[Detection]Pytorch_version.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import torch.optim as optim\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from sklearn.model_selection import train_test_split\n",
    "from torchvision import datasets, transforms\n",
    "from torch.autograd import Variable"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### data load"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>filename</th>\n",
       "      <th>db</th>\n",
       "      <th>e9</th>\n",
       "      <th>48</th>\n",
       "      <th>89</th>\n",
       "      <th>9d</th>\n",
       "      <th>b2</th>\n",
       "      <th>72</th>\n",
       "      <th>68</th>\n",
       "      <th>63</th>\n",
       "      <th>...</th>\n",
       "      <th>_free_locale</th>\n",
       "      <th>FlatSB_GetScrollRange</th>\n",
       "      <th>_ultoa</th>\n",
       "      <th>JetIndexRecordCount</th>\n",
       "      <th>SetUserGeoID</th>\n",
       "      <th>DispInvoke</th>\n",
       "      <th>DebugActiveProcessStop</th>\n",
       "      <th>OleCreateFromFileEx</th>\n",
       "      <th>AVIFileAddRef</th>\n",
       "      <th>KeQuerySystemTime</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>8c2da36f62b096324cab6f23a376c162.vir</td>\n",
       "      <td>0.000108</td>\n",
       "      <td>0.000107</td>\n",
       "      <td>0.000427</td>\n",
       "      <td>0.000746</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000637</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>b44bcdd49349d1ddb7019e0f0c8a4cce.vir</td>\n",
       "      <td>0.035262</td>\n",
       "      <td>0.039464</td>\n",
       "      <td>0.047589</td>\n",
       "      <td>0.065637</td>\n",
       "      <td>0.033758</td>\n",
       "      <td>0.039422</td>\n",
       "      <td>0.116690</td>\n",
       "      <td>0.155006</td>\n",
       "      <td>0.113825</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>9130ead71757488a4bcd16465014e88b.vir</td>\n",
       "      <td>0.156950</td>\n",
       "      <td>0.264558</td>\n",
       "      <td>0.370679</td>\n",
       "      <td>0.896217</td>\n",
       "      <td>0.031559</td>\n",
       "      <td>0.095725</td>\n",
       "      <td>0.226040</td>\n",
       "      <td>0.462257</td>\n",
       "      <td>0.113502</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4b5b336bed173f5a32e3cad21cc2d45f.vir</td>\n",
       "      <td>0.005949</td>\n",
       "      <td>0.009973</td>\n",
       "      <td>0.009283</td>\n",
       "      <td>0.084710</td>\n",
       "      <td>0.000660</td>\n",
       "      <td>0.001221</td>\n",
       "      <td>0.014041</td>\n",
       "      <td>0.011785</td>\n",
       "      <td>0.022270</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>37273913476c520aca9b2f26e95e80cb.vir</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 4911 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                               filename        db        e9        48  \\\n",
       "0  8c2da36f62b096324cab6f23a376c162.vir  0.000108  0.000107  0.000427   \n",
       "1  b44bcdd49349d1ddb7019e0f0c8a4cce.vir  0.035262  0.039464  0.047589   \n",
       "2  9130ead71757488a4bcd16465014e88b.vir  0.156950  0.264558  0.370679   \n",
       "3  4b5b336bed173f5a32e3cad21cc2d45f.vir  0.005949  0.009973  0.009283   \n",
       "4  37273913476c520aca9b2f26e95e80cb.vir  0.000000  0.000000  0.000000   \n",
       "\n",
       "         89        9d        b2        72        68        63  \\\n",
       "0  0.000746  0.000000  0.000000  0.000000  0.000637  0.000000   \n",
       "1  0.065637  0.033758  0.039422  0.116690  0.155006  0.113825   \n",
       "2  0.896217  0.031559  0.095725  0.226040  0.462257  0.113502   \n",
       "3  0.084710  0.000660  0.001221  0.014041  0.011785  0.022270   \n",
       "4  0.000000  0.000000  0.000000  0.000000  0.000000  0.000000   \n",
       "\n",
       "         ...          _free_locale  FlatSB_GetScrollRange  _ultoa  \\\n",
       "0        ...                     0                      0       0   \n",
       "1        ...                     0                      0       0   \n",
       "2        ...                     0                      0       0   \n",
       "3        ...                     0                      0       0   \n",
       "4        ...                     0                      0       0   \n",
       "\n",
       "   JetIndexRecordCount  SetUserGeoID  DispInvoke  DebugActiveProcessStop  \\\n",
       "0                    0             0           0                       0   \n",
       "1                    0             0           0                       0   \n",
       "2                    0             0           0                       0   \n",
       "3                    0             0           0                       0   \n",
