XOr16WIyGFot/nIOZd4JBt3zs3jgL4H59x9rV474PbQl/QCPp9xzqRB/PrdTZTXNJCSENPxm0Skx9B1g9LlThqdjnPwcV6pbmYTiTAKBelyk7NTALj5mWVc9YSuGRCJJAoF6XJRfh/fPD44SN78jcXkl9V4XJGIdJZCQcLi51+fyPwfngrAG7oSSSRiKBQkbIYOSGBydjKvrzzw9hQR6akUChJW504azOqdlWwrqfa6FBHpBIWChNU5kwYB8OhHW9hcXOVxNSLSEYWChNXglHimDUvl6YU7uPSRT6lvCnhdkoi0Q6EgYffLSyZz6+mjKa1u4O01e7wuR0TaoVCQsMtN78ctXxlNdmo8c+dvobymweuSROQgFArSLXw+486zxrF+dyWXPPIpTYFmr0sSkTYoFKTbnDd5ML+9fAqbiqp46bOdXpcjIm1QKEi3+upRA5k4JInffbBZ4yKJ9EAKBelWZsaVM4axtaSatYXtjr4uIh5QKEi3O31CFj6Dt3QlkkiPo1CQbpfeP5Zpw9J4dflOXYkk0sMoFMQTN84ewa7yOi5/dKH6FkR6EIWCeOIr47L48bnjWVdYycY9Gv5CpKdQKIhnTp+QBcD7G4o8rkRE9lMoiGcGJcczbmAiHygURHoMhYJ46swJWSzcWsb764vUtyDSAygUxFM3zR7F2KxErvnjYs76v4+orm/yuiSRPk2hIJ6Kj/Hz1HXHcevpo9mwZx9PfrrN65JE+jSFgnguMzGOW08fw6ljM/j9h1vYVV7rdUkifZZCQXqMH587gUCz46anl9HcrP4FES8oFKTHGJHRnx+cOYYV+eVsK9UznUW8oFCQHuX4kQMAWFFQ7nElIn2TQkF6lNGZiSTE+FmRX+F1KSJ9UlhDwczOMrMNZpZnZne1s97FZubMbFo465Gez+8zJg5JZnm+zhREvBC2UDAzP/AQcDYwAbjczCa0sV4icAuwMFy1SGQ5JieFtYWV7Cit4ZPNJbqpTaQbhfNM4Tggzzm3xTnXADwLnN/Gej8H/huoC2MtEkEumpoNwFd+9QFXPLqQm/68jIraRo+rEukbwhkKQ4D8VtMFoXktzGwKkOOcez2MdUiEGTswkZ+ffxSDU+K57sThvLNuD+c9+E/d7SzSDaLCuG1rY15LO4CZ+YD/Ba7ucENmNwA3AAwdOrSLypOe7LLpQ7lsevBYT89N5cY/L+OTzaWcERpZVUTCI5xnCgVATqvpbGBXq+lEYCLwgZltA44HXm2rs9k5N9c5N805Ny0jIyOMJUtPdOq4TOKj/XycV+J1KSK9XjhDYTEw2syGm1kM8A3g1f0LnXMVzrl051yucy4XWACc55xbEsaaJALFRvk5bngaH20q9roUkV4vbKHgnGsCvgu8BawDnnfOrTGz+83svHB9rvROJ41OZ3NxNS8sLfC6FJFeLZx9Cjjn5gHzDph330HWnR3OWiSyXTo9h3fW7eEHf13ByIx+TBma6nVJIr2S7miWiJAUF81jV00nMTaKn722luufXMze6gavyxLpdRQKEjH6xUZx8bRslueX8866It5br8d4inQ1hYJElO/MHsW/njwCgE+3lHpcjUjvE9Y+BZGulpEYy91zxrOttJoFCgWRLqczBYlIJ4wYQMHeWn7yt9W601mkC+lMQSLSnEmD+GhTCU9+up3YaD/nThrE0UOSMWvrRnoR6SyLtBEop02b5pYs0f1tEnTbc8t5+bOdAHz/tNFsLq7injnjGZwS73FlIj2LmS11znX4eAKdKUhEu/vscQCs372PX7+7CYCxWYl877TRXpYlErHUpyARLTMpjv+97Bj+66KjyUyMBeCfGiNJ5LApFKRXmJSdwsJ7TuOm2SNZuLWMZxbuoK4x4HVZIhFHoSC9hpkxe0