mobilenet_v1.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "# PyTorch for CIFAR10\n",
    "We test ResNet classifier.\n",
    "First init some basic environment."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "%matplotlib inline\n",
    "plt.rcParams['figure.figsize'] = (10.0, 8.0) # set default size of plots\n",
    "plt.rcParams['image.interpolation'] = 'nearest'\n",
    "plt.rcParams['image.cmap'] = 'gray'\n",
    "\n",
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# mobilenetV1 classifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "MobileNetV1(\n",
      "  (layers): Sequential(\n",
      "    (0): Sequential(\n",
      "      (0): Conv2d(3, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (2): ReLU6(inplace)\n",
      "    )\n",
      "    (1): Sequential(\n",
      "      (0): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)\n",
      "      (1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (2): ReLU6(inplace)\n",
      "      (3): Conv2d(32, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (4): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (5): ReLU6(inplace)\n",
      "    )\n",
      "    (2): Sequential(\n",
      "      (0): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=64, bias=False)\n",
      "      (1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (2): ReLU6(inplace)\n",
      "      (3): Conv2d(64, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (4): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (5): ReLU6(inplace)\n",
      "    )\n",
      "    (3): Sequential(\n",
      "      (0): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=128, bias=False)\n",
      "      (1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (2): ReLU6(inplace)\n",
      "      (3): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (4): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (5): ReLU6(inplace)\n",
      "    )\n",
      "    (4): Sequential(\n",
      "      (0): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=128, bias=False)\n",
      "      (1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (2): ReLU6(inplace)\n",
      "      (3): Conv2d(128, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (4): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (5): ReLU6(inplace)\n",
      "    )\n",
      "    (5): Sequential(\n",
      "      (0): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256, bias=False)\n",
      "      (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (2): ReLU6(inplace)\n",
      "      (3): Conv2d(256, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (4): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (5): ReLU6(inplace)\n",
      "    )\n",
      "    (6): Sequential(\n",
      "      (0): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=256, bias=False)\n",
      "      (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (2): ReLU6(inplace)\n",
      "      (3): Conv2d(256, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (4): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (5): ReLU6(inplace)\n",
      "    )\n",
      "    (7): Sequential(\n",
      "      (0): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512, bias=False)\n",
      "      (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (2): ReLU6(inplace)\n",
      "      (3): Conv2d(512, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (4): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (5): ReLU6(inplace)\n",
      "    )\n",
      "    (8): Sequential(\n",
      "      (0): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512, bias=False)\n",
      "      (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (2): ReLU6(inplace)\n",
      "      (3): Conv2d(512, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (4): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (5): ReLU6(inplace)\n",
      "    )\n",
      "    (9): Sequential(\n",
      "      (0): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512, bias=False)\n",
      "      (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (2): ReLU6(inplace)\n",
      "      (3): Conv2d(512, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (4): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (5): ReLU6(inplace)\n",
      "    )\n",
      "    (10): Sequential(\n",
      "      (0): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512, bias=False)\n",
      "      (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (2): ReLU6(inplace)\n",
      "      (3): Conv2d(512, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (4): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (5): ReLU6(inplace)\n",
      "    )\n",
      "    (11): Sequential(\n",
      "      (0): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512, bias=False)\n",
      "      (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (2): ReLU6(inplace)\n",
      "      (3): Conv2d(512, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (4): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (5): ReLU6(inplace)\n",
      "    )\n",
      "    (12): Sequential(\n",
      "      (0): Conv2d(512, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=512, bias=False)\n",
      "      (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (2): ReLU6(inplace)\n",
      "      (3): Conv2d(512, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (4): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (5): ReLU6(inplace)\n",
      "    )\n",
      "    (13): Sequential(\n",
      "      (0): Conv2d(1024, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1024, bias=False)\n",
      "      (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (2): ReLU6(inplace)\n",
      "      (3): Conv2d(1024, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (4): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      (5): ReLU6(inplace)\n",
      "    )\n",
      "  )\n",
      "  (avg_pool): AvgPool2d(kernel_size=4, stride=4, padding=0)\n",
      "  (classifier): Sequential(\n",
      "    (0): Linear(in_features=1024, out_features=10, bias=True)\n",
      "  )\n",
      ")\n",
      "params num: 3217226\n"
     ]
    }
   ],
   "source": [
    "import torch.optim as optim\n",
    "import torch.nn as nn\n",
    "import time\n",