       "4                    0             0           0                       0   \n",
       "\n",
       "   OleCreateFromFileEx  AVIFileAddRef  KeQuerySystemTime  \n",
       "0                    0              0                  0  \n",
       "1                    0              0                  0  \n",
       "2                    0              0                  0  \n",
       "3                    0              0                  0  \n",
       "4                    0              0                  0  \n",
       "\n",
       "[5 rows x 4911 columns]"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "opcode_api_csv = pd.read_csv('api_opcode.csv')\n",
    "opcode_api_csv.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "label = opcode_api_csv.pop('label')\n",
    "filename = opcode_api_csv.pop('filename')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Net(nn.Module):\n",
    "    \n",
    "    def __init__(self):\n",
    "        super(Net, self).__init__()\n",
    "        self.fc1 = nn.Linear(4909, 3000)\n",
    "        self.dout = nn.Dropout(0.2)\n",
    "        self.fc2 = nn.Linear(3000, 1500)\n",
    "        self.dout2 = nn.Dropout(0.2)\n",
    "        self.fc3 = nn.Linear(1500, 700)\n",
    "        self.dout3 = nn.Dropout(0.2)\n",
    "        self.fc4 = nn.Linear(700, 300)\n",
    "        self.fc5 = nn.Linear(300, 1)\n",
    "        \n",
    "    def forward(self, input_):\n",
    "        a1 = F.relu(self.fc1(input_))\n",
    "        dout1 = self.dout(a1)\n",
    "        a2 = F.relu(self.fc2(dout1))\n",
    "        dout2 = self.dout2(a2)\n",
    "        a3 = F.relu(self.fc3(dout2))\n",
    "        dout3 = self.dout3(a3)\n",
    "        a4 = F.relu(self.fc4(dout3))\n",
    "        return torch.sigmoid(self.fc5(a4))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = Net()\n",
    "\n",
    "criterion = nn.BCELoss()\n",
    "optimizer = optim.Adam(model.parameters(), lr=0.0001)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train, x_test, y_train, y_test = train_test_split(opcode_api_csv, label, stratify=label, test_size = 0.2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train_tensor = Variable(torch.FloatTensor(x_train.values))\n",
    "x_test_tensor = Variable(torch.FloatTensor(x_test.values))\n",
    "y_train_tensor = Variable(torch.FloatTensor(y_train.values))\n",
    "y_test_tensor = Variable(torch.FloatTensor(y_test.values))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "losses = []\n",
    "def train(epoch):\n",
    "    model.train()\n",
    "    optimizer.zero_grad()\n",
    "    output = model(x_train_tensor)\n",
    "    loss = criterion(output, y_train_tensor)\n",
    "    loss.backward()\n",
    "    optimizer.step()\n",
    "    losses.append(loss.data[0])\n",
    "    print(\"Epoch: {} loss: {}\".format(epoch, loss.data[0]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "def test():\n",
    "    model.eval()\n",
    "    correct = 0\n",
    "    output = model(x_test_tensor)\n",
    "#     print(output)\n",
    "    for i in range(len(output)):\n",
    "        if output[i].data[0] < 0.5:\n",
    "            result = 0\n",
    "        else:\n",
    "            result = 1\n",
    "        if result == y_test_tensor[i]:\n",
    "            correct += 1\n",
    "    return correct/len(x_test_tensor)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/jupyter/.local/lib/python3.5/site-packages/torch/nn/functional.py:1594: UserWarning: Using a target size (torch.Size([8000])) that is different to the input size (torch.Size([8000, 1])) is deprecated. Please ensure they have the same size.\n",
      "  \"Please ensure they have the same size.\".format(target.size(), input.size()))\n",
      "/home/jupyter/.local/lib/python3.5/site-packages/ipykernel_launcher.py:9: UserWarning: invalid index of a 0-dim tensor. This will be an error in PyTorch 0.5. Use tensor.item() to convert a 0-dim tensor to a Python number\n",
      "  if __name__ == '__main__':\n",
      "/home/jupyter/.local/lib/python3.5/site-packages/ipykernel_launcher.py:10: UserWarning: invalid index of a 0-dim tensor. This will be an error in PyTorch 0.5. Use tensor.item() to convert a 0-dim tensor to a Python number\n",
      "  # Remove the CWD from sys.path while we load stuff.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 0 loss: 0.7034906148910522\n",
      "Epoch: 1 loss: 0.6971545219421387\n",
      "Epoch: 2 loss: 0.6907854676246643\n",
      "Epoch: 3 loss: 0.6833271384239197\n",
      "Epoch: 4 loss: 0.6743916869163513\n",
      "Epoch: 5 loss: 0.6635416746139526\n",
      "Epoch: 6 loss: 0.6510078310966492\n",
      "Epoch: 7 loss: 0.6377418041229248\n",
      "Epoch: 8 loss: 0.6252204775810242\n",
      "Epoch: 9 loss: 0.6157894134521484\n",
      "Epoch: 10 loss: 0.6102935075759888\n",
      "Epoch: 11 loss: 0.6074745655059814\n",