xwFN17Xl7Fg+/lMXf+Zkqq6j2uTCRyqE9BepXjhqfxxDXTeXT+Fh58Pw+A/LJafv71iR5XJhIZdKYgvYqZcerYTL4f6mhOjI3ixWUFepynSCcpFKRXmjFiAO/cfjJ/un4GNQ0BZv7nu7oDWqQTFArSa43KTOSYnBSe+fYMEmKjeHT+Fq9LEunxFArS680cmc6FU4bw4cZiStXpLNIuhYL0CRdMHUJTs+Objy3i208t4e+rCr0uSaRHUihInzBuYBL/c+lkHLCqoIKbnl7G//5jI5E2zItIuOmSVOkzLpyazYVTs2loaubel1fx63c3UdsYoCngKKmqJyk+ikun5TApO8XrUkU8o1CQPicmysd/XTSJ6Cgfc+dvIcpnZKfGU7yvnr99tov/d/Y45hw9iLR+MV6XKtLtNEqq9FnOOf6yKJ+JQ5KYlJ3CrvJavjF3ATvKahg/KIlXbp5JbJTf6zJFukRnR0lVn4L0WWbGFTOGtjQXDU6J5707TuGhK6ayrrCSH/x1pR7gI32Omo9EWony+zhn0iC2lY7ll29voKa+iceunu51WSLdRqEg0oabTx1FjN/Hv89bx2srdvHe+iKyU+O548yxXpcmElYKBZGDuHpWLi8sLeB7f/msZd6CLaXkpCXw43MmYAYpCeqMlt5FHc0i7aiobeTB9zYxMDmeJz7eSvG+euqbmonyGRmJsbxy8yyykuK8LlOkQ53taFYoiHRSaVU9PjMemb+Zgr21vL++iKZmx1ePGsgDF08iLlpXKknPpWc0i3SxAf2Dj/u8++zxAKwsKOf5Jfk8vXAHn24uIb1/LENS4jlxdDpXz8zFzLwsV+SwKBREDtOk7BQmZadw4qh03l6zh8q6RraV1vDua2v556YSRmT041sn5JKTluB1qSKdFtbmIzM7C/g14Af+4Jz7xQHLbweuB5qAYuBa59z29rap5iPpyZxz/Pz1dby6YhcVtQ3ERfl58TszGZ7ej2j/F28Lyi+r4W/Ld/Kd2aPw+XRWIeHleZ+CmfmBjcAZQAGwGLjcObe21TqnAgudczVmdhMw2zl3WXvbVShIpNhRWsMFD39MaXUDqQnRzByVztbiap6/8QQCAcd/zFvHc0vymfvNYznzqIFelyu9XE/oUzgOyHPObQkV9CxwPtASCs6591utvwC4Moz1iHSroQMSePrbM3hjZSH/WLuHv68qpNnBV/93PiVV9fhDZwePfrSF08dn6WxBeoRwDnMxBMhvNV0Qmncw1wF/b2uBmd1gZkvMbElxcXEXligSXuMGJnHHmWN55eZZfPjDUzltXCY7y2vx+4yahgBnTxzI4m17+frDH/P84nwN5S2eC+eZQlt/9rT5E29mVwLTgFPaWu6cmwvMhWDzUVcVKNJd4qL95KQl8IuLJrFoaxkjM/vx7roibjxlJC9/tpOHP8jjzhdXsqOshu+fPpoon7FsRzlrCyu5dFq2BuaTbhPOUCgAclpNZwO7DlzJzE4H7gVOcc7pWYnSq2UkxnLOpEFA8CwC4OJjs7lo6hDufGElD76fx+Mfb2VA/xjyy2oBWJFfzvTcVIan92fasFQ1M0lYhbOjOYpgR/NpwE6CHc1XOOfWtFpnCvACcJZzblNntquOZumtmgLNvLNuD59uLmVraQ3nThrEmp0VPPnp5xfkHZOTwm8vn0K030dKQnS7N8xV1zeREOPX/RIC9ICrj0JFzAH+j+AlqY875/7dzO4HljjnXjWzd4Cjgf0PzN3hnDuvvW0qFKQvaQw08976IkZm9Gfp9jL+Y956GpqaqW0MEO03rjtxBJmJsTzxyVZOHJXB2l0VZKcmcNGxQ/j+X5ZzbG4qD14xlf6xuiWpr+sRoRAOCgXpyzbs3sf9r6/h+OED2FxcxSvLgy