    "from cifar10.tnt_solver import *\n",
    "from cifar10.classifiers.mobilenet import MobileNetV1\n",
    "\n",
    "model = MobileNetV1(num_classes=10)\n",
    "print(model)\n",
    "num_params = count_parameters(model)\n",
    "print('params num:', num_params)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# train "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Files already downloaded and verified\n",
      "Files already downloaded and verified\n",
      "Files already downloaded and verified\n",
      "Epoch: 1/150, lr:1.00e-01\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.72it/s, loss:1.6046, acc:40.7975%]\n",
      "Val loss: 1.4866, accuracy: 49.36%\n",
      "Epoch: 2/150, lr:1.00e-01\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.77it/s, loss:1.1188, acc:59.9875%]\n",
      "Val loss: 1.0533, accuracy: 62.78%\n",
      "Epoch: 3/150, lr:1.00e-01\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.75it/s, loss:0.8686, acc:69.7350%]\n",
      "Val loss: 0.8882, accuracy: 71.43%\n",
      "Epoch: 4/150, lr:1.00e-01\n",
      "100%|██████████| 625/625 [00:44<00:00, 13.94it/s, loss:0.7434, acc:74.5650%]\n",
      "Val loss: 0.6856, accuracy: 77.11%\n",
      "Epoch: 5/150, lr:1.00e-01\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.27it/s, loss:0.6552, acc:77.6625%]\n",
      "Val loss: 0.6193, accuracy: 79.25%\n",
      "Epoch: 6/150, lr:1.00e-01\n",
      "100%|██████████| 625/625 [00:46<00:00, 13.39it/s, loss:0.5981, acc:79.6150%]\n",
      "Val loss: 0.5879, accuracy: 80.12%\n",
      "Epoch: 7/150, lr:1.00e-01\n",
      "100%|██████████| 625/625 [00:44<00:00, 13.99it/s, loss:0.5585, acc:81.0375%]\n",
      "Val loss: 0.6320, accuracy: 79.16%\n",
      "Epoch: 8/150, lr:1.00e-01\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.76it/s, loss:0.5231, acc:82.0050%]\n",
      "Val loss: 0.5646, accuracy: 81.40%\n",
      "Epoch: 9/150, lr:1.00e-01\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.34it/s, loss:0.4952, acc:83.1550%]\n",
      "Val loss: 0.5330, accuracy: 81.67%\n",
      "Epoch: 10/150, lr:1.00e-01\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.67it/s, loss:0.4743, acc:83.8550%]\n",
      "Val loss: 0.5243, accuracy: 82.33%\n",
      "Epoch: 11/150, lr:1.00e-01\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.69it/s, loss:0.4552, acc:84.4000%]\n",
      "Val loss: 0.4602, accuracy: 84.69%\n",
      "Epoch: 12/150, lr:1.00e-01\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.72it/s, loss:0.4338, acc:85.2000%]\n",
      "Val loss: 0.4419, accuracy: 85.05%\n",
      "Epoch: 13/150, lr:5.00e-02\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.17it/s, loss:0.3284, acc:88.8100%]\n",
      "Val loss: 0.3848, accuracy: 87.42%\n",
      "Epoch: 14/150, lr:5.00e-02\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.20it/s, loss:0.3014, acc:89.6125%]\n",
      "Val loss: 0.3527, accuracy: 87.88%\n",
      "Epoch: 15/150, lr:5.00e-02\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.78it/s, loss:0.2856, acc:90.3950%]\n",
      "Val loss: 0.3769, accuracy: 87.50%\n",
      "Epoch: 16/150, lr:5.00e-02\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.38it/s, loss:0.2791, acc:90.2175%]\n",
      "Val loss: 0.3814, accuracy: 87.85%\n",
      "Epoch: 17/150, lr:5.00e-02\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.28it/s, loss:0.2756, acc:90.4450%]\n",
      "Val loss: 0.3707, accuracy: 87.77%\n",
      "Epoch: 18/150, lr:5.00e-02\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.86it/s, loss:0.2660, acc:90.7950%]\n",
      "Val loss: 0.3736, accuracy: 87.56%\n",
      "Epoch: 19/150, lr:5.00e-02\n",
      "100%|██████████| 625/625 [00:44<00:00, 13.97it/s, loss:0.2628, acc:90.7900%]\n",
      "Val loss: 0.3610, accuracy: 88.48%\n",
      "Epoch: 20/150, lr:5.00e-02\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.09it/s, loss:0.2617, acc:90.7225%]\n",
      "Val loss: 0.3679, accuracy: 88.14%\n",
      "Epoch: 21/150, lr:5.00e-02\n",
      "100%|██████████| 625/625 [00:42<00:00, 14.54it/s, loss:0.2529, acc:91.2975%]\n",
      "Val loss: 0.3922, accuracy: 87.47%\n",
      "Epoch: 22/150, lr:5.00e-02\n",
      "100%|██████████| 625/625 [00:46<00:00, 13.54it/s, loss:0.2575, acc:91.0450%]\n",
      "Val loss: 0.3773, accuracy: 87.88%\n",
      "Epoch: 23/150, lr:5.00e-02\n",
      "100%|██████████| 625/625 [00:44<00:00, 13.97it/s, loss:0.2478, acc:91.2325%]\n",
      "Val loss: 0.3373, accuracy: 88.50%\n",
      "Epoch: 24/150, lr:2.50e-02\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.38it/s, loss:0.1715, acc:94.2900%]\n",
      "Val loss: 0.2902, accuracy: 90.86%\n",
      "Epoch: 25/150, lr:2.50e-02\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.09it/s, loss:0.1454, acc:94.9800%]\n",
      "Val loss: 0.3245, accuracy: 90.38%\n",
      "Epoch: 26/150, lr:2.50e-02\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.50it/s, loss:0.1422, acc:95.1225%]\n",
      "Val loss: 0.3211, accuracy: 90.28%\n",
      "Epoch: 27/150, lr:2.50e-02\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.30it/s, loss:0.1376, acc:95.2375%]\n",
      "Val loss: 0.3186, accuracy: 90.45%\n",
      "Epoch: 28/150, lr:2.50e-02\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.89it/s, loss:0.1299, acc:95.4900%]\n",
      "Val loss: 0.3434, accuracy: 89.89%\n",
      "Epoch: 29/150, lr:2.50e-02\n",
      "100%|██████████| 625/625 [00:47<00:00, 13.03it/s, loss:0.1268, acc:95.5100%]\n",
      "Val loss: 0.3472, accuracy: 89.75%\n",
      "Epoch: 30/150, lr:2.50e-02\n",
      "100%|██████████| 625/625 [00:42<00:00, 14.54it/s, loss:0.1320, acc:95.3175%]\n",
      "Val loss: 0.3231, accuracy: 90.64%\n",
      "Epoch: 31/150, lr:2.50e-02\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.39it/s, loss:0.1257, acc:95.6625%]\n",
      "Val loss: 0.3468, accuracy: 89.86%\n",
      "Epoch: 32/150, lr:2.50e-02\n",
      "100%|██████████| 625/625 [00:46<00:00, 13.49it/s, loss:0.1223, acc:95.7125%]\n",
      "Val loss: 0.3426, accuracy: 90.64%\n",
      "Epoch: 33/150, lr:2.50e-02\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.77it/s, loss:0.1183, acc:95.8650%]\n",
      "Val loss: 0.3427, accuracy: 90.23%\n",
      "Epoch: 34/150, lr:2.50e-02\n",