      "Epoch: 12 loss: 0.6037571430206299\n",
      "Epoch: 13 loss: 0.5966548919677734\n",
      "Epoch: 14 loss: 0.5866448283195496\n",
      "Epoch: 15 loss: 0.576124370098114\n",
      "Epoch: 16 loss: 0.5664183497428894\n",
      "Epoch: 17 loss: 0.5580946803092957\n",
      "Epoch: 18 loss: 0.5505063533782959\n",
      "Epoch: 19 loss: 0.5434680581092834\n",
      "Epoch: 20 loss: 0.5365344285964966\n",
      "Epoch: 21 loss: 0.5297293066978455\n",
      "Epoch: 22 loss: 0.5220815539360046\n",
      "Epoch: 23 loss: 0.5138059258460999\n",
      "Epoch: 24 loss: 0.5053076148033142\n",
      "Epoch: 25 loss: 0.4976179301738739\n",
      "Epoch: 26 loss: 0.4889509975910187\n",
      "Epoch: 27 loss: 0.4799712598323822\n",
      "Epoch: 28 loss: 0.4716658890247345\n",
      "Epoch: 29 loss: 0.4618196189403534\n",
      "Epoch: 30 loss: 0.45111778378486633\n",
      "Epoch: 31 loss: 0.44090980291366577\n",
      "Epoch: 32 loss: 0.4314689636230469\n",
      "Epoch: 33 loss: 0.42252248525619507\n",
      "Epoch: 34 loss: 0.41388192772865295\n",
      "Epoch: 35 loss: 0.4060359001159668\n",
      "Epoch: 36 loss: 0.39883118867874146\n",
      "Epoch: 37 loss: 0.39143675565719604\n",
      "Epoch: 38 loss: 0.3841865062713623\n",
      "Epoch: 39 loss: 0.37630873918533325\n",
      "Epoch: 40 loss: 0.37029212713241577\n",
      "Epoch: 41 loss: 0.36399364471435547\n",
      "Epoch: 42 loss: 0.3571876585483551\n",
      "Epoch: 43 loss: 0.3510580062866211\n",
      "Epoch: 44 loss: 0.3439529836177826\n",
      "Epoch: 45 loss: 0.33773311972618103\n",
      "Epoch: 46 loss: 0.3321707844734192\n",
      "Epoch: 47 loss: 0.32467570900917053\n",
      "Epoch: 48 loss: 0.3170414865016937\n",
      "Epoch: 49 loss: 0.3102419078350067\n",
      "Epoch: 50 loss: 0.30246150493621826\n",
      "Epoch: 51 loss: 0.2957218587398529\n",
      "Epoch: 52 loss: 0.28702083230018616\n",
      "Epoch: 53 loss: 0.2808891236782074\n",
      "Epoch: 54 loss: 0.2725963294506073\n",
      "Epoch: 55 loss: 0.26316437125205994\n",
      "Epoch: 56 loss: 0.25552791357040405\n",
      "Epoch: 57 loss: 0.24837322533130646\n",
      "Epoch: 58 loss: 0.2404487580060959\n",
      "Epoch: 59 loss: 0.2325943410396576\n",
      "Epoch: 60 loss: 0.22820237278938293\n",
      "Epoch: 61 loss: 0.22303776443004608\n",
      "Epoch: 62 loss: 0.21605901420116425\n",
      "Epoch: 63 loss: 0.21105851233005524\n",
      "Epoch: 64 loss: 0.20879843831062317\n",
      "Epoch: 65 loss: 0.20349545776844025\n",
      "Epoch: 66 loss: 0.20373649895191193\n",
      "Epoch: 67 loss: 0.19964277744293213\n",
      "Epoch: 68 loss: 0.19516782462596893\n",
      "Epoch: 69 loss: 0.19238445162773132\n",
      "Epoch: 70 loss: 0.1881597936153412\n",
      "Epoch: 71 loss: 0.1859806925058365\n",
      "Epoch: 72 loss: 0.18281656503677368\n",
      "Epoch: 73 loss: 0.17712771892547607\n",
      "Epoch: 74 loss: 0.17582760751247406\n",
      "Epoch: 75 loss: 0.1713038831949234\n",
      "Epoch: 76 loss: 0.1684529185295105\n",
      "Epoch: 77 loss: 0.16548700630664825\n",
      "Epoch: 78 loss: 0.1609751433134079\n",
      "Epoch: 79 loss: 0.15766575932502747\n",
      "Epoch: 80 loss: 0.15613077580928802\n",
      "Epoch: 81 loss: 0.15490567684173584\n",
      "Epoch: 82 loss: 0.1520448476076126\n",
      "Epoch: 83 loss: 0.15112744271755219\n",
      "Epoch: 84 loss: 0.14787302911281586\n",
      "Epoch: 85 loss: 0.14553794264793396\n",
      "Epoch: 86 loss: 0.14465877413749695\n",
      "Epoch: 87 loss: 0.14015163481235504\n",
      "Epoch: 88 loss: 0.13939763605594635\n",
      "Epoch: 89 loss: 0.1375330686569214\n",
      "Epoch: 90 loss: 0.13600493967533112\n",
      "Epoch: 91 loss: 0.1339719146490097\n",
      "Epoch: 92 loss: 0.13198669254779816\n",
      "Epoch: 93 loss: 0.13072262704372406\n",
      "Epoch: 94 loss: 0.12936177849769592\n",
      "Epoch: 95 loss: 0.1270846277475357\n",
      "Epoch: 96 loss: 0.12609022855758667\n",
      "Epoch: 97 loss: 0.12276768684387207\n",
      "Epoch: 98 loss: 0.12111049145460129\n",
      "Epoch: 99 loss: 0.11947270482778549\n",
      "Epoch: 100 loss: 0.11879464983940125\n",
      "Epoch: 101 loss: 0.1175481304526329\n",
      "Epoch: 102 loss: 0.11478841304779053\n",
      "Epoch: 103 loss: 0.11426545679569244\n",
      "Epoch: 104 loss: 0.11178924143314362\n",
      "Epoch: 105 loss: 0.11181551963090897\n",
      "Epoch: 106 loss: 0.10868370532989502\n",
      "Epoch: 107 loss: 0.10999268293380737\n",
      "Epoch: 108 loss: 0.10696283727884293\n",
      "Epoch: 109 loss: 0.10728182643651962\n",
      "Epoch: 110 loss: 0.10645892471075058\n",
      "Epoch: 111 loss: 0.10709967464208603\n",
      "Epoch: 112 loss: 0.10423851758241653\n",
      "Epoch: 113 loss: 0.10404498875141144\n",
      "Epoch: 114 loss: 0.10361270606517792\n",
      "Epoch: 115 loss: 0.10291419923305511\n",