2yw9P7sbWkmtwBCZRVN1BZ10S/GD91Tc0cPyKNh684lmcW7WDGiDSWbtvLxCHJnDBygMd7I91JoSDSB2wursI5GJnRjy0l1eSkJrC7oo67XlrJt07Ipaq+iR/8dUXL+tF+ozHgSIqL4s1bT6ZoXz31jQG2lVYzKjORY4elerg3Ek4KBREB4N11e1i1s4KxWYnc//pajhqczCebSwCoaQi0rDcoOY7zjhlMeXUj1500nH6xUdTUNzEqsz95RVVU1TcxZahCI1IpFETkSxoDwWG/1xXu44mPt5KTlsDYgYkU7K3l568H7yv1+4xAs8MMnIOspFhKqxoIOMdxuWmMH5TED746lsraRmKifNQ2BDS+UwRQKIhIpznnuOFPS4mL9vOTr03gucX5BJodmYmxfJRXQkp8NGawcEsZm4qqiI/2U9sYIMpnNDvHJcfmMCw9gZkj06mpb+L4EcH+iuqGJhLjoj3eOwGFgogcIudch5ev7h/wb+HWUs46aiDVDQH2VNYxb1Uh9U3NLeuNyepPlM/H5uIqnr5+BtUNAX777iYumZZNwd5aov0+rj9pOHWNzby4tIDLZwzVFVJhplAQkW61cc8+1u6qpCHQzAtLC6ioaaSmsYny6kYcUNPQRLP7vHnqhBEDqKpvYtXOCs6YkMW9c8Zzz8uruPvs8TQEAgxN60fm8O5bAAAL2ElEQVRGYmzL9gv21pCVFEeg2fHLtzZw8pgMTh6T4d0ORxiFgoh4Lr+shl+8uZ4txdU8eMUU9lY3cNTgZF5dsZMfvbKaKJ+PsyYO5OXPdpLeP4aSqgZionw0NDVjBl+bNJixAxNJjo/mx39bzciM/mQlxfJxXilmcPfZ47jh5JFsLq5iwZZSpuSkMn5QIn9euIPcAQnERft58L08Thg5gBtPGdlmjZ9sLqGuMcBXxmV183eneykURKRHq28KXvkU4/fxg7+u5MVlBcw5eiCLtpbxzeNzqapv5M8LdlDbGFxvbFYiPp+xtaSK78wexYbd+3hjVSFDUuLZWR4cEsTvMyZnJ7NsR3nL50T5DAfMu+Ukxg5M/EIN63dXcv6DH+Mz441bTiQ22s+QlHiWbt/L5uIqLp2Ww6Gobwr02HGqFAoiEjHqGgO8t76I08dnEe23lr6N5mZHSVU9L322kwumDCErKa6l7yPQ7Hji4618ll/OxMHJfGVcJs8u3sH7oe34fUZqvxjOmzyYc37zEX6fMWxAP0qr6mloaiYjMZatJdX4fcbemkb8PiPab9wzZzy/eXcTJVUNPHDxJGKifGSnJjB1aAoA1Q2Blv6Pl5YV8NSn2/njNdN5dcUuHnhrA6/cPIuRGf275PtSsLeGgUlxRPmPfEBrhYKISMjaXZX8+7y1VNUHGJqWQIzfR15xFSnx0fzkaxP41dsbWZ5fzqDkOJZs34vfZwzoF0PRvs/H6DwuN42k+Cg+3FjMyIz+7K6so6Y+QEOgmTlHD+S99UXUNTZzybHZ/OeFR7N0+17W797Hm6t385PzJhAX5Sc+xk9WUhx5Rfv4rzc3cMb4LNL6xTBjRFrLVVr1TQGeWbiDnNQEbvzzUs6ZNIhff2PKEX8PFAoiIp3UFGjGzDDg3fVFOOfITk1g6fYyjhs+gEVbS/nl2xupqG3kjAlZlNc0MDA5nj2VdaQlxPDmmt1kJMZyXG4ab67ZzdC0BLaWVAPB5rGm5maaQ79qj8lJIa+oitrGAIHQzP6xUSTE+JkxYgB5RVWsK6z8Qn2/vGQy03NTifb7GJwSf1j7qFAQEelCRfvqyC+r4dhhaV+YX1nXyIcbijltfCaNAccv/r6e1TsruGpmLuMGJhIf4+c3725iWm4aFTUNvL6ykPGDkrj9jDEUVtRR3xTg9RWF1DYG+GBDEQOT47h