      "100%|██████████| 625/625 [00:44<00:00, 13.92it/s, loss:0.1186, acc:95.7350%]\n",
      "Val loss: 0.3517, accuracy: 89.77%\n",
      "Epoch: 35/150, lr:1.25e-02\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.77it/s, loss:0.0763, acc:97.3675%]\n",
      "Val loss: 0.3247, accuracy: 91.25%\n",
      "Epoch: 36/150, lr:1.25e-02\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.61it/s, loss:0.0571, acc:98.0550%]\n",
      "Val loss: 0.3078, accuracy: 91.58%\n",
      "Epoch: 37/150, lr:1.25e-02\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.35it/s, loss:0.0520, acc:98.2525%]\n",
      "Val loss: 0.3269, accuracy: 91.56%\n",
      "Epoch: 38/150, lr:1.25e-02\n",
      "100%|██████████| 625/625 [00:44<00:00, 13.95it/s, loss:0.0546, acc:98.1000%]\n",
      "Val loss: 0.3322, accuracy: 91.58%\n",
      "Epoch: 39/150, lr:1.25e-02\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.44it/s, loss:0.0489, acc:98.3050%]\n",
      "Val loss: 0.3459, accuracy: 91.34%\n",
      "Epoch: 40/150, lr:1.25e-02\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.17it/s, loss:0.0466, acc:98.3150%]\n",
      "Val loss: 0.3580, accuracy: 91.30%\n",
      "Epoch: 41/150, lr:1.25e-02\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.87it/s, loss:0.0466, acc:98.3750%]\n",
      "Val loss: 0.3549, accuracy: 91.24%\n",
      "Epoch: 42/150, lr:1.25e-02\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.25it/s, loss:0.0446, acc:98.5100%]\n",
      "Val loss: 0.3560, accuracy: 91.59%\n",
      "Epoch: 43/150, lr:1.25e-02\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.19it/s, loss:0.0426, acc:98.5050%]\n",
      "Val loss: 0.3639, accuracy: 91.23%\n",
      "Epoch: 44/150, lr:1.25e-02\n",
      "100%|██████████| 625/625 [00:42<00:00, 14.57it/s, loss:0.0445, acc:98.3850%]\n",
      "Val loss: 0.3663, accuracy: 91.49%\n",
      "Epoch: 45/150, lr:1.25e-02\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.26it/s, loss:0.0402, acc:98.5925%]\n",
      "Val loss: 0.3649, accuracy: 91.25%\n",
      "Epoch: 46/150, lr:6.25e-03\n",
      "100%|██████████| 625/625 [00:44<00:00, 13.91it/s, loss:0.0277, acc:99.1200%]\n",
      "Val loss: 0.3478, accuracy: 92.10%\n",
      "Epoch: 47/150, lr:6.25e-03\n",
      "100%|██████████| 625/625 [00:44<00:00, 13.90it/s, loss:0.0210, acc:99.3250%]\n",
      "Val loss: 0.3516, accuracy: 92.10%\n",
      "Epoch: 48/150, lr:6.25e-03\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.22it/s, loss:0.0196, acc:99.3750%]\n",
      "Val loss: 0.3485, accuracy: 92.33%\n",
      "Epoch: 49/150, lr:6.25e-03\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.86it/s, loss:0.0180, acc:99.4450%]\n",
      "Val loss: 0.3487, accuracy: 92.17%\n",
      "Epoch: 50/150, lr:6.25e-03\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.35it/s, loss:0.0157, acc:99.5150%]\n",
      "Val loss: 0.3516, accuracy: 92.16%\n",
      "Epoch: 51/150, lr:6.25e-03\n",
      "100%|██████████| 625/625 [00:44<00:00, 13.90it/s, loss:0.0146, acc:99.5650%]\n",
      "Val loss: 0.3589, accuracy: 92.27%\n",
      "Epoch: 52/150, lr:6.25e-03\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.02it/s, loss:0.0143, acc:99.5725%]\n",
      "Val loss: 0.3579, accuracy: 92.26%\n",
      "Epoch: 53/150, lr:6.25e-03\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.29it/s, loss:0.0135, acc:99.5875%]\n",
      "Val loss: 0.3733, accuracy: 91.97%\n",
      "Epoch: 54/150, lr:6.25e-03\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.87it/s, loss:0.0147, acc:99.5375%]\n",
      "Val loss: 0.3660, accuracy: 92.23%\n",
      "Epoch: 55/150, lr:6.25e-03\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.09it/s, loss:0.0123, acc:99.6350%]\n",
      "Val loss: 0.3646, accuracy: 92.41%\n",
      "Epoch: 56/150, lr:6.25e-03\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.08it/s, loss:0.0149, acc:99.5475%]\n",
      "Val loss: 0.3595, accuracy: 92.41%\n",
      "Epoch: 57/150, lr:3.13e-03\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.01it/s, loss:0.0103, acc:99.6875%]\n",
      "Val loss: 0.3712, accuracy: 92.35%\n",
      "Epoch: 58/150, lr:3.13e-03\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.79it/s, loss:0.0086, acc:99.7450%]\n",
      "Val loss: 0.3567, accuracy: 92.31%\n",
      "Epoch: 59/150, lr:3.13e-03\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "100%|██████████| 625/625 [00:44<00:00, 13.97it/s, loss:0.0090, acc:99.7525%]\n",
      "Val loss: 0.3631, accuracy: 92.40%\n",
      "Epoch: 60/150, lr:3.13e-03\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.51it/s, loss:0.0081, acc:99.7675%]\n",
      "Val loss: 0.3562, accuracy: 92.76%\n",
      "Epoch: 61/150, lr:3.13e-03\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.85it/s, loss:0.0080, acc:99.7725%]\n",
      "Val loss: 0.3623, accuracy: 92.52%\n",
      "Epoch: 62/150, lr:3.13e-03\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.00it/s, loss:0.0070, acc:99.8150%]\n",
      "Val loss: 0.3555, accuracy: 92.60%\n",
      "Epoch: 63/150, lr:3.13e-03\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.07it/s, loss:0.0072, acc:99.8050%]\n",
      "Val loss: 0.3573, accuracy: 92.58%\n",
      "Epoch: 64/150, lr:3.13e-03\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.84it/s, loss:0.0076, acc:99.7775%]\n",
      "Val loss: 0.3712, accuracy: 92.47%\n",
      "Epoch: 65/150, lr:3.13e-03\n",
      "100%|██████████| 625/625 [00:42<00:00, 14.79it/s, loss:0.0065, acc:99.8325%]\n",
      "Val loss: 0.3619, accuracy: 92.61%\n",
      "Epoch: 66/150, lr:3.13e-03\n",
      "100%|██████████| 625/625 [00:44<00:00, 13.93it/s, loss:0.0055, acc:99.8625%]\n",
      "Val loss: 0.3679, accuracy: 92.48%\n",
      "Epoch: 67/150, lr:3.13e-03\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.17it/s, loss:0.0069, acc:99.8050%]\n",
      "Val loss: 0.3756, accuracy: 92.43%\n",
      "Epoch: 68/150, lr:1.56e-03\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.21it/s, loss:0.0057, acc:99.8575%]\n",
      "Val loss: 0.3691, accuracy: 92.63%\n",
      "Epoch: 69/150, lr:1.56e-03\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.03it/s, loss:0.0053, acc:99.8500%]\n",
      "Val loss: 0.3615, accuracy: 92.76%\n",
      "Epoch: 70/150, lr:1.56e-03\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.87it/s, loss:0.0054, acc:99.8575%]\n",