      "Epoch: 116 loss: 0.10120899230241776\n",
      "Epoch: 117 loss: 0.10239804536104202\n",
      "Epoch: 118 loss: 0.10181006044149399\n",
      "Epoch: 119 loss: 0.0997437909245491\n",
      "Epoch: 120 loss: 0.09975595027208328\n",
      "Epoch: 121 loss: 0.09949031472206116\n",
      "Epoch: 122 loss: 0.09837225824594498\n",
      "Epoch: 123 loss: 0.09827689081430435\n",
      "Epoch: 124 loss: 0.0985187292098999\n",
      "Epoch: 125 loss: 0.09598085284233093\n",
      "Epoch: 126 loss: 0.09728624671697617\n",
      "Epoch: 127 loss: 0.0972466692328453\n",
      "Epoch: 128 loss: 0.09756078571081161\n",
      "Epoch: 129 loss: 0.09640148282051086\n",
      "Epoch: 130 loss: 0.09628680348396301\n",
      "Epoch: 131 loss: 0.09573043137788773\n",
      "Epoch: 132 loss: 0.09480232745409012\n",
      "Epoch: 133 loss: 0.09505493938922882\n",
      "Epoch: 134 loss: 0.09463086724281311\n",
      "Epoch: 135 loss: 0.09454813599586487\n",
      "Epoch: 136 loss: 0.09375406801700592\n",
      "Epoch: 137 loss: 0.09513773024082184\n",
      "Epoch: 138 loss: 0.09449281543493271\n",
      "Epoch: 139 loss: 0.09453947097063065\n",
      "Epoch: 140 loss: 0.09148015081882477\n",
      "Epoch: 141 loss: 0.0944453626871109\n",
      "Epoch: 142 loss: 0.09239565581083298\n",
      "Epoch: 143 loss: 0.0929763987660408\n",
      "Epoch: 144 loss: 0.09165210276842117\n",
      "Epoch: 145 loss: 0.09080437570810318\n",
      "Epoch: 146 loss: 0.09295041114091873\n",
      "Epoch: 147 loss: 0.09231268614530563\n",
      "Epoch: 148 loss: 0.09065846353769302\n",
      "Epoch: 149 loss: 0.09158209711313248\n",
      "Epoch: 150 loss: 0.09027893841266632\n",
      "Epoch: 151 loss: 0.09065470099449158\n",
      "Epoch: 152 loss: 0.09027822315692902\n",
      "Epoch: 153 loss: 0.08979298919439316\n",
      "Epoch: 154 loss: 0.09040874242782593\n",
      "Epoch: 155 loss: 0.09001773595809937\n",
      "Epoch: 156 loss: 0.09052065014839172\n",
      "Epoch: 157 loss: 0.0895896926522255\n",
      "Epoch: 158 loss: 0.0898764431476593\n",
      "Epoch: 159 loss: 0.0891517847776413\n",
      "Epoch: 160 loss: 0.08815798908472061\n",
      "Epoch: 161 loss: 0.08888909220695496\n",
      "Epoch: 162 loss: 0.0886683240532875\n",
      "Epoch: 163 loss: 0.08775609731674194\n",
      "Epoch: 164 loss: 0.08876501768827438\n",
      "Epoch: 165 loss: 0.08855918794870377\n",
      "Epoch: 166 loss: 0.08901268243789673\n",
      "Epoch: 167 loss: 0.08798564225435257\n",
      "Epoch: 168 loss: 0.08775147795677185\n",
      "Epoch: 169 loss: 0.08802195638418198\n",
      "Epoch: 170 loss: 0.08791891485452652\n",
      "Epoch: 171 loss: 0.08742786943912506\n",
      "Epoch: 172 loss: 0.08729388564825058\n",
      "Epoch: 173 loss: 0.08766861259937286\n",
      "Epoch: 174 loss: 0.08812962472438812\n",
      "Epoch: 175 loss: 0.08755504339933395\n",
      "Epoch: 176 loss: 0.08652415126562119\n",
      "Epoch: 177 loss: 0.08786805719137192\n",
      "Epoch: 178 loss: 0.0872928649187088\n",
      "Epoch: 179 loss: 0.08618340641260147\n",
      "Epoch: 180 loss: 0.08743332326412201\n",
      "Epoch: 181 loss: 0.08745431154966354\n",
      "Epoch: 182 loss: 0.08667850494384766\n",
      "Epoch: 183 loss: 0.08606745302677155\n",
      "Epoch: 184 loss: 0.08577928692102432\n",
      "Epoch: 185 loss: 0.08589600771665573\n",
      "Epoch: 186 loss: 0.0862802192568779\n",
      "Epoch: 187 loss: 0.08512768149375916\n",
      "Epoch: 188 loss: 0.085887610912323\n",
      "Epoch: 189 loss: 0.0851324275135994\n",
      "Epoch: 190 loss: 0.08623159676790237\n",
      "Epoch: 191 loss: 0.08483611047267914\n",
      "Epoch: 192 loss: 0.08600163459777832\n",
      "Epoch: 193 loss: 0.08567793667316437\n",
      "Epoch: 194 loss: 0.08584589511156082\n",
      "Epoch: 195 loss: 0.08536700159311295\n",
      "Epoch: 196 loss: 0.08495458960533142\n",
      "Epoch: 197 loss: 0.08475518971681595\n",
      "Epoch: 198 loss: 0.08503162860870361\n",
      "Epoch: 199 loss: 0.08510657399892807\n",
      "Epoch: 200 loss: 0.08581960201263428\n",
      "Epoch: 201 loss: 0.08496811240911484\n",
      "Epoch: 202 loss: 0.08414904773235321\n",
      "Epoch: 203 loss: 0.0852581262588501\n",
      "Epoch: 204 loss: 0.08462855219841003\n",
      "Epoch: 205 loss: 0.08466541767120361\n",
      "Epoch: 206 loss: 0.08423614501953125\n",
      "Epoch: 207 loss: 0.08411901444196701\n",
      "Epoch: 208 loss: 0.08336856961250305\n",
      "Epoch: 209 loss: 0.0850711464881897\n",
      "Epoch: 210 loss: 0.08532068878412247\n",
      "Epoch: 211 loss: 0.08468801528215408\n",
      "Epoch: 212 loss: 0.08520448952913284\n",
      "Epoch: 213 loss: 0.08448870480060577\n",
      "Epoch: 214 loss: 0.08476405590772629\n",
      "Epoch: 215 loss: 0.08385846763849258\n",
      "Epoch: 216 loss: 0.0846051499247551\n",
      "Epoch: 217 loss: 0.08398392051458359\n",