0Wg5PfbqdG08ZyWsrdvHpllIArj9xOD86d8Jh1a9QEBHpBeoaA/zug80kxPg5d/JghoT5TEF3i4iI9GBx0X5uO2NMt31eOJ/RLCIiEUahICIiLRQKIiLSQqEgIiItFAoiItJCoSAiIi0UCiIi0kKhICIiLSLujmYzKwa2H+bb04GSLizHS9qXnkn70jNpX2CYc67DpxJFXCgcCTNb0pnbvCOB9qVn0r70TNqXzlPzkYiItFAoiIhIi74WCnO9LqALaV96Ju1Lz6R96aQ+1acgIiLt62tnCiIi0o4+EwpmdpaZbTCzPDO7y+t6DpWZbTOzVWa23MyWhOalmdk/zGxT6N9Ur+tsi5k9bmZFZra61bw2a7eg34SO00ozm+pd5V92kH35qZntDB2b5WY2p9Wyu0P7ssHMvupN1V9mZjlm9r6ZrTOzNWb2/dD8iDsu7exLJB6XODNbZGYrQvvys9D84Wa2MHRcnjOzmND82NB0Xmh57hEX4Zzr9V+AH9gMjABigBXABK/rOsR92AakHzDvv4G7Qq/vAv7L6zoPUvvJwFRgdUe1A3OAvwMGHA8s9Lr+TuzLT4EftLHuhNDPWiwwPPQz6Pd6H0K1DQKmhl4nAhtD9UbccWlnXyLxuBjQP/Q6GlgY+n4/D3wjNP8R4KbQ6+8Aj4RefwN47khr6CtnCscBec65Lc65BuBZ4HyPa+oK5wNPhl4/CXzdw1oOyjk3Hyg7YPbBaj8feMoFLQBSzGxQ91TasYPsy8GcDzzrnKt3zm0F8gj+LHrOOVfonFsWer0PWAcMIQKPSzv7cjA9+bg451xVaDI69OWArwAvhOYfeFz2H68XgNPMzI6khr4SCkOA/FbTBbT/Q9MTOeBtM1tqZjeE5mU55woh+B8DyPSsukN3sNoj9Vh9N9Ss8nirZryI2JdQk8MUgn+VRvRxOWBfIAKPi5n5zWw5UAT8g+CZTLlzrim0Sut6W/YltLwCGHAkn99XQqGt5Iy0y65mOeemAmcDN5vZyV4XFCaReKx+B4wEjgEKgV+F5vf4fTGz/sCLwK3Oucr2Vm1jXk/fl4g8Ls65gHPuGCCb4BnM+LZWC/3b5fvSV0KhAMhpNZ0N7PKolsPinNsV+rcIeJngD8ue/afwoX+LvKvwkB2s9og7Vs65PaH/yM3Ao3zeFNGj98XMogn+En3aOfdSaHZEHpe29iVSj8t+zrly4AOCfQopZhYVWtS63pZ9CS1PpvPNm23qK6GwGBgd6sGPIdgh86rHNXWamfUzs8T9r4EzgdUE9+Gq0GpXAX/zpsLDcrDaXwW+Fbra5XigYn9zRk91QNv6BQSPDQT35RuhK0SGA6OBRd1dX1tC7c6PAeucc//TalHEHZeD7UuEHpcMM0sJvY4HTifYR/I+cHFotQOPy/7jdTHwngv1Oh82r3vbu+uL4NUTGwm2z93rdT2HWPsIgldLrADW7K+fYNvhu8Cm0L9pXtd6kPr/QvD0vZHgXzbXHax2gqfDD4WO0ypgmtf1d2Jf/hSqdWXoP+mgVuvfG9qXDcDZXtffqq4TCTYzrASWh77mROJxaWdfIvG4TAI+C9W8GrgvNH8EweDKA/4KxIbmx4Wm80LLRxxpDbqjWUREWvSV5iMREekEhYKIiLRQKIiISAuFgoiItFAoiIhIC4WCiIi0UChIRDKzT0L/5prZFV287Xva+qxwMbOvm9l9HaxzSWgo5WYzm3bAsi8NA21mMWY2v9VdsCKdolCQiOScmxl6mQscUiiYmb+DVb4QCq0+K1zuBB7uYJ3VwIXA/NYzzWwCwTv0jwLOAh42M78Ljgb8LnBZ15crvZlCQSKSme0fXvgXwEmhh6jcFhph8gEzWxwaHfNfQ+vPDj2I5RmCd7liZq+ERp1ds3/kWTP7BRAf2t7TrT8rNMTDA2a22oIPPLqs1bY/MLMXzGy9mT29f/hiM/uFma0N1fLLNvZjDFDvnCsJTf/NzL4Vev2v+2twzq1zzm1o41vR3jDQrwD/ckTfaOlzdGopke4ugg9SORcg9Mu9wjk33cxigY/N7O3QuscBE0O/PAGudc6VhcaYWWxmLzrn7jKz77rgKJUHupDgiJuTgfTQe/b/5T6F4F/ru4CPgVlmtpbgmDvjnHNu/5g2B5gFLGs1fUOo5q3AHQQHQ2vPEGBBq+nWwyqvBqZ38H6RL9CZgvQ2ZxIcuG05wTH1BxAc8AxgUatAALjFzFYQ/KWa02q9gzkR+IsLjry5B/iQz3/pLnLOFbjgiJzLCTZrVQJ1wB/M7EKgpo1tDgKK90+EtnsfwQHQ7nDOdTTi5UGHTnbOBYCG/YMpinSGQkF6GwO+55w7JvQ13Dm3/0yhumUls9kER6A8wTk3meAgZHGd2PbB1Ld6HQCiXPChJ8cRHNL568Cbbbyvto3PPRooBQZ3UA90PAx0LMFgEukUhYJEun0En8u731vATaHx9TGzMaHhxg+UDOx1ztWY2Ti+2EzTuP/9B5gPXBbqt8gg+Lzmgw65HHroS7Jzbh5wK8GmpwOtA0a1es9xBB+kNAX4QWho5/YcdBhoMxsAFDvnGjvYhkgLhYJEupVAk5mtMLPbgD8Aa4FlZrYa+D1t9529CUSZ2Urg53yxXX4usHJ/J28rL4c+bwXwHnCnc253O7UlAq+HPuND4LY21pkPTAl1YscSfBjMtS74UKU7gMdDyy4wswLgBOANM3sLwDm3huBD3deG9unmULMRwKnAvHbqE/kSDZ0t4jEz+zXwmnPunS7e7kvA3Qe5akmkTTpTEPHefwAJXblBCz5h8BUFghwqnSmIiEgLnSmIiEgLhYKIiLRQKIiISAuFgoiItFAoiIhIi/8PqQ/Lq+I2wAQAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# coding: utf-8\n",
    "import sys\n",
    "sys.path.append('..')  # 为导入父目录中的文件而进行的设置\n",
    "import numpy as np\n",
    "from common.optimizer import SGD\n",
    "from dataset import spiral\n",
    "import matplotlib.pyplot as plt\n",
    "#from two_layer_net import TwoLayerNet\n",
    "\n",
    "\n",
    "# 设置超参数\n",
    "max_epoch = 300\n",
    "batch_size = 30\n",
    "hidden_size = 10\n",
    "learning_rate = 1.0\n",
    "\n",
    "x, t = spiral.load_data()\n",
    "model = TwoLayerNet(input_size=2, hidden_size=hidden_size, output_size=3)\n",
    "optimizer = SGD(lr=learning_rate)\n",
    "\n",
    "# 学习时使用的变量\n",
    "data_size = len(x)\n",
    "max_iters = data_size // batch_size\n",
    "total_loss = 0\n",
    "loss_count = 0\n",
    "loss_list = []\n",
    "\n",
    "for epoch in range(max_epoch):\n",
    "    # 打乱数据\n",
    "    idx = np.random.permutation(data_size)\n",