      "Val loss: 0.3612, accuracy: 92.72%\n",
      "Epoch: 71/150, lr:1.56e-03\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.82it/s, loss:0.0052, acc:99.8525%]\n",
      "Val loss: 0.3648, accuracy: 92.57%\n",
      "Epoch: 72/150, lr:1.56e-03\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.41it/s, loss:0.0045, acc:99.9000%]\n",
      "Val loss: 0.3605, accuracy: 92.79%\n",
      "Epoch: 73/150, lr:1.56e-03\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.24it/s, loss:0.0050, acc:99.8825%]\n",
      "Val loss: 0.3631, accuracy: 92.69%\n",
      "Epoch: 74/150, lr:1.56e-03\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.17it/s, loss:0.0045, acc:99.8800%]\n",
      "Val loss: 0.3643, accuracy: 92.68%\n",
      "Epoch: 75/150, lr:1.56e-03\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.09it/s, loss:0.0041, acc:99.8950%]\n",
      "Val loss: 0.3658, accuracy: 92.79%\n",
      "Epoch: 76/150, lr:1.56e-03\n",
      "100%|██████████| 625/625 [00:44<00:00, 13.94it/s, loss:0.0042, acc:99.8950%]\n",
      "Val loss: 0.3634, accuracy: 92.65%\n",
      "Epoch: 77/150, lr:1.56e-03\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.08it/s, loss:0.0045, acc:99.9025%]\n",
      "Val loss: 0.3665, accuracy: 92.70%\n",
      "Epoch: 78/150, lr:1.56e-03\n",
      "100%|██████████| 625/625 [00:42<00:00, 14.57it/s, loss:0.0036, acc:99.9125%]\n",
      "Val loss: 0.3685, accuracy: 92.52%\n",
      "Epoch: 79/150, lr:7.81e-04\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.22it/s, loss:0.0043, acc:99.8975%]\n",
      "Val loss: 0.3661, accuracy: 92.75%\n",
      "Epoch: 80/150, lr:7.81e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.72it/s, loss:0.0037, acc:99.9200%]\n",
      "Val loss: 0.3656, accuracy: 92.79%\n",
      "Epoch: 81/150, lr:7.81e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.14it/s, loss:0.0042, acc:99.8950%]\n",
      "Val loss: 0.3681, accuracy: 92.66%\n",
      "Epoch: 82/150, lr:7.81e-04\n",
      "100%|██████████| 625/625 [00:40<00:00, 15.37it/s, loss:0.0038, acc:99.9025%]\n",
      "Val loss: 0.3701, accuracy: 92.59%\n",
      "Epoch: 83/150, lr:7.81e-04\n",
      "100%|██████████| 625/625 [00:41<00:00, 15.10it/s, loss:0.0035, acc:99.9350%]\n",
      "Val loss: 0.3681, accuracy: 92.73%\n",
      "Epoch: 84/150, lr:7.81e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.82it/s, loss:0.0036, acc:99.9125%]\n",
      "Val loss: 0.3665, accuracy: 92.59%\n",
      "Epoch: 85/150, lr:7.81e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.82it/s, loss:0.0041, acc:99.8850%]\n",
      "Val loss: 0.3670, accuracy: 92.79%\n",
      "Epoch: 86/150, lr:7.81e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.15it/s, loss:0.0039, acc:99.9100%]\n",
      "Val loss: 0.3657, accuracy: 92.79%\n",
      "Epoch: 87/150, lr:7.81e-04\n",
      "100%|██████████| 625/625 [00:46<00:00, 13.38it/s, loss:0.0032, acc:99.9350%]\n",
      "Val loss: 0.3627, accuracy: 92.63%\n",
      "Epoch: 88/150, lr:7.81e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.64it/s, loss:0.0035, acc:99.9175%]\n",
      "Val loss: 0.3641, accuracy: 92.87%\n",
      "Epoch: 89/150, lr:7.81e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.09it/s, loss:0.0034, acc:99.9200%]\n",
      "Val loss: 0.3657, accuracy: 92.75%\n",
      "Epoch: 90/150, lr:3.91e-04\n",
      "100%|██████████| 625/625 [00:47<00:00, 13.24it/s, loss:0.0037, acc:99.9325%]\n",
      "Val loss: 0.3624, accuracy: 92.84%\n",
      "Epoch: 91/150, lr:3.91e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.15it/s, loss:0.0037, acc:99.9200%]\n",
      "Val loss: 0.3646, accuracy: 92.70%\n",
      "Epoch: 92/150, lr:3.91e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.75it/s, loss:0.0032, acc:99.9275%]\n",
      "Val loss: 0.3668, accuracy: 92.78%\n",
      "Epoch: 93/150, lr:3.91e-04\n",
      "100%|██████████| 625/625 [00:46<00:00, 13.50it/s, loss:0.0034, acc:99.9250%]\n",
      "Val loss: 0.3651, accuracy: 92.73%\n",
      "Epoch: 94/150, lr:3.91e-04\n",
      "100%|██████████| 625/625 [00:42<00:00, 14.61it/s, loss:0.0033, acc:99.9425%]\n",
      "Val loss: 0.3671, accuracy: 92.77%\n",
      "Epoch: 95/150, lr:3.91e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.86it/s, loss:0.0032, acc:99.9300%]\n",
      "Val loss: 0.3663, accuracy: 92.79%\n",
      "Epoch: 96/150, lr:3.91e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 13.91it/s, loss:0.0033, acc:99.9350%]\n",
      "Val loss: 0.3709, accuracy: 92.70%\n",
      "Epoch: 97/150, lr:3.91e-04\n",
      "100%|██████████| 625/625 [00:46<00:00, 13.37it/s, loss:0.0030, acc:99.9325%]\n",
      "Val loss: 0.3650, accuracy: 92.75%\n",
      "Epoch: 98/150, lr:3.91e-04\n",
      "100%|██████████| 625/625 [00:42<00:00, 14.54it/s, loss:0.0032, acc:99.9350%]\n",
      "Val loss: 0.3649, accuracy: 92.69%\n",
      "Epoch: 99/150, lr:3.91e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.69it/s, loss:0.0029, acc:99.9550%]\n",
      "Val loss: 0.3636, accuracy: 92.79%\n",
      "Epoch: 100/150, lr:3.91e-04\n",
      "100%|██████████| 625/625 [00:46<00:00, 13.53it/s, loss:0.0027, acc:99.9475%]\n",
      "Val loss: 0.3667, accuracy: 92.63%\n",
      "Epoch: 101/150, lr:1.95e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.88it/s, loss:0.0032, acc:99.9300%]\n",
      "Val loss: 0.3644, accuracy: 92.73%\n",
      "Epoch: 102/150, lr:1.95e-04\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.26it/s, loss:0.0029, acc:99.9450%]\n",
      "Val loss: 0.3645, accuracy: 92.79%\n",
      "Epoch: 103/150, lr:1.95e-04\n",
      "100%|██████████| 625/625 [00:46<00:00, 13.37it/s, loss:0.0029, acc:99.9450%]\n",
      "Val loss: 0.3654, accuracy: 92.74%\n",
      "Epoch: 104/150, lr:1.95e-04\n",
      "100%|██████████| 625/625 [00:48<00:00, 13.00it/s, loss:0.0032, acc:99.9225%]\n",
      "Val loss: 0.3648, accuracy: 92.83%\n",
      "Epoch: 105/150, lr:1.95e-04\n",
      "100%|██████████| 625/625 [00:41<00:00, 15.00it/s, loss:0.0035, acc:99.9200%]\n",
      "Val loss: 0.3679, accuracy: 92.61%\n",
      "Epoch: 106/150, lr:1.95e-04\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.40it/s, loss:0.0033, acc:99.9175%]\n",
      "Val loss: 0.3666, accuracy: 92.71%\n",