      "Epoch: 218 loss: 0.08355067670345306\n",
      "Epoch: 219 loss: 0.08337641507387161\n",
      "Epoch: 220 loss: 0.08389639854431152\n",
      "Epoch: 221 loss: 0.08381924778223038\n",
      "Epoch: 222 loss: 0.08443630486726761\n",
      "Epoch: 223 loss: 0.08480577915906906\n",
      "Epoch: 224 loss: 0.08393730968236923\n",
      "Epoch: 225 loss: 0.08310941606760025\n",
      "Epoch: 226 loss: 0.08420458436012268\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 227 loss: 0.08345043659210205\n",
      "Epoch: 228 loss: 0.08378884941339493\n",
      "Epoch: 229 loss: 0.08476356416940689\n",
      "Epoch: 230 loss: 0.08316129446029663\n",
      "Epoch: 231 loss: 0.08399701863527298\n",
      "Epoch: 232 loss: 0.08320348709821701\n",
      "Epoch: 233 loss: 0.08415865898132324\n",
      "Epoch: 234 loss: 0.08456573635339737\n",
      "Epoch: 235 loss: 0.08324511349201202\n",
      "Epoch: 236 loss: 0.08336759358644485\n",
      "Epoch: 237 loss: 0.08244500309228897\n",
      "Epoch: 238 loss: 0.08350621163845062\n",
      "Epoch: 239 loss: 0.08231078833341599\n",
      "Epoch: 240 loss: 0.08317428082227707\n",
      "Epoch: 241 loss: 0.08316711336374283\n",
      "Epoch: 242 loss: 0.0828072726726532\n",
      "Epoch: 243 loss: 0.08218219131231308\n",
      "Epoch: 244 loss: 0.08276072889566422\n",
      "Epoch: 245 loss: 0.08213475346565247\n",
      "Epoch: 246 loss: 0.08262613415718079\n",
      "Epoch: 247 loss: 0.08247216790914536\n",
      "Epoch: 248 loss: 0.08255460113286972\n",
      "Epoch: 249 loss: 0.08241745829582214\n",
      "Epoch: 250 loss: 0.08312124758958817\n",
      "Epoch: 251 loss: 0.0824136734008789\n",
      "Epoch: 252 loss: 0.08323855698108673\n",
      "Epoch: 253 loss: 0.08237472176551819\n",
      "Epoch: 254 loss: 0.08248629420995712\n",
      "Epoch: 255 loss: 0.08293568342924118\n",
      "Epoch: 256 loss: 0.0826055109500885\n",
      "Epoch: 257 loss: 0.08249296247959137\n",
      "Epoch: 258 loss: 0.08394568413496017\n",
      "Epoch: 259 loss: 0.0811343640089035\n",
      "Epoch: 260 loss: 0.08227953314781189\n",
      "Epoch: 261 loss: 0.08155507594347\n",
      "Epoch: 262 loss: 0.08328551799058914\n",
      "Epoch: 263 loss: 0.08304763585329056\n",
      "Epoch: 264 loss: 0.08478441834449768\n",
      "Epoch: 265 loss: 0.08227290213108063\n",
      "Epoch: 266 loss: 0.08925263583660126\n",
      "Epoch: 267 loss: 0.08199905604124069\n",
      "Epoch: 268 loss: 0.0874256044626236\n",
      "Epoch: 269 loss: 0.08327188342809677\n",
      "Epoch: 270 loss: 0.08230127394199371\n",
      "Epoch: 271 loss: 0.08356504142284393\n",
      "Epoch: 272 loss: 0.08606594055891037\n",
      "Epoch: 273 loss: 0.08222034573554993\n",
      "Epoch: 274 loss: 0.08285151422023773\n",
      "Epoch: 275 loss: 0.08367996662855148\n",
      "Epoch: 276 loss: 0.08291271328926086\n",
      "Epoch: 277 loss: 0.08311636745929718\n",
      "Epoch: 278 loss: 0.08247392624616623\n",
      "Epoch: 279 loss: 0.08337192982435226\n",
      "Epoch: 280 loss: 0.0823005884885788\n",
      "Epoch: 281 loss: 0.08215230703353882\n",
      "Epoch: 282 loss: 0.08202942460775375\n",
      "Epoch: 283 loss: 0.08188476413488388\n",
      "Epoch: 284 loss: 0.08245696127414703\n",
      "Epoch: 285 loss: 0.08167150616645813\n",
      "Epoch: 286 loss: 0.08300863951444626\n",
      "Epoch: 287 loss: 0.08210480213165283\n",
      "Epoch: 288 loss: 0.08125333487987518\n",
      "Epoch: 289 loss: 0.08102250844240189\n",
      "Epoch: 290 loss: 0.08167039602994919\n",
      "Epoch: 291 loss: 0.08109275996685028\n",
      "Epoch: 292 loss: 0.08079764991998672\n",
      "Epoch: 293 loss: 0.08183188736438751\n",
      "Epoch: 294 loss: 0.08143987506628036\n",
      "Epoch: 295 loss: 0.08174607157707214\n",
      "Epoch: 296 loss: 0.08039185404777527\n",
      "Epoch: 297 loss: 0.08123613148927689\n",
      "Epoch: 298 loss: 0.08188732713460922\n",
      "Epoch: 299 loss: 0.08054710924625397\n"
     ]
    }
   ],
   "source": [
    "for epoch in range(300):\n",
    "    train(epoch)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAABIQAAAEyCAYAAACLeQv5AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4wLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvqOYd8AAAIABJREFUeJzs3Xd81fWh//H35+zkZE8gYW8ERAhDcFVtL9ZWW/euOLBau+69/bW3t7f3tr1d2vZ2uXBg67aOOqrVtm6GEBDZYJgBAtk7Z39+f5yYRkWImuSb8Xo+HnmQc84n57xzMuC8+QxjrRUAAAAAAAAGD5fTAQAAAAAAANC7KIQAAAAAAAAGGQohAAAAAACAQYZCCAAAAAAAYJChEAIAAAAAABhkKIQAAAAAAAAGGQohAAAAAACAQYZCCAAAAAAAYJChEAIAAAAAABhkPE49cF5enh01apRTDw8AAAAAADDgrFmzptpam3+0cY4VQqNGjVJpaalTDw8AAAAAADDgGGP2dGUcS8YAAAAAAAAGGQohAAAAAACAQYZCCAAAAAAAYJChEAIAAAAAABhkKIQAAAAAAAAGGQohAAAAAACAQYZCCAAAAAAAYJChEAIAAAAAABhkulQIGWMWGmO2GWPKjDHfOczt/2eMWdf+tt0YU9/9UQEAAAAAANAdjloIGWPckm6RdIakKZIuNsZM6TzGWvtNa+0Ma+0MSb+T9ERPhO2LHl1drvLaVqdjAAAAAAAAdFlXZgjNkVRmrd1prY1IeljS2UcYf7Gkh7ojXF/X0BrVT57fos/+9nU9v6HC6TgAAAAAAABd0pVCqEhSeafL+9qv+wBjzEhJoyW99Mmj9X2ZqV49c+MJGpOfpusfWKvvP7VRoWjc6VgAAAAAAABH1N2bSl8k6TFr7WFbEWPMYmNMqTGmtKqqqpsf2hnDc1L1p+uO17UnjtYfV+zRubct167qFqdjAQAAAAAAfKiuFEL7JQ3vdLm4/brDuUhHWC5mrV1irS2x1pbk5+d3PWUf5/O49J9nTtHdXyrR/vo2fe63r+updR/2FAEAAAAAADirK4XQaknjjTGjjTE+JUufp98/yBgzSVK2pBXdG7H/OG1yoZ772omaMixDX394nb792HqWkAEAAAAAgD7nqIWQtTYm6UZJL0jaIulRa+0mY8wPjTFndRp6kaSHrbW2Z6L2D8OyUvTQtfN046fG6dE15brkzpWqaQ47HQsAAAAAAKCDcaq/KSkpsaWlpY48dm95fkOFvvHIOg3NDGjpojkanRd0OhIAAAAAABjAjDFrrLUlRxvX3ZtKo5Mzpg3Vg9fOU2MopnNuXaY1e+qcjgQAAAAAAEAh1NNmjczWE9fPV1aqT1fc/abW7qUUAgAAAAAAzqIQ6gWj8oJ6ePE85aX7deU9q7TpQIPTkQAAAAAAwCBGIdRLCjMCeuCauUrze3T53av0zqEmpyMBAAAAAIBBikKoFxVnp+rBa+fJ7TK69K43VdHQ5nQkAAAAAAAwCFEI9bJReUHdf/VctYRj+vL9axWOxZ2OBAAAAAAABhkKIQdMHJKuX15wrN4ur9cPntnsdBwAAAAAADDIUAg5ZOHUobr+lLF68M29enR1udNxAAAAAADAIEIh5KB//8xEnTAuT997aqM27OPkMQAAAAAA0DsohBzkdhn99uLjlBf06esPv6W2CPsJAQAAAACAnkch5LCcoE+/OP9Y7axu0U0vbHU6DgAAAAAAGAQohPqA+ePydOX8UVq6bLeW76h2Og4AAAAAABjgKIT6iG8vnKTReUF960/r1RSKOh0HAAAAAAAMYBRCfUSKz61fXnCsKhra9KNnOYoeAAAAAAD0HAqhPmTmiGxdd/JYPVq6T69tr3I6DgAAAAAAGKAohPqYr582XmPyg/rukxvUEo45HQcAAAAAAAxAFEJ9TMDr1s/Oma59dW365YvbnY4DAAAAAAAGIAqhPmjO6BxdNm+Eli7fpbf21jkdBwAAAAAADDAUQn3UtxdO0pCMgL79+HpFYgmn4wAAAAAAgAGEQqiPSg949b9fmKrth5r1u5fecToOAAAAAAAYQCiE+rDTJhfq3JnFuvWVHVpXXu90HAAAAAAAMEBQCPVx/33WFBWm+/Wvj65TKBp3Og4AAAAAABgAKIT6uIyAVzeff6x2VrXo53/d6nQcAAAAAAAwAFAI9QMLxuXpS8eP1NJlu7W8rNrpOAAAAAAAoJ+jEOonvnPGZI3JC+rGh97Szqpmp+MAAAAAAIB+jEKon0jxuXXXl0okSVfcs0qVjSGHEwEAAAAAgP6KQqgfGZOfpqVXzlZtS0RX3LNKjaGo05EAAAAAAEA/RCHUzxw7PEu3XzZLZZXNuubeUlU1hZ2OBAAAAAAA+hkKoX7opAn5+tWFM7SuvF6n/+pV/am0XNZap2MBAAAAAIB+gkKonzrr2GF67usnakJhmr712Hpdfvcqle6uVSJBMQQAAAAAAI7MODWzpKSkxJaWljry2ANJImH1wKq9+vnzW9Ucjqkww68zpg7VWTOG6bjhWTLGOB0RAAAAAAD0EmPMGmttyVHHdaUQMsYslPQbSW5Jd1lrf3aYMRdI+h9JVtLb1tpLjnSfFELdqykU1UtbK/WX9RV6ZXuVIrGEphVlatGCUfrc9GHyeZgMBgAAAADAQNdthZAxxi1pu6RPS9onabWki621mzuNGS/pUUmnWmvrjDEF1trKI90vhVDPaQ7H9ORb+7V02S7trGpRQbpfvzj/WJ00Id/paAAAAAAAoAd1tRDqyrSROZLKrLU7rbURSQ9LOvt9Y66VdIu1tk6SjlYGoWel+T26fN5I/f2bJ+veRbOVnerTlUtXaemyXWw+DQAAAAAAulQIFUkq73R5X/t1nU2QNMEYs8wYs7J9idkHGGMWG2NKjTGlVVVVHy8xuszlMjplYoEev2G+Tp1UqB88s1nffXKjovGE09EAAAAAAICDumtjGY+k8ZJOkXSxpDuNMVnvH2StXWKtLbHWluTns3ypt6T5PVpy+Sxdf8pYPbRqr66/f43inEYGAAAAAMCg1ZVCaL+k4Z0uF7df19k+SU9ba6PW2l1K7jk0vnsioju4XEbfXjhJ//P5Kfr7lkr96NnNR/8gAAAAAAAwIHWlEFotabwxZrQxxifpIklPv2/Mn5WcHSRjTJ6SS8h2dmNOdJMrF4zWVQtG697lu/WH5budjgMAAAAAABxw1ELIWhuTdKOkFyRtkfSotXaTMeaHxpiz2oe9IKnGGLNZ0suSvmWtremp0Phk/vPMyTp9cqF+8MwmvbyV/b8BAAAAABhsjnrsfE/h2HlntYRjuuCOFdpd3aInbligiUPSnY4EAAAAAAA+oe48dh4DUNDv0d1fmq1Uv0fX/rFU9a0RpyMBAAAAAIBeQiE0iA3JDOj2y2bpYENINz74lmIcRw8AAAAAwKBAITTIzRqZrf/94lS9UVatnzy31ek4AAAAAACgF3icDgDnXVAyXJsPNOqeZbs0cUiaLpw9wulIAAAAAACgB1EIQZL0vTMna0dVs7775EYVZgR0ysQCpyMBAAAAAIAewpIxSJI8bpduvXSmJham64YH1mrj/ganIwEAAAAAgB5CIYQO6QGvli6arexUnxbdu1rlta1ORwIAAAAAAD2AQgjvUZgR0NJFsxWOxnXFPatU0dDmdCQAAAAAANDNKITwARMK07V00WxVNYV14R0rta+OmUIAAAAAAAwkFEI4rFkjc3Tf1XNU1xrRhXes1N4aSiEAAAAAAAYKCiF8qONGZOuha+epJRLTBXes0DuHmpyOBAAAAAAAugGFEI5oalGmHrp2nuLW6tzblmvVrlqnIwEAAAAAgE+IQghHNXlohp64fr7y0v267O439fyGCqcjAQAAAACAT4BCCF0yPCdVj395vqYVZeqGB9fqrtd3ylrrdCwAAAAAAPAxUAihy7KDPj1wzVwtPGaI/vcvW/QfT2xQJJZwOhYAAAAAAPiIKITwkQS8bt1yyUx99dRxenh1uS67+03VtkScjgUAAAAAAD4CCiF8ZC6X0b99ZqJ+c9EMrSuv11m/f0PbDnICGQAAAAAA/QWFED62s2cU6dHrjlckltA5ty7TP7YccjoSAAAAAADoAgohfCIzhmfp6RtP0Jj8NF3zx1Ld8eoONpsGAAAAAKCPoxDCJzYkM6BHrzten506VD99fqv+/U/rFY7FnY4FAAAAAAA+BIUQukWKz63fX3KcvnH6eD2+dp8uufNNVTeHnY4FAAAAAAAOg0II3cYYo2+cPkG3XDJTmw406OzfL9OWikanYwEAAAAAgPehEEK3O3P6UP3puvmKJ6zOvW25Xth00OlIAAAAAACgEwoh9IhpxZl6+sYFGl+QpuvuW6NbXi5js2kAAAAAAPoICiH0mIKMgB657niddeww3fzCNn3zkXVsNg0AAAAAQB/gcToABraA163fXDRDEwrT9IsXt6u+LarbL5ulgNftdDQAAAAAAAYtZgihxxljdOOp4/XTc6bp1e1Vuure1WqNxJyOBQAAAADAoEUhhF5z8ZwR+uX5x2rlzhp96Z5VagpFnY4EAAAAAMCgRCGEXnXOzGL99uLjtHZvvS6/e5Ua2iiFAAAAAADobRRC6HWfmz5Mt146U5sONOjSu1aqriXidCQAAAAAAAYVCiE44l+OGaIlV5Ro+6FmXXznSlU1hZ2OBAAAAADAoNGlQsgYs9AYs80YU2aM+c5hbr/SGFNljFnX/nZN90fFQPOpiQVaeuVs7alp1UVLVuhgQ8jpSAAAAAAADApHLYSMMW5Jt0g6Q9IUSRcbY6YcZugj1toZ7W93dXNODFALxuXpD1fN0aHGsC5cskL769ucjgQAAAAAwIDXlRlCcySVWWt3Wmsjkh6WdHbPxsJgMmd0ju67eo5qWyK64PYV2lPT4nQkAAAAAAAGtK4UQkWSyjtd3td+3fuda4xZb4x5zBgz/HB3ZIxZbIwpNcaUVlVVfYy4GKiOG5Gth66dp9ZITBfesVI7qpqdjgQAAAAAwIDVXZtKPyNplLV2uqS/SfrD4QZZa5dYa0ustSX5+fnd9NAYKKYWZeqhxfMUSyR04R0rtf1Qk9ORAAAAAAAYkLpSCO2X1HnGT3H7dR2stTXW2nePibpL0qzuiYfBZtKQDD28+Hi5XdJFS1Zq04EGpyMBAAAAADDgdKUQWi1pvDFmtDHGJ+kiSU93HmCMGdrp4lmStnRfRAw24wrS9Oh1xyvF69bFS1ZqXXm905EAAAAAABhQjloIWWtjkm6U9IKSRc+j1tpNxpgfGmPOah/2NWPMJmPM25K+JunKngqMwWFkblCPXDdPWak+XX73m9q4n5lCAAAAAAB0F2OtdeSBS0pKbGlpqSOPjf5jf32bLrh9hdqicT163TyNK0h3OhIAAAAAAH2WMWaNtbbkaOO6a1NpoEcUZaXogWvmyu0yuuTONzmSHgAAAACAbkAhhD5vVF5Q9189V9F4Qpfc+abKa1udjgQAAAAAQL9GIYR+YeKQdP3xqrlqCkV10ZKV2l3NTCEAAAAAAD4uCiH0G9OKM/XgtfPUGonpwiUrVFbZ7HQkAAAAAAD6JQoh9CtTizL18OLjFU9YXbRkpbYfanI6EgAAAAAA/Q6FEPqdiUPS9fDi4+Uy0pfuWaWDDSGnIwEAAAAA0K9QCKFfGleQpnsXzVFjW1RX3btazeGY05EAAAAAAOg3KITQb00ZlqFbLp2pbYeadOODaxWLJ5yOBAAAAABAv0AhhH7tlIkF+tHZU/XKtip9/+lNstY6HQkAAAAAgD7P43QA4JO6ZO4I7a1t1e2v7tDkoRm6fN5IpyMBAAAAANCnMUMIA8K3/mWiPjUxXz94epNW7651Og4AAAAAAH0ahRAGBLfL6NcXHafi7BRdf/9aTh4DAAAAAOAIKIQwYGSmeLXkihK1RWL68v1rFIrGnY4EAAAAAECfRCGEAWVCYbp+ecGxWlder+8+uYFNpgEAAAAAOAwKIQw4C6cO1TdPn6An1u7Xba/ucDoOAAAAAAB9DqeMYUD62mnjVFbVrJv+uk1j8tK0cOoQpyMBAAAAANBnMEMIA5IxRjefN10zhmfpm4+s08b9DU5HAgAAAACgz6AQwoAV8Lq15IpZyk716qp7V6u8ttXpSAAAAAAA9AkUQhjQCtIDWrpojkLRuL50zyrVtkScjgQAAAAAgOMohDDgTRySrruvnK399W1adO9qtYRjTkcCAAAAAMBRFEIYFGaPytHvL5mpDfvqdf0DaxWJJZyOBAAAAACAYyiEMGh8ekqhfnrONL22vUo3PrhW0TilEAAAAABgcKIQwqBy4ewR+p/PT9GLmw/pG4+sU4xSCAAAAAAwCHmcDgD0tisXjFYkntBPntsqn9ulX5x/rNwu43QsAAAAAAB6DYUQBqXFJ41VJJbQL17cLq/b6GfnTJeLUggAAAAAMEhQCGHQuvHU8YrEEvrtS2XyeVz60dlTZQylEAAAAABg4KMQwqD2zU9PUDie0B2v7pTX7dL3PzeFUggAAAAAMOBRCGFQM8boOwsnKRJLaOmy3fK6XfqPMyZRCgEAAAAABjQKIQx6xhh9/3NTFItbLXltp5rDMf3o7KlsNA0AAAAAGLAohAAlS6Efnn2M0gMe3frKDtW3RvR/F86Q3+N2OhoAAAAAAN3O1ZVBxpiFxphtxpgyY8x3jjDuXGOMNcaUdF9EoHcYY/T/Fk7S986crOc2HNSipavVFIo6HQsAAAAAgG531ELIGOOWdIukMyRNkXSxMWbKYcalS/q6pDe7OyTQm645cYz+78JjtWpXrc67bYX21bU6HQkAAAAAgG7VlRlCcySVWWt3Wmsjkh6WdPZhxv1I0s8lhboxH+CILx5XrD9cNUcHGtr0hVuWa115vdORAAAAAADoNl0phIoklXe6vK/9ug7GmJmShltr/9KN2QBHLRiXpydvmK9Un1sX3rFCz22ocDoSAAAAAADdokt7CB2JMcYl6VeS/q0LYxcbY0qNMaVVVVWf9KGBHjeuIF1P3jBfU4sydcMDa3XLy2Wy1jodCwAAAACAT6QrhdB+ScM7XS5uv+5d6ZKmSnrFGLNb0jxJTx9uY2lr7RJrbYm1tiQ/P//jpwZ6UW6aXw9cM1dnzximm1/Ypm89tl6RWMLpWAAAAAAAfGxdOXZ+taTxxpjRShZBF0m65N0brbUNkvLevWyMeUXSv1trS7s3KuCcgNetX184Q6Pzgvr1399ReW2r7rh8lrJSfU5HAwAAAADgIzvqDCFrbUzSjZJekLRF0qPW2k3GmB8aY87q6YBAX2GM0TdOn6BfXzhDb+2t1zm3Lteu6hanYwEAAAAA8JEZp/ZDKSkpsaWlTCJC/7R6d62uu2+NEtbqjstmae6YXKcjAQAAAAAgY8waa+0HtvF5v0+8qTQwGM0elaMnb5iv3KBPl939pp5at//oHwQAAAAAQB9BIQR8TCNzg3ri+gWaNTJbX394nZYu2+V0JAAAAAAAuoRCCPgEMlO9unfRHH1mSqF+8Mxm/erFbRxLDwAAAADo8yiEgE8o4HXr1ktn6vxZxfrtS2X67pMbFIrGnY4FAAAAAMCH6sqx8wCOwuN26abzpisv3a/bXtmht/bW63cXH6fxhelORwMAAAAA4AOYIQR0E2OMvr1wku65skRVTWF97ndv6P6Ve1hCBgAAAADocyiEgG526qRCPf/1EzVndI6+9+eNuu6+NapriTgdCwAAAACADhRCQA8oyAjoD4vm6HtnTtbL2yq18DevaXlZtdOxAAAAAACQRCEE9BiXy+iaE8foyRsWKOj36NK739QvXtimeIIlZAAAAAAAZ1EIAT1salGmnv3qCTpvZrF+/3KZrrp3tRpao07HAgAAAAAMYhRCQC9I9Xl003nT9eMvTtXyHdX6/O/f0NaDjU7HAgAAAAAMUhRCQC8xxujSuSP18OLjFYrG9cVbluvhVXs5hQwAAAAA0OsohIBeNmtktp796gmaOTJL33lig66/f63qWzmFDAAAAADQeyiEAAcUZAR031Vz9R9nTNI/th7Swl+/rtLdtU7HAgAAAAAMEhRCgENcLqPrTh6rJ29YoIDXpYvvXKk/lZY7HQsAAAAAMAhQCAEOm1qUqae+coLmjs7Vtx5br588t4Wj6QEAAAAAPYpCCOgDMlO9Wrpotq44fqSWvLZTV927WjXNYadjAQAAAAAGKAohoI/wul364dlT9eMvTtWKnTU64zeva1lZtdOxAAAAAAADEIUQ0MdcOnek/nzDAqUHPLrs7jf1879uVSSWcDoWAAAAAGAAoRAC+qApwzL0zFdP0IUlw3XbKzv0+d+9obfL652OBQAAAAAYICiEgD4q1efRz86drruuKFFDW1RfvHWZfvyXzWqLxJ2OBgAAAADo5yiEgD7u9CmFevFfT9JFc0boztd36dzblutQY8jpWAAAAACAfoxCCOgHMgJe/eSL07T0ytnaXdOic25drrLKJqdjAQAAAAD6KQohoB/51KQCPbL4eIVjCZ172wqt2lXrdCQAAAAAQD9EIQT0M9OKM/XkDfOVm+bTZXe/qec2VDgdCQAAAADQz1AIAf3Q8JxUPf7l+ZpWlKmvPLhWd7+xy+lIAAAAAIB+hEII6Keygz49cM1c/cuUIfrRs5v1o2c3K5GwTscCAAAAAPQDFEJAPxbwunXLpTN15fxRuvuNXVp8X6kaQ1GnYwEAAAAA+jgKIaCfc7uM/vvzU/SDs47RK9uq9IXfL9M7hziBDAAAAADw4SiEgAHAGKMvzR+lB6+dp8ZQVF+4ZZn+uvGg07EAAAAAAH0UhRAwgMwZnaNnvnqCxhWm6/oH1ui+FbudjgQAAAAA6IO6VAgZYxYaY7YZY8qMMd85zO1fNsZsMMasM8a8YYyZ0v1RAXTF0MwUPXztPJ06sUD/9dQm/epv22Utm00DAAAAAP7pqIWQMcYt6RZJZ0iaIuniwxQ+D1prp1lrZ0i6SdKvuj0pgC5L8bl1x+WzdP6sYv32H+/oP/+8UbF4wulYAAAAAIA+wtOFMXMklVlrd0qSMeZhSWdL2vzuAGttY6fxQUlMRwAc5nG7dNN505Wb5tftr+7QrqoW/f6S45Sb5nc6GgAAAADAYV1ZMlYkqbzT5X3t172HMeYrxpgdSs4Q+trh7sgYs9gYU2qMKa2qqvo4eQF8BMYYfeeMSbr5vOlas7dOn//dG3q7vN7pWAAAAAAAh3XbptLW2lustWMlfVvS9z5kzBJrbYm1tiQ/P7+7HhrAUZxfMlxPXD9fxhidf/sK/fmt/U5HAgAAAAA4qCuF0H5JwztdLm6/7sM8LOkLnyQUgO43tShTz371BM0cmaVvPrpOT62jFAIAAACAwaorhdBqSeONMaONMT5JF0l6uvMAY8z4ThfPlPRO90UE0F2ygz4tvXKO5o7O0TcfWadn3j7gdCQAAAAAgAOOWghZa2OSbpT0gqQtkh611m4yxvzQGHNW+7AbjTGbjDHrJP2rpC/1WGIAn0iKz617rpytkpE5+sYj6/TchgqnIwEAAAAAepmx1pkDwUpKSmxpaakjjw1Aag7HdOU9q7Rmb51uOGWsvnH6BHnd3batGAAAAADAAcaYNdbakqON49UfMEil+T36w1VzdP6sYt3y8g6dd/sK7alpcToWAAAAAKAXUAgBg1jQ79FN5x2rWy6ZqV1Vzfrsb17XS1sPOR0LAAAAANDDKIQA6MzpQ/X8N07SmPw0XXffGv11I/sKAQAAAMBARiEEQJJUlJWiB66dq2lFmfrKg29xLD0AAAAADGAUQgA6ZAS8+uPVczVrZLa+8cg6Pb5mn9ORAAAAAAA9gEIIwHuk+T36w6I5mj82V9967G29uOmg05EAAAAAAN2MQgjAB6T43FpyeYmmFWfpqw+9pVW7ap2OBAAAAADoRhRCAA4r6Pdo6ZWzVZSdomv+sFpbDzY6HQkAAAAA0E0ohAB8qJygT3+8ao5SfG596Z5VKqtsdjoSAAAAAKAbUAgBOKLi7FT98aq5iieszrt9udbsYfkYAAAAAPR3FEIAjmrikHQ9cf0CZaV4dcmdb7LRNAAAAAD0cxRCALpkRG6qHr9+viYNzdCX71+jP67Y7XQkAAAAAMDHRCEEoMty0/x66Nq5OnVSgb7/1CZ9/6mNisUTTscCAAAAAHxEFEIAPpJUn0d3XF6ixSeN0R9X7NGie1eroS3qdCwAAAAAwEdAIQTgI3O7jL772cm66dzpWrmzRufcukz769ucjgUAAAAA6CIKIQAf2wWzh+u+q+eqsims825brrLKJqcjAQAAAAC6gEIIwCcyb0yuHll8vKJxq/NvX6G3y+udjgQAAAAAOAoKIQCf2JRhGXr8+uOVFvDo4jtXallZtdORAAAAAABHQCEEoFuMzA3qsS/P1/DsVC1aulrPb6hwOhIAAAAA4ENQCAHoNoUZAT163fGaVpyprzy4Vg+t2ut0JAAAAADAYVAIAehWmale3Xf1HJ00IV//8cQG/fT5LWoMcSw9AAAAAPQlFEIAul2qz6M7ryjRBSXFuuPVnTrx5y/r1lfK1BKOOR0NAAAAACAKIQA9xOt26abzjtUzN56gmSOydNNft+nkm1/Wn0rLlUhYp+MBAAAAwKBGIQSgR00rztTSRXP0+PXzNSInVd96bL0uuGOFNh9odDoaAAAAAAxaFEIAesWskdl67MvzddN507WzukWf+93r+tXftjNbCAAAAAAcQCEEoNe4XEYXlAzXy/92ir5wXJF++4939OX716iZvYUAAAAAoFdRCAHodZmpXv3y/GP135+fon9srdQ5ty7TnpoWp2MBAAAAwKBBIQTAEcYYLVowWn+8ao4ONYZ15m/f0J2v7VQklnA6GgAAAAAMeBRCABy1YFyenv3qCZozOkc/fm6LFv7mNb28rVLWsrcQAAAAAPQU49SLrpKSEltaWurIYwPom17aekg/enaLdlW3KD/drwVjc7VgXJ5Om1yonKDP6XgAAAAA0OcZY9ZYa0uONs7TG2EAoCtOnVSoE8bl66l1+/X6O9V6/Z1q/XndAaUHPPr2wkm6ZM4IuVzG6ZgAAAAA0O91aYaQMWahpN8X8cbVAAAgAElEQVRIcku6y1r7s/fd/q+SrpEUk1Ql6Spr7Z4j3SczhAAcTSJhtelAo376/BYt31GjGcOz9L0zJysr1atIzCqWSGhCYboCXrfTUQEAAACgT+jqDKGjFkLGGLek7ZI+LWmfpNWSLrbWbu405lOS3rTWthpjrpd0irX2wiPdL4UQgK6y1urP6/brf5/dopqWyHtuy0vz6eoTxuiyeSOUHvA6lBAAAAAA+obuXDI2R1KZtXZn+x0/LOlsSR2FkLX25U7jV0q67KPFBYAPZ4zRF48r1qcmFuiVbVUyRvJ7XIolrB5ZXa6f/3WrbnulTIsWjNbVJ45WBsUQAAAAABxRVwqhIknlnS7vkzT3COOvlvT84W4wxiyWtFiSRowY0cWIAJCUlerTF44res91n5s+TG+X1+uWl8v0m3+8oz+s2K3rTx6rK44fpRQfS8kAAAAA4HC69dh5Y8xlkkok3Xy42621S6y1Jdbakvz8/O58aACD2LHDs7TkihI9+9UTNGN4ln76/FaddPPL+vFfNmvNnjolEhxhDwAAAACddWWG0H5JwztdLm6/7j2MMadL+k9JJ1trw90TDwC6bmpRpu5dNEerdtXq9ld36N7lu3Xn67s0JCOgc2cV6eoTxnB8PQAAAACoa5tKe5TcVPo0JYug1ZIusdZu6jTmOEmPSVporX2nKw/MptIAelpDW1QvbT2kv6yv0D+2VirF69YVx4/StSeOVm6a3+l4AAAAANDtuu2UsfY7+6ykXyt57Pw91tofG2N+KKnUWvu0MebvkqZJqmj/kL3W2rOOdJ8UQgB60zuHmvS7l8r0zPoD8riMji3O0vFjc3X8mFzNHp0jr7tbV9ACAAAAgCO6tRDqCRRCAJxQVtmsx9bs04qdNdqwr14JK+UEfTrr2GE6d2axphZlyBjjdEwAAAAA+FgohADgKJpCUS3fUaOn1x3Q3zYfUiSe0IzhWfrZudM0aUiG0/EAAAAA4COjEAKAj6ChNaqn1x/Qr/+2XY2hqL526nh9+ZSxLCUDAAAA0K9QCAHAx1DbEtH3n9qoZ9dXaNKQdH1qUoHG5AU1Jj9NxwzLUMDrdjoiAAAAAHyorhZCXTl2HgAGjZygT7+/ZKbOnFahX//9Hd352k7FEsnivDDDr6+dNl4XlAxn5hAAAACAfo0ZQgBwBNF4QuW1rdp2sEl3vbFLa/bUaVRuqr5++ngtPGaoUnzMGAIAAADQd7BkDAC6mbVWL22t1M0vbNPWg01K9bl1+uRCfW76UJ08MV9+D+UQAAAAAGexZAwAupkxRqdNLtSnJhZo5c4aPbO+Qn/dWKGn3z6gdL9Hnz6mUJ+fPkwLxuXJ52FJGQAAAIC+ixlCAPAJROMJLd9Ro2ffPqAXNh1UYyimIRkBfe208Tq/pJi9hgAAAAD0KpaMAUAvi8QSem17lW59pUxr99ZrdF5Q3zh9vM6cNlQeiiEAAAAAvYBCCAAcYq3VP7ZU6hcvJvcayk/369yZxbpw9nCNzgs6HQ8AAADAAEYhBAAOSySs/r7lkB4tLdfL26oUT1iNyQtq8tAMTRySrgmF6RqTH9SInFQFvGxIDQAAAOCTY1NpAHCYy2X0mWOG6DPHDFFlY0h/Xrdfa/fUa9OBBv1lQ0XHOGOkYZkpOmFcnj47fajmj81l7yEAAAAAPYpCCAB6QUFGQItPGttxuSUc046qZu2qbtGemlZtO9Skv2yo0COl5cpK9WrOqBwNy0pRYUZAI3JSdfLEfKX5+ZUNAAAAoHvw6gIAHBD0ezS9OEvTi7M6rgtF43r9nWr9Zf0BbTzQqBU7atQUjkmSUn1unTF1qC4oKdac0TkyxjgVHQAAAMAAQCEEAH1EwOvWp6cU6tNTCjuuaw7HtLWiUY+v3adn3q7Q42v3KS/NrxPH5+nE8XmaPSpHQzIDLDEDAAAA8JGwqTQA9BOtkZhe3HRIL2+r1BvvVKumJdJxW16aT0MzU3TerGJdPGeEfB4KIgAAAGAw4pQxABjAEgmrzRWN2rC/QYcaQzrUGNaWikatK69XcXaKvnn6BH3huCK5XSwtAwAAAAYTCiEAGGSstXrtnWrd/MJWbdzfKL/Hpbw0v/LS/SpM92taUaaOG5Gt6cMzlRHwOh0XAAAAQA/g2HkAGGSMMTp5Qr5OGp+nFzYd0tq9dapuCqu6JaKyqma9uPlQ+zhpVG5Qk4aka+KQdB0zLFOzRmYrJ+hz+DMAAAAA0FsohABggDHGaOHUIVo4dch7rm9oi2r9vnq9tbdemw80auvBJv1100G9O1F0XEGaZo/KVsnIHM0elaPhOSmcZgYAAAAMUCwZA4BBrDUS08b9jSrdU6vS3XUq3V2rxlDyqPvCjOQys7H5aRqTH9SY/DSNyQsqJ+ijKAIAAAD6KJaMAQCOKtXn0ZzROZozOkdScrPq7ZVNWt1eDm2taNJr26sViSc6PiYzxaux+UEdNyJbJ4zL05zROQr6+esEAAAA6E+YIQQAOKJ4wmp/XZt2VDdrZ1WLdlQ1q+xQs9btq1cklpDXbTR5aIaKslI0LCtFw7NTdPqUQhVnpzodHQAAABh0OGUMANCjQtG4SnfX6Y2yam060KAD9W06UB9SWzQuY6QTxuXpwtnDdfrkQgW8bqfjAgAAAIMChRAAoNdZa1Ve26bH1+7TY2v2aX99m9wuo3H5aTpmWIamDMvQ1KJMTRmWoYyA1+m4AAAAwIBDIQQAcFQ8YbV8R7Xe3FmrTQcatOlAoyqbwh23j8hJ1bCsgHKDfmUHvSpMD2hkXlBj8oIalRdUGvsSAQAAAB8Zm0oDABzldhmdOD5fJ47P77iusimkTQcatflAozZXNKqqMaytBxtV1xpVbUukY5wxUsnIbC2cOlT/cgz7EQEAAADdjRlCAIA+oS0S157aFu2ubtHmA416cfMhbT3YJEnKS/MpPeBVRsCj/PSAThyfp1MnFWh4DkURAAAA0BlLxgAA/d6u6ha9uOmg9tS2qikUU2NbVLtrWrSnplWSNCYvqNw0n979q2xoVoq+MGOYTpqQL6/b1XE/8YSVy0jGGCc+DQAAAKDXsGQMANDvjc4L6rqTx37g+l3VLXplW6WWlVWrJZw81UySlpVV65m3Dyg36NOpkwpU1xrVzqpm7a1t1bCsFJ09Y5jOnjFM4wrSe/kzAQAAAPqWLs0QMsYslPQbSW5Jd1lrf/a+20+S9GtJ0yVdZK197Gj3yQwhAEB3i8YTenVblZ54a5/eeKdaQzIDGpOXppF5qdp8oFHLyqqVsNKwzIASVmqLxhWOxZWZ4lVBekAF6X65XUb1rVHVtUYUjSd04vh8fXbaUM0ZnSO3ixlGAAAA6Nu6bcmYMcYtabukT0vaJ2m1pIuttZs7jRklKUPSv0t6mkIIANAXVTaG9Oz6Cq3fVy+fx6UUr1tet0sNbVFVNoVV2RRWImGVlepVdqpPsYTVG2VVCkUTykvza3xBmlJ8bgW8LgU8bgV8bgU8bqX4XMpK8Skn6FNOmk/Ds1M1Ji8o1/sKJJauAQAAoKd155KxOZLKrLU72+/4YUlnS+oohKy1u9tvS3ystAAA9IKCjICuOmH0R/qY1khML2+t0l83HdTBhjZVNkUViiYUisbb3xJqi8YVT7z3P1iCPreOGZapUXmpqmgIaW9tq/bVtUmSMlO8ykzxamhmQCWjcjRnVI5mjsxSqo+V3AAAAOgdXfmXZ5Gk8k6X90ma2zNxAADoW1J9Hp05fajOnD70Q8dYa9UUjqm2OaKalrB2VrVo4/4GbdjfoJe2VmlYVkDTijJ15rShMkZqaIuqvjW5QfbvX3pHCSsZI3ndLnlcRm6X6fjT7TIKeN0qykrR8OxUDc9JUdDvkdtlkjONrFU4llAknpC10qyR2SoZmS1Pp021380o/XN2UjSeLLJC0bj8brcyU7099yQCAACgz+nV/4o0xiyWtFiSRowY0ZsPDQBAjzHGKCPgVUbAq1F5Qc0amaPzS4Z36WObQlGt3VuvdXvr22caJRRLWMUTVrGEVSJh1RKJa39dq/6xtVLVzeGj3mdGwKOTJxYoM8WjnVUt2lnVooONoU55pfevGM8IeDQiN1XFWakKeF1ytZdSqT6PMgIeZaQkl9ENyQyoMCOgwgy/XMYoEksoHEtOEA763Qr6PB9YKnc48YRVRUObfB6XslN97zkVDgAAAD2vK4XQfkmd/1Vb3H7dR2atXSJpiZTcQ+jj3AcAAANJesCrkyfk6+QJ+V0aH4rG1RaJK2Gt4u2tjt/tls/jUjSR