    "    x = x[idx]\n",
    "    t = t[idx]\n",
    "\n",
    "    for iters in range(max_iters):\n",
    "        batch_x = x[iters*batch_size:(iters+1)*batch_size]\n",
    "        batch_t = t[iters*batch_size:(iters+1)*batch_size]\n",
    "\n",
    "        # 求梯度、更新参数\n",
    "        loss = model.forward(batch_x, batch_t)\n",
    "        model.backward()\n",
    "        optimizer.update(model.params, model.grads)\n",
    "\n",
    "        total_loss += loss\n",
    "        loss_count += 1\n",
    "\n",
    "        # 定期输出学习进度\n",
    "        if (iters+1) % 10 == 0:\n",
    "            avg_loss = total_loss / loss_count\n",
    "            print('| epoch %d |  iter %d / %d | loss %.2f'\n",
    "                  % (epoch + 1, iters + 1, max_iters, avg_loss))\n",
    "            loss_list.append(avg_loss)\n",
    "            total_loss, loss_count = 0, 0\n",
    "\n",
    "\n",
    "# 显示学习结果\n",
    "plt.plot(np.arange(len(loss_list)), loss_list, label='train')\n",
    "plt.xlabel('iterations (x10)')\n",
    "plt.ylabel('loss')\n",
    "plt.show()\n",
    "\n",
    "# 显示边界区域\n",
    "h = 0.001\n",
    "x_min, x_max = x[:, 0].min() - .1, x[:, 0].max() + .1\n",
    "y_min, y_max = x[:, 1].min() - .1, x[:, 1].max() + .1\n",
    "xx, yy = np.meshgrid(np.arange(x_min, x_max, h), np.arange(y_min, y_max, h))\n",
    "X = np.c_[xx.ravel(), yy.ravel()]\n",
    "score = model.predict(X)\n",
    "predict_cls = np.argmax(score, axis=1)\n",
    "Z = predict_cls.reshape(xx.shape)\n",
    "plt.contourf(xx, yy, Z)\n",
    "plt.axis('off')\n",
    "\n",
    "# 绘制数据点\n",
    "x, t = spiral.load_data()\n",
    "N = 100\n",
    "CLS_NUM = 3\n",
    "markers = ['o', 'x', '^']\n",
    "for i in range(CLS_NUM):\n",
    "    plt.scatter(x[i*N:(i+1)*N, 0], x[i*N:(i+1)*N, 1], s=40, marker=markers[i])\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
}