      "Epoch: 107/150, lr:1.95e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.81it/s, loss:0.0029, acc:99.9400%]\n",
      "Val loss: 0.3628, accuracy: 92.84%\n",
      "Epoch: 108/150, lr:1.95e-04\n",
      "100%|██████████| 625/625 [00:41<00:00, 14.91it/s, loss:0.0034, acc:99.9250%]\n",
      "Val loss: 0.3710, accuracy: 92.66%\n",
      "Epoch: 109/150, lr:1.95e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.15it/s, loss:0.0036, acc:99.9100%]\n",
      "Val loss: 0.3660, accuracy: 92.72%\n",
      "Epoch: 110/150, lr:1.95e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.89it/s, loss:0.0032, acc:99.9150%]\n",
      "Val loss: 0.3652, accuracy: 92.75%\n",
      "Epoch: 111/150, lr:1.95e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.08it/s, loss:0.0033, acc:99.9125%]\n",
      "Val loss: 0.3679, accuracy: 92.72%\n",
      "Epoch: 112/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.60it/s, loss:0.0033, acc:99.9125%]\n",
      "Val loss: 0.3669, accuracy: 92.75%\n",
      "Epoch: 113/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.81it/s, loss:0.0031, acc:99.9200%]\n",
      "Val loss: 0.3646, accuracy: 92.72%\n",
      "Epoch: 114/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:47<00:00, 13.28it/s, loss:0.0033, acc:99.9175%]\n",
      "Val loss: 0.3685, accuracy: 92.57%\n",
      "Epoch: 115/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.74it/s, loss:0.0031, acc:99.9375%]\n",
      "Val loss: 0.3652, accuracy: 92.79%\n",
      "Epoch: 116/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:46<00:00, 13.52it/s, loss:0.0032, acc:99.9225%]\n",
      "Val loss: 0.3743, accuracy: 92.55%\n",
      "Epoch: 117/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.16it/s, loss:0.0027, acc:99.9550%]\n",
      "Val loss: 0.3696, accuracy: 92.76%\n",
      "Epoch: 118/150, lr:1.00e-04\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "100%|██████████| 625/625 [00:42<00:00, 14.78it/s, loss:0.0030, acc:99.9325%]\n",
      "Val loss: 0.3662, accuracy: 92.80%\n",
      "Epoch: 119/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.88it/s, loss:0.0031, acc:99.9425%]\n",
      "Val loss: 0.3631, accuracy: 92.69%\n",
      "Epoch: 120/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.47it/s, loss:0.0029, acc:99.9525%]\n",
      "Val loss: 0.3661, accuracy: 92.72%\n",
      "Epoch: 121/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:42<00:00, 14.80it/s, loss:0.0029, acc:99.9450%]\n",
      "Val loss: 0.3673, accuracy: 92.76%\n",
      "Epoch: 122/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.08it/s, loss:0.0027, acc:99.9425%]\n",
      "Val loss: 0.3664, accuracy: 92.71%\n",
      "Epoch: 123/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.08it/s, loss:0.0029, acc:99.9400%]\n",
      "Val loss: 0.3670, accuracy: 92.73%\n",
      "Epoch: 124/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:42<00:00, 14.74it/s, loss:0.0028, acc:99.9400%]\n",
      "Val loss: 0.3690, accuracy: 92.69%\n",
      "Epoch: 125/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.87it/s, loss:0.0027, acc:99.9550%]\n",
      "Val loss: 0.3670, accuracy: 92.67%\n",
      "Epoch: 126/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.27it/s, loss:0.0032, acc:99.9175%]\n",
      "Val loss: 0.3632, accuracy: 92.71%\n",
      "Epoch: 127/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:40<00:00, 15.25it/s, loss:0.0029, acc:99.9450%]\n",
      "Val loss: 0.3672, accuracy: 92.71%\n",
      "Epoch: 128/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.36it/s, loss:0.0032, acc:99.9300%]\n",
      "Val loss: 0.3671, accuracy: 92.70%\n",
      "Epoch: 129/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.12it/s, loss:0.0029, acc:99.9275%]\n",
      "Val loss: 0.3647, accuracy: 92.66%\n",
      "Epoch: 130/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.00it/s, loss:0.0029, acc:99.9450%]\n",
      "Val loss: 0.3664, accuracy: 92.70%\n",
      "Epoch: 131/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:42<00:00, 14.65it/s, loss:0.0031, acc:99.9200%]\n",
      "Val loss: 0.3653, accuracy: 92.73%\n",
      "Epoch: 132/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.76it/s, loss:0.0033, acc:99.9350%]\n",
      "Val loss: 0.3686, accuracy: 92.66%\n",
      "Epoch: 133/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 13.90it/s, loss:0.0032, acc:99.9225%]\n",
      "Val loss: 0.3639, accuracy: 92.73%\n",
      "Epoch: 134/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.85it/s, loss:0.0029, acc:99.9475%]\n",
      "Val loss: 0.3654, accuracy: 92.69%\n",
      "Epoch: 135/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.82it/s, loss:0.0029, acc:99.9450%]\n",
      "Val loss: 0.3659, accuracy: 92.68%\n",
      "Epoch: 136/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.74it/s, loss:0.0030, acc:99.9225%]\n",
      "Val loss: 0.3719, accuracy: 92.64%\n",
      "Epoch: 137/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.15it/s, loss:0.0032, acc:99.9250%]\n",
      "Val loss: 0.3695, accuracy: 92.71%\n",
      "Epoch: 138/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:48<00:00, 12.99it/s, loss:0.0029, acc:99.9425%]\n",
      "Val loss: 0.3640, accuracy: 92.65%\n",
      "Epoch: 139/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:46<00:00, 13.44it/s, loss:0.0028, acc:99.9350%]\n",
      "Val loss: 0.3676, accuracy: 92.62%\n",
      "Epoch: 140/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.05it/s, loss:0.0030, acc:99.9400%]\n",
      "Val loss: 0.3646, accuracy: 92.71%\n",
      "Epoch: 141/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.02it/s, loss:0.0027, acc:99.9425%]\n",
      "Val loss: 0.3672, accuracy: 92.78%\n",
      "Epoch: 142/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 13.94it/s, loss:0.0026, acc:99.9550%]\n",
      "Val loss: 0.3664, accuracy: 92.75%\n",
      "Epoch: 143/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 14.14it/s, loss:0.0034, acc:99.9125%]\n",
      "Val loss: 0.3634, accuracy: 92.77%\n",
      "Epoch: 144/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:44<00:00, 13.95it/s, loss:0.0030, acc:99.9325%]\n",
      "Val loss: 0.3680, accuracy: 92.74%\n",
      "Epoch: 145/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.25it/s, loss:0.0030, acc:99.9225%]\n",
      "Val loss: 0.3663, accuracy: 92.63%\n",
      "Epoch: 146/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:43<00:00, 14.31it/s, loss:0.0028, acc:99.9325%]\n",
      "Val loss: 0.3642, accuracy: 92.73%\n",
      "Epoch: 147/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:46<00:00, 13.49it/s, loss:0.0029, acc:99.9450%]\n",
      "Val loss: 0.3667, accuracy: 92.63%\n",
      "Epoch: 148/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:47<00:00, 13.18it/s, loss:0.0032, acc:99.9150%]\n",
      "Val loss: 0.3667, accuracy: 92.72%\n",
      "Epoch: 149/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:47<00:00, 13.18it/s, loss:0.0033, acc:99.9325%]\n",
      "Val loss: 0.3688, accuracy: 92.76%\n",
      "Epoch: 150/150, lr:1.00e-04\n",
      "100%|██████████| 625/625 [00:45<00:00, 13.85it/s, loss:0.0028, acc:99.9450%]\n",
      "Val loss: 0.3709, accuracy: 92.65%\n",
      "Test loss: 0.3945, accuracy: 92.45%\n",
      "CPU times: user 1h 32min 8s, sys: 28min 30s, total: 2h 38s\n",
      "Wall time: 1h 59min 39s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "import torch.nn.functional as F\n",
    "opt = optim.SGD(model.parameters(), lr=1e-1, momentum=0.9, weight_decay=1e-4, nesterov=True)\n",
    "loss_fn = F.cross_entropy\n",
    "\n",
    "lr_scheduler = optim.lr_scheduler.ReduceLROnPlateau(opt, 'min', factor=0.5, min_lr=1e-4)\n",
    "history = main(model, opt, epoch=150, loss_fn=loss_fn, lr_scheduler=lr_scheduler)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# plot loss history\n",
    "We check the classifier by plotting the loss and acc."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYUAAAEKCAYAAAD9xUlFAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAAIABJREFUeJzt3Xl8HNWV6PHf6VZrl7V7lY1sbBabxTayMYGwhM0QAmSBsGV7SZx1EmYmDPBCyDIv80KSRxJmIOBkHEIWEgIkQxKHNWYNmwzG2NjGxqu8SZa1WXt3n/fHLcltIVltWaVquc/389FH3VW3q05Xd9epurfqXlFVjDHGGIBQ0AEYY4xJHZYUjDHG9LKkYIwxppclBWOMMb0sKRhjjOllScEYY0wvSwrGGGN6WVIwxhjTy5KCMcaYXhlBB3CoysrKtLKyMugwjDFmVFm+fPkeVS0frNyoSwqVlZVUV1cHHYYxxowqIrIlmXJWfWSMMaaXJQVjjDG9LCkYY4zpNeraFPrT3d1NTU0NHR0dQYfiu+zsbCoqKohEIkGHYow5Ah0RSaGmpoaCggIqKysRkaDD8Y2qUl9fT01NDVOnTg06HGPMEci36iMRWSIitSKy6iBlzhaRFSKyWkSeGeq6Ojo6KC0tPaITAoCIUFpamhZnRMaYYPjZpnAvsHCgmSJSBNwFXKqqs4ArDmdlR3pC6JEu79MYEwzfkoKqPgvsPUiRa4CHVXWrV77Wr1gAOrpj7GrqIBqL+7kaY4wZ1YK8+ugYoFhEnhaR5SLycT9X1tkdo7alg+748I9J3djYyF133XXIr7v44otpbGwc9niMMWaogkwKGcApwPuBC4FviMgx/RUUkUUiUi0i1XV1dUNaWU+1i+rIJYVoNHrQ1y1dupSioqJhj8cYY4YqyKRQAzymqq2qugd4Fji5v4KqulhVq1S1qrx80K47+hXyquJ9OFHgpptu4p133mH27NnMmzeP9773vVx66aXMnDkTgMsvv5xTTjmFWbNmsXjx4t7XVVZWsmfPHjZv3szxxx/PZz/7WWbNmsUFF1xAe3v78AdqjDGDCPKS1P8B/ktEMoBM4FTgR4e70G//eTVv7Wh+1/S4Ku1dMbIjYcKhQ2usnTlxDN/8wKwB53/ve99j1apVrFixgqeffpr3v//9rFq1qvey0SVLllBSUkJ7ezvz5s3jwx/+MKWlpQcsY/369dx///387Gc/48orr+Shhx7iuuuuO6Q4jTHmcPmWFETkfuBsoExEaoBvAhEAVb1bVdeIyKPASiAO/FxVB7x8dTSZP3/+AfcR3HHHHfzxj38EYNu2baxfv/5dSWHq1KnMnj0bgFNOOYXNmzePWLzGGNPDt6SgqlcnUeYHwA+Gc70DHdF3dMd4e3cLU0pyKcrNHM5VvkteXl7v46effponn3ySF198kdzcXM4+++x+7zPIysrqfRwOh636yBgTiLTp+8jPNoWCggJaWlr6ndfU1ERxcTG5ubmsXbuWl156afgDMMaYYXJEdHORDD+vPiotLeX000/nhBNOICcnh3HjxvXOW7hwIXfffTfHH388xx57LAsWLBj29RtjzHARP3aSfqqqqtK+g+ysWbOG448//qCvi8XjrN7RzITCHMoLsg5aNtUl836NMSaRiCxX1arByqVN9ZGfZwrGGHOkSJukEBJBEF/aFIwx5kiRNkkBQMTdr2CMMaZ/aZUUQiJWfWSMMQeRVknBnSkEHYUxxqSu9EkK7Y0coxsJxzuDjsQYY1JW+iQFIIyCDv94CkPtOhvgxz/+MW1tbcMckTHGDE36JAXx3qolBWOMGVDa3NG8PykMf6NCYtfZ559/PmPHjuWBBx6gs7OTD37wg3z729+mtbWVK6+8kpqaGmKxGN/4xjfYvXs3O3bs4JxzzqGsrIxly5YNe2zGGHMojryk8LebYNeb756uMehuo5QsyDzEDvHGnwgXfW/A2YldZz/++OM8+OCDvPLKK6gql156Kc8++yx1dXVMnDiRv/71r4DrE6mwsJDbb7+dZcuWUVZWdmgxGWOMD9Kn+oieMRT8vfzo8ccf5/HHH2fOnDnMnTuXtWvXsn79ek488USeeOIJbrzxRp577jkKCwt9jcMYY4biyDtTGOiIPtoFtavZK+VMmFDh2+pVlZtvvpnPfe5z75r32muvsXTpUm655RbOPfdcbr31Vt/iMMaYoUifMwWvTUF8aFNI7Dr7wgsvZMmSJezbtw+A7du3U1tby44dO8jNzeW6667jhhtu4LXXXnvXa40xJmh+jry2BLgEqFXVEw5Sbh7wInCVqj7oVzx4HeIJw3/1UWLX2RdddBHXXHMNp512GgD5+fn8+te/ZsOGDdxwww2EQiEikQg//elPAVi0aBELFy5k4sSJ1tBsjAmcb11ni8iZwD7gvoGSgoiEgSeADmBJMklhqF1nowo7V1CrxYydVJnUe0hV1nW2MeZQBd51tqo+C+wdpNg/AQ8BtX7F0UsEdf2kWv9HxhgzgMDaFERkEvBB4KdJlF0kItUiUl1XVzfkdSohBPX5+iNjjBm9gmxo/jFwo+rgtxir6mJVrVLVqvLy8oHKDLpCFSFEfFR3n21nOcYYPwV5SWoV8DtvRLQy4GIRiarqnw51QdnZ2dTX11NaWto7wlq/JEQI9eOm5hGhqtTX15OdnR10KMaYI1RgSUFVp/Y8FpF7gb8MJSEAVFRUUFNTw2BVS/Hm3XTGQ0Qau8kIjc6rcbOzs6mo8O8+C2NMevPzktT7gbOBMhGpAb4JRABU9e7hXFckEmHq1KmDltt7x+dZXRdl/JcfZca4guEMwRhjjgi+JQVVvfoQyn7SrzgOWE9GDjlST0f38N+rYIwxR4LRWYcyVJEcsumiIxoLOhJjjElJaZUUJJJDDp102pmCMcb0K62SApFcsqWLjm47UzDGmP6kVVIIZeaSQ6dVHxljzACOvK6zDyKUmUsWXdbQbIwxA0izM4UccqSLzu5o0KEYY0xKSqukkJGVB0B3Z3vAkRhjTGpKq6QQzs4FINrZGnAkxhiTmtIqKfScKcQ72wKOxBhjUlNaJQWJ5ACgXVZ9ZIwx/UmrpICXFOJdVn1kjDH9ScukoN12pmCMMf1Js6TgGprptjYFY4zpT5olBXemgJ0pGGNMv9IrKWR4SSHaEWwcxhiTotIrKXhnChK1MwVjjOmPb0lBRJaISK2IrBpg/rUislJE3hSRf4jIyX7F0strUwhbUjDGmH75eaZwL7DwIPM3AWep6onAvwOLfYzF8c4UQjGrPjLGmP74ORznsyJSeZD5/0h4+hLg/2j0XlIIW1Iwxph+pUqbwqeBvw00U0QWiUi1iFTX1dUNfS2hMFGJkGFJwRhj+hV4UhCRc3BJ4caByqjqYlWtUtWq8vLyw1pfdyiLSNySgjHG9CfQQXZE5CTg58BFqlo/EuvsDmWTYZekGmNMvwI7UxCRKcDDwMdU9e2RWm8snE1mvHOkVmeMMaOKb2cKInI/cDZQJiI1wDeBCICq3g3cCpQCd4kIQFRVq/yKp0cslE0WXXTH4kTCgdeeGWNMSvHz6qOrB5n/GeAzfq1/ILGMbHLopDNqScEYY/pKu71iPCOHbOmiozsWdCjGGJNy0i4paEYOOXRaUjDGmH6kYVLIJptuOqPxoEMxxpiUk3ZJgUguOWJnCsYY05+0SwoSySWbLtq6LCkYY0xfaZcUMrJzyaGTprbuoEMxxpiUk3ZJITM7jxy6aGrrCjoUY4xJOWmXFLJy8giJ0tJm4zQbY0xfaZcUMnPyAWhrbQk4EmOMST1plxTCmW70tfbWfQFHYowxqSftkkLPkJyd7ZYUjDGmrzRMCtkAdFlSMMaYd0m/pJBVAEC8oyngQIwxJvWkX1LIHwdAZvuegAMxxpjUk35JoWACAHldhzHWszHGHKF8SwoiskREakVk1QDzRUTuEJENIrJSROb6FcsBcoqJSoSCbjtTMMaYvvw8U7gXWHiQ+RcBM7y/RcBPfYxlPxHaMssp0QY6o9b/kTHGJPItKajqs8DegxS5DLhPnZeAIhGZ4Fc8iTpyyhlHA03t1v+RMcYkCrJNYRKwLeF5jTfNd9HccYyTBpotKRhjzAEGTQoicoyIPNXTNiAiJ4nILf6HdkAMi0SkWkSq6+oOv4E4nj+OsdJoZwrGGNNHMmcKPwNuBroBVHUlcNUwrHs7MDnheYU37V1UdbGqVqlqVXl5+WGvOFQwnjHSRkuL3atgjDGJkkkKuar6Sp9p0WFY9yPAx72rkBYATaq6cxiWO6iMookAdDXsGInVGWPMqJGRRJk9InI0oAAi8hFg0J23iNwPnA2UiUgN8E0gAqCqdwNLgYuBDUAb8KkhxD8k2cWu6SLatGukVmmMMaNCMknhS8Bi4DgR2Q5sAq4b7EWqevUg89Vb9ojLKa0AQPZZUjDGmESDJgVV3QicJyJ5QEhVR/1ABJFCd+Vr2JKCMcYcYNCkICK39nkOgKp+x6eY/JdTTCcRIu3W1YUxxiRKpvqoNeFxNnAJsMafcEaICHtDJeR01gYdiTHGpJRkqo/+X+JzEfkh8JhvEY2Q5nAp+V3W/5ExxiQayh3Nubh7Cka1lswyxkTrgw7DGGNSSjJtCm/iXY4KhIFyYPS2J3jas8opaa0OOgxjjEkpybQpXJLwOArsVtXhuHktUF0548inDbpaITMv6HCMMSYlDFh9JCIlIlICtCT8tQNjvOmjWixvLADdTSNyE7UxxowKBztTWI6rNpJ+5ikwzZeIRoo3AlvrnhqKyqcHHIwxxqSGAZOCqk4dyUBGWkaBO1PoaNodcCTGGJM6kmlTQESKcSOkZfdM8wbRGbWyCl1vq13Ndq+CMcb0SObqo88AX8VdhroCWAC8CLzP39D8VVQ6HoDORksKxhjTI5n7FL4KzAO2qOo5wByg0deoRkBFWRHNmkunnSkYY0yvZJJCh6p2AIhIlqquBY71Nyz/FeZGaJAxxPfZXc3GGNMjmTaFGhEpAv4EPCEiDcAWf8MaGW0ZRYQ77K5mY4zpkUzfRx/0Hn5LRJYBhcCjvkY1QrqySshvt9HXjDGmx6DVRyJyh4i8B0BVn1HVR1S1K5mFi8hCEVknIhtE5KZ+5k8RkWUi8rqIrBSRiw/9LQyd5paSH2siHtfBCxtjTBpIpk1hOXCLiLwjIj8UkapkFiwiYeBO4CJgJnC1iMzsU+wW4AFVnQNcBdyVfOiHL1IwlmKa2d3cPpKrNcaYlDVoUlDVX6rqxbgrkNYBt4nI+iSWPR/YoKobvTOL3wGX9V08MMZ7XAiMaF1OduFYMiXG9l12A5sxxsChdZ09HTgOOApYm0T5ScC2hOc13rRE3wKuE5EaYCnwT4cQz2Eb492rULdr+0iu1hhjUlYybQrf984MvgO8CVSp6geGaf1XA/eqagVwMfArEXlXTCKySESqRaS6rm74htAsKnP9HzXusU7xjDEGkrsk9R3gNFU91Av6twOTE55XeNMSfRpYCKCqL4pINlAGHHBHmaouBhYDVFVVDVurcKTAdXWxr8Gqj4wxBpJrU7hnCAkB4FVghohMFZFMXEPyI33KbAXOBRCR43F9Kw3fqcBg8soA6G4ZuVUaY0wqG8pwnEnxBuL5Mm485zW4q4xWi8h3RORSr9i/Ap8VkTeA+4FPqurIXR+a65JCfJ8lBWOMgSR7SR0qVV2Ka0BOnHZrwuO3gNP9jOGgMnPpDmWT1dVAe1eMnMxwYKEYY0wqSKah+WgRyfIeny0iX/G6vTgidGeVUCLNbGtoCzoUY4wJXDLVRw8BMRGZjmvsnQz81teoRlJeGaW0sGlPa9CRGGNM4JJJCnGvfeCDwH+q6g3ABH/DGjmZY8opkWbW7mwJOhRjjAlcMkmhW0SuBj4B/MWbFvEvpJGVkV/OuPA+Vu9oCjoUY4wJXDJJ4VPAacB3VXWTiEwFfuVvWCMor4ximnlrZ3PQkRhjTOCS6Tr7LeAr0DtWc4Gq3uZ3YCMmt5RM7aS+oYGm9m4Kc46YkyBjjDlkyVx99LSIjBGREuA14Gcicrv/oY0Q7wa2UmlhjZ0tGGPSXDLVR4Wq2gx8CLhPVU8FzvM3rBHk3cBWQjNv7bCkYIxJb8kkhQwRmQBcyf6G5iNH/lgAZuTss3YFY0zaSyYpfAfXVcU7qvqqiEwDkhlPYXQYOxNCEc7J32xnCsaYtJdMQ/MfgD8kPN8IfNjPoEZUZi5UVDGnfhXr61voisbJzPCtSyhjjElpyTQ0V4jIH0Wk1vt7SEQqRiK4EVN5BhPa1pIZa2ND7b6gozHGmMAkc0j8C1yX1xO9vz97044clWcQ0hhVobd5bWtD0NEYY0xgkkkK5ar6C1WNen/3AuU+xzWyKuajoQjn5bzN0+usG21jTPpKJinUi8h1IhL2/q4D6v0ObERl5iIVVZyV9TYvbNhDZzQGqrDyAWg9st6qMcYcTDJJ4X/hLkfdBewEPgJ80seYglF5BhXta5HuVl7ZtBdqquHhz8LLdwcdmTHGjJhkhuPcoqqXqmq5qo5V1ctJ8uojEVkoIutEZIOI3DRAmStF5C0RWS0iwXXJ7bUrnBV5i2Vr62DFr930Lf8ILCRjjBlpQ7328l8GKyAiYeBO4CJgJnC1iMzsU2YGcDNwuqrOAq4fYjyHb8p7oHAKN+T8mRfXboNVD4OEoOZV6O4ILCxjjBlJQ00KkkSZ+cAGVd2oql3A74DL+pT5LHCnqjYAqGrtEOM5fBmZcPZNTOtax5eab4fOZjj18xDrhO3LAwvLGGNG0lCTgiZRZhKwLeF5jTct0THAMSLygoi8JCIL+1uQiCwSkWoRqa6r8/HqoJOvort4BpeEX6I5awKceQMgsOUF/9ZpjDEpZMCkICItItLcz18L7n6F4ZABzADOBq7G9cD6rvGfVXWxqlapalV5uY9Xw4bCRM7/BgAPRc8gll0M406Azc/7t05jjEkhAyYFVS1Q1TH9/BWo6qDdYwDbceM596jwpiWqAR5R1W5V3QS8jUsSwTn+Ul5f8GN+2LqQv6+thaPeA9tegWhXoGEZY8xI8LOTn1eBGSIyVUQygatwd0Yn+hPuLAERKcNVJ230MabBiXDi+Z+gYEwx9724GSpPh2g77FwRaFjGGDMSfEsKqhoFvozrYXUN8ICqrhaR74jIpV6xx3A3x70FLANuUNXA7xbLCIe45tQpPLd+D5vzZwMCa/4cdFjGGOM7UU2mzTh1VFVVaXV1te/rqW3p4IzvLeOj8ybz79Efwbql8JUVUDDO93UbY8xwE5Hlqlo1WDnrI3oAYwuy+dDcSTxQvY29p34Nop3w3P8LOixjjPGVJYWDWHTmNLpicZa8FYK5H4PqJdCwOeiwjDHGN5YUDmJaeT4LZ43nvhc307rgXyCUAfecBc//yO5yNsYckSwpDOLzZx1Nc0eUe1d1w2eehMmnwpPfgqf/b9ChGWPMsLOkMIiTJxdx4axx/NffN7Aj+2i49gGYdjasfyLo0IwxZthZUkjCLe+fSVyV7y5d4yZMPQtqV8M+G5DHGHNksaSQhMkluXzpnOn8deVOnltf55ICwOZngw3MGGOGmSWFJC06cxrTyvP459+/we78YyFrDGyypGCMObJYUkhSdiTMPdedQltXlC/89g3iU95jScEYc8SxpHAIZowr4IdXnMxrWxv5a8sM2LsRGrcN/kJjjBklLCkcootPnMA/vW86d27xhobY/FywARljzDCypDAE/3L+MRx30nx2axEtT/0Q2vYGHZIxxgwLSwpDICLcdsVs7iq5mazmLbTfdwV0tQUdljHGHDZLCkOUlRHmS5/6JLeErydr13K6Hrs16JCMMeawWVI4DGPHZPOh677Io/H5dK78I4yybsiNMaYvX5OCiCwUkXUiskFEbjpIuQ+LiIrIoH19p5oF00rZVHw6Bd17YNebQYdjjDGHxbekICJh4E7gImAmcLWIzOynXAHwVeBlv2LxW9Mkd4ezrn884EiMMebw+HmmMB/YoKobVbUL+B1wWT/l/h24DRi1fVGPm3gUK+NTia57LOhQjDHmsPiZFCYBiXd21XjTeonIXGCyqv7Vxzh8d3R5Hsvis8nYUW2XpxpjRrXAGppFJATcDvxrEmUXiUi1iFTX1aVez6TTx+bzdGw2onF45+9DX9DKB+Dn58OeDcMXnDHGHAI/k8J2YHLC8wpvWo8C4ATgaRHZDCwAHumvsVlVF6tqlapWlZeX+xjy0EwszGFdxgxaM4pgzSP7Z6jCrlXwj/+E138N8djAC9m7Cf58PdS8AksugO3L/Q/cGGP6yPBx2a8CM0RkKi4ZXAVc0zNTVZuAsp7nIvI08DVVrfYxJl+EQkJl2Riejl3I+996AGrXQtkx8Lur4e1H9xdcfi9UfRrqN0BuCcz/HIQzIB6HP30RQmH4xJ/hf74Mv7wMvroC8soGXO8RqasNXvwvmHQKHP0+EHl3mV1vwphJbhseipbd0L4Xxh6/f1r9O259detg1gfhxI9ATvHhvYdkxbqhaRuEIlAwHsIR/9fZ0eTe6/iTIJLtzzoaNkN3O5Qf1//ndzDxOKx+2J1xT54PR58LhRWHvpxYtxs+91BflwxV2PG6++4UV0L5MZBdOPzr6RGLQnMNNG6F/HFQfqx/68LHpKCqURH5MvAYEAaWqOpqEfkOUK2qjxx8CaPL0WPzuWvrxbw/cyk8/R9w1BkuIZx5A5zyKdjyAvztRvjT50FCoHFY8xc4/Suw/Jew9R9w+U9h6plwze/hrgXw+q/gjH8O+q0dXO1aePtvLtG11kNeKYydBad+ziW5Q9HRBL+9ym0LcDuViir3Q6iYD1MWwDO3wUt3QdEUuPYhKJvhkkR2IRQf5V7XstvtEPJK9y9768vw+2uhtQ5OvMLtbFY/DBuedGVLpsHSr8ETt8Kcj8Epn3TJobsNtr8GjVtg0lyYOAead8Led2BfLbTVQ+seF/uYCS6uhi0upsYt0LwDckqgdBpkF0E408XQsAmaatz3AABx73PMRIjkuJ1qONMdFBRWwLhZbjmtde5vX6172fgToWACNG11O+OGzdDWAPljIXsMNG2H1lqI5EKsC3asAI1BViEcdzHklrrvY06R6w6+/h3Ys86tq7gSMrJdjBrz/schHnVntrtXuee5ZVA6HSbPc9t51YNuetEUl9yzCqCr1SWj9kb3PK/MLV9CsO0VaNkJ406AzmbYtRIy82HFb9x7zCuHCSe7v5Jpbt171rnvW0cTZGRBZp5LBN1tbpu37YFIHhRNdgmwYh40b3fr6mhy680rg9Kj3U6+ZReg7nvUWgc734BwFkx/H5Qc7aa1N0K0w82rX7//uxXKcL/biXPcvKYad1BYWOE+j6Ya95lmF7kEMmYSbH0Rtr3qvmOFFe6vYLyLfe9G91405rZL03b3GOA9/wQX/J9D+10dItFRdsNVVVWVVlen3snET55cz4+fepv1560g47nvuy/U1PfCtQ/uP1ppb3RfkNLprprpz191X+KcYjj1C3DWv+0ve+8lbqfylRWHvnMdbvE4LF8C21+HcTPdl7ej2e1Q1/7Flckf53YObfWwbxfMvg4u/U8IHaSGsqsN3rgfVj3kfqTN212vs5ff5araXvul+1G11rkdEQIozL4W1j/ufjgF46FurZs3/TxXbuPTbpsdf6lLKg1bYPkv3A/v2Ivh1Z+7H/eYSXDyVe6MrWCc22G+sti17cS7k98+WYVuB9yy060/FIGxx7mdyZiJLmns3QidLRDrdNupZKrbKRYd5X7wzTvd+2/e7t5XRhZEO932bNgC3a0HrjO7yG2jrpb908JZbpm5JbBvt/uMCie5z6a73e38pixwCWb94/D2Y24dGnPbA1zyKJvhfVe3JSQt3GckIbetCytg/AkuabTucQmitc69fv5noXiqW379eujc595P+XFuR9zZ7JLa3k1ue1TMc5/F7tUueZxxvUvce9523dPvXOl2tnVr3PaVsNuZ549zO/Foh3tdKMOtv2C82+4dTe77s3252x6hiNtx549176tlp/tcJOxeI2HoaHTJccLJbttufMbFG8l1v9OMbLdNT/iIi7txi9vBr/6TO5Ife7z7TPesc59pcaVLTNFOdxHKnnUuORdMhMrTXdxN29x+ob3BraPkaJdExDtQKDrKHfAUTXHbsGB88t/NBCKyXFUHvRfMksIw+cvKHXz5t6/z6OdP5rjfnQ4ofPEl9+UcSP07UPuW25lFcg6ct/qP8IdPwjV/gGMuSC6Ixm3uxzntnP1VAztWuC9az1G0qvuxPvdDOOHDsOALB19m6x740xfcTiS7yP1oemQXuh3q/EWQn9DWs+w/3BH9MQu9o8qN7sip7Bi3s+5qczvyHSugswnGznTLinXD2TfDjPMOjCHaCZufh03PuCOy6ee5H/vvr4OMHJh9jTvSe+0+VwVz8lXux/b6r7wjyWw3rvblP3U7zOYd7uhr0in9J62m7S6xxLrc8saf5H6QNdXuKLZoirdTGu+OtDMy3etiUWjZ4ab3TBsO8bg7s+hshryx7sg5I3P/9NY6t+PIH3fwJHww3e0uieSV7T8Iicfc90VCbgd1sKoYVfeZ5BT5V/3W3eF2nkWTXZJJlqp7XW4pZOYe2jpj3e77l5U/+DqinYNXycW6XUIcM/Hd2zPaeWjv6xBZUhhha3Y2c9FPnuM/r57DB8q9o5LxJwx9gbFu+NEsmDAbrn3g4GX3boLffMRV4YCrrvrAj92O7T7v1pCiKe4ItbPFHb2FvS/fF190O7j+7HwD7r/G7XQu/C7M+4x73FbvduK5Zf3v/FTh7/8OL9zhjjpLj3ZXVNVvcD+EcBaUTXfVBbOvdUevftT9dntHkLkl/izfmFEk2aTgZ0NzWplalocIvFO3D06ec/gLDEdg7ifg2R+4M4Ciyf2Xi8ddw/S+Wlh4mzvzWP4LqDwDHv+GOzqf9xnY8g+3g8wf5+r7j1kId53m6tGve/jAnWZrvatvf+JWd9T36cfcaTe4U+/8sQePXQTOvRXO+XqwVV+RbP8aU405QllSGCbZkTBTSnJZvqVh+BY6+xp49vuuzv2M610978v37G94LD/OVZlsed7V38/9uDs63vZxwDbDAAARcElEQVQKPPRpV8d61W9cA+mpn3v38t93Czx6I7z5IJx0hasueOQr8MZvXZ3r5AVw5X2uvn0ogm4LMcYcMksKw+ij8ybz/UfXsWxtLeccN8jRdDJKpsKkKnc1xxnXw/O3wws/2T+/p9Fv+nnuihlwR8YfugeWXARn3eASwkDmfcYt+3++5K7UWf8ErPg1zPsszP2Yq0u3ahdj0oq1KQyjrmici+94js5ojMevP4uczGE4Un7pbnc0/6lHXbvBjAtctczuVe5qjaYaOPcb727Q7u5IruqkbS/88tL9V3ac+gW46HuHH7cxJqVYQ3NAXtpYz1WLX+JTp1dy6yUzkcM90m7ZDbcf5xp1W2vhCy+6y0KHU+sel3BKpsGHfmbVPsYcgayhOSALppXy8dOO4hcvbCYzI8RNC487vMRQMA6mngUbl8Fxlwx/QgB3GeJnl1lVkTHGkoIfvvWBWcTiyj3PbGRrfRufPmMqpxxVPPTkMOc6dxPPmV8b3kATWUIwxmBJwRehkPB/Lj+B8oIsfvbsRv62ahcTCrOZM6WIc44dy4fmVhAOHcJO+IQPu5u2BrsU1BhjDpO1KfistTPKX9/cybNv1/H61ka2N7Zz/IQxXH/eDOZOKaa8wL87GI0xpoc1NKcgVWXpm7v4j6Vr2N7YDkBRboTi3EwqS3P50jnTqaosobalgz0tXcycOCbgiI0xRwpLCimsozvG61sbWb2jic31rTS1R3lpYz11LZ1MLslh216XMM4+tpxvXDKTo8sH6XfFGGMGYUlhlGnrinLvPzazfHMDVZUliMCdf99ATJXnb3wfJXnD2MGaMSbt2CWpo0xuZgZfPHv6AdPmTC7io4tfYvmWBs6fOcSuJowx5hAENkazGdzJk4vICAmvbx3G/pSMMeYgfE0KIrJQRNaJyAYRuamf+f8iIm+JyEoReUpEjvIzntEmOxLm+AljeH1r4+CFjTFmGPiWFEQkDNwJXATMBK4Wkb63474OVKnqScCDwPf9ime0mjOliJU1jcTio6vtxxgzOvl5pjAf2KCqG1W1C/gdcFliAVVdpqpt3tOXgAof4xmVZk8uorUrxvralsELG2PMYfIzKUwCtiU8r/GmDeTTwN98jGdUmjPFDW24wqqQjDEjICUamkXkOqAK+MEA8xeJSLWIVNfV1Y1scAGrLM2lKDdi7QrGmBHhZ1LYDiSOIVnhTTuAiJwHfB24VFU7+1uQqi5W1SpVrSovL++vyBFLRJg9uYgV2ywpGGP852dSeBWYISJTRSQTuAp4JLGAiMwB7sElhFofYxnVZk8u4u3aFprau4MOxRhzhPMtKahqFPgy8BiwBnhAVVeLyHdE5FKv2A+AfOAPIrJCRB4ZYHFp7dzjxqEKdzy1PuhQjDFHOF/vaFbVpcDSPtNuTXh8np/rP1KcWFHItadO4RcvbOKy2RM5qaIo6JCMMUeolGhoNoO78aLjKMvP4t8eXMkLG/bQ2NYVdEjGmCOQ9X00SozJjvDdD57IF369nGt//jIAsyaO4axjyrlq3hSmlOYGHKEx5khgvaSOMg2tXaze0cyKbQ08u34Pr21pIK7KRSdM4Kxjyqksy+PVzXv5xzt7OGFiIVdUVTB9bEHQYRtjAmZdZ6eJ3c0dLHlhE/e/vJXmjmjv9Blj89m4p5VYXJlcksPcKcWcclQxc6cUU5gTIRZXKopzyAhbDaIx6cCSQpqJx5Ute9t4p3YfsyaNYUJhDnUtnfxl5Q5e3byX6s0N1LYceBvIJ047im9fdkJAERtjRpIlBXMAVWV7YzsrtjXS1hXjLyt38tqWBl75+rnkZlrTkjFHOhtkxxxARKgozqWi2DVIV5bmceU9L/LXlTu5omryIK82xqQLq1BOU/Mqi5lWnsfvX902eGFjTNqwpJCmRISr5k2meksDG6xbbmOMx5JCGvvQ3AoyQsJdT7/DaGtbMsb4w5JCGivLz2LRmdN4+LXt/Pfzm4IOxxiTAqyhOc197YJj2VzfyneXrqE7plw+ZyLFuZls2tPKmJwIk4pygg7RGDOCLCmkuVBIuP3K2TS1v8ptj67ltkfXIgI9tUlzphRx5oxyppblUZqfiSDkZIaZVJRDeUEW4ZAE+waMMcPKkoIhOxLmN59ZwIbafTy5ZjftXTGOHpvP9oZ2/vzGDn4yQJfdkbAwvjCbvMwMdjV3EI0p8yqLed/x47hm/hRLGMaMQnbzmhlUR3eMmoY2GtrcID/7OqJsb2x3fw3ttHZGGV+YTVzh5Y31bNzTymnTSvnJ1bMZW5AdcPTGGLCb18wwyo6ED6lTvQeX13DLn97kwh89y9Xzp/DhUyqYUpJLxPpZMibl+XqmICILgZ8AYeDnqvq9PvOzgPuAU4B64KOquvlgy7QzhdHh7d0tfP/Rtfx9bS1x7ytWkpdJWX4mYwuymVySy+SSHAqyI+RGwuRlhcnJzCA3M0xOJMy+zijN7d2U5mcxpSSXxrYuNte3Mbkkh2PHFSBiVVPGHIrAzxREJAzcCZwP1ACvisgjqvpWQrFPAw2qOl1ErgJuAz7qV0xm5BwzroCff2IeOxrbWbaultrmTvbs66SupZPdLZ08tnoXe1uHNlDQ2IIsjpswhrEFWWRmhOiOxskIC9kRl1ByImFyMsNkR8JkhIRoXInFlWhcyQgJY3IyGJMdYUxOhJDAnn1dtHVFyc3MIC8zw0tQYUIiCCACIXEN7NkZYbpjcbpiccIhIRwSMkIh7797Ho0rsZiSnRkiKyOMqrryItYrrUl5flYfzQc2qOpGABH5HXAZkJgULgO+5T1+EPgvEREdbQ0dZkATi3K49tSj+p3X2hmltStKe1eMtq4YbV1R2rpitHfFyMtyO+7alg627m2jODeTKaW5bNi9j+c37GHL3jbW726hO6ZEwm5H3NEVo607RiyeOl+fzHCIaDzee7YUDgkh7+quxChDAoK4K79wHRiqQtz7KbjXSe9/ARC8pOVe1/sYl8ggcbpbPr2P6T3bEtk/v2eei8GtX9WLJ3EaPVeo6bumDfTzTTy761lPqCf2A+IenCDveh9930sPTdjSiaH1F2ZP7NqnjHrvs7954H1+3sFDyIsnFNr/vCfEts4Y7d0xRCAsQsg7mOj5bAd7/9fMn8Lnzjr64IUOk59JYRKQ2LFODXDqQGVUNSoiTUApsCexkIgsAhYBTJkyxa94zQjLy8ogL2uwr2DhAc/mTinmynkH78CvOxanvTtGR1fMnR2E9x/NR2NxWjqiNHd009wepTsepywvi7yssJeYYrR2Rmnvjh2ws4vH1S2zO0ZmRohIKERc9YCzkFg8TjSuhL0feGfUrSsj5M4yYnGlozuGcuCOWnt3rG6n1LOTD3k7GVWIqRL31hXXPjupxB2yN73vTrvvzi2xPL3P9y+nZwfXk3z6njUlJp2QDLyTHkhvssF7Pz2JRwdPDH3fZ+K0xPdyQLJIXIAkPjwwoSTO3v/8wOSZWGr/5dtKPL7//cS94OKqxBKSam4kTG5mGAXvs9z//YkncTAzYQTuGxoVDc2quhhYDK5NIeBwTIqLhENEwiHGZEf6nV+anzXCERkzevhZwbkdSDykq/Cm9VtGRDJwh4X1PsZkjDHmIPxMCq8CM0RkqohkAlcBj/Qp8wjwCe/xR4C/W3uCMcYEx7fqI6+N4MvAY7hLUpeo6moR+Q5QraqPAP8N/EpENgB7cYnDGGNMQHxtU1DVpcDSPtNuTXjcAVzhZwzGGGOSZxdNG2OM6WVJwRhjTC9LCsYYY3pZUjDGGNNr1HWdLSJ1wJYhvryMPndLpyCLcXhYjMPDYjx8qRLfUapaPlihUZcUDoeIVCfTS2CQLMbhYTEOD4vx8KV6fH1Z9ZExxphelhSMMcb0SreksDjoAJJgMQ4Pi3F4WIyHL9XjO0BatSkYY4w5uHQ7UzDGGHMQaZMURGShiKwTkQ0iclPQ8QCIyGQRWSYib4nIahH5qje9RESeEJH13v/igOMMi8jrIvIX7/lUEXnZ25a/93rBDTK+IhF5UETWisgaETktBbfhP3uf8SoRuV9EsoPejiKyRERqRWRVwrR+t5s4d3ixrhSRuQHG+APvs14pIn8UkaKEeTd7Ma4TkQuDijFh3r+KiIpImfc8kO14KNIiKSSMF30RMBO4WkRmBhsVAFHgX1V1JrAA+JIX103AU6o6A3jKex6krwJrEp7fBvxIVacDDbixtoP0E+BRVT0OOBkXa8psQxGZBHwFqFLVE3C9BveMSR7kdrwXWNhn2kDb7SJghve3CPhpgDE+AZygqicBbwM3A3i/nauAWd5r7vJ++0HEiIhMBi4AtiZMDmo7Ji0tkgIJ40WrahfQM150oFR1p6q+5j1uwe3MJuFi+6VX7JfA5cFECCJSAbwf+Ln3XID34cbUhuDjKwTOxHXDjqp2qWojKbQNPRlAjjeYVC6wk4C3o6o+i+uyPtFA2+0y4D51XgKKRGRCEDGq6uOqGvWevoQbwKsnxt+paqeqbgI24H77Ix6j50fAv3HgcNyBbMdDkS5Job/xoicFFEu/RKQSmAO8DIxT1Z3erF3AuIDCAvgx7osd956XAo0JP8qgt+VUoA74hVfF9XMRySOFtqGqbgd+iDti3Ak0ActJre3YY6Dtlqq/of8F/M17nDIxishlwHZVfaPPrJSJcSDpkhRSmojkAw8B16tqc+I8byS6QC4RE5FLgFpVXR7E+pOUAcwFfqqqc4BW+lQVBbkNAbx6+ctwCWwikEc/1Q2pJujtNhgR+TquCvY3QceSSERygf8N3DpY2VSULkkhmfGiAyEiEVxC+I2qPuxN3t1zSun9rw0ovNOBS0VkM67K7X24+vsirxoEgt+WNUCNqr7sPX8QlyRSZRsCnAdsUtU6Ve0GHsZt21Tajj0G2m4p9RsSkU8ClwDXJgzhmyoxHo07AHjD++1UAK+JyHhSJ8YBpUtSSGa86BHn1c//N7BGVW9PmJU4dvUngP8Z6dgAVPVmVa1Q1UrcNvu7ql4LLMONqR1ofACqugvYJiLHepPOBd4iRbahZyuwQERyvc+8J8aU2Y4JBtpujwAf966eWQA0JVQzjSgRWYir0rxUVdsSZj0CXCUiWSIyFdeY+8pIx6eqb6rqWFWt9H47NcBc77uaMttxQKqaFn/AxbgrFd4Bvh50PF5MZ+BOz1cCK7y/i3H19k8B64EngZIUiPVs4C/e42m4H9sG4A9AVsCxzQaqve34J6A41bYh8G1gLbAK+BWQFfR2BO7HtXF043Zcnx5ouwGCu4LvHeBN3JVUQcW4AVcv3/ObuTuh/Ne9GNcBFwUVY5/5m4GyILfjofzZHc3GGGN6pUv1kTHGmCRYUjDGGNPLkoIxxphelhSMMcb0sqRgjDGmlyUFY0aQiJwtXm+zxqQiSwrGGGN6WVIwph8icp2IvCIiK0TkHnFjSuwTkR954yI8JSLlXtnZIvJSQv/+PWMQTBeRJ0XkDRF5TUSO9hafL/vHf/iNd5ezMSnBkoIxfYjI8cBHgdNVdTYQA67FdWRXraqzgGeAb3ovuQ+4UV3//m8mTP8NcKeqngy8B3fXK7jecK/Hje0xDdcPkjEpIWPwIsaknXOBU4BXvYP4HFzHcHHg916ZXwMPe+M5FKnqM970XwJ/EJECYJKq/hFAVTsAvOW9oqo13vMVQCXwvP9vy5jBWVIw5t0E+KWq3nzARJFv9Ck31D5iOhMex7DfoUkhVn1kzLs9BXxERMZC77jFR+F+Lz29ml4DPK+qTUCDiLzXm/4x4Bl1I+nViMjl3jKyvH72jUlpdoRiTB+q+paI3AI8LiIhXO+XX8IN4DPfm1eLa3cA18X03d5OfyPwKW/6x4B7ROQ73jKuGMG3YcyQWC+pxiRJRPapan7QcRjjJ6s+MsYY08vOFIwxxvSyMwVjjDG9LCkYY4zpZUnBGGNML0sKxhhjellSMMYY08uSgjHGmF7/H38oUHqT1gXuAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_history(history)"
   ]
  }
 ],
 "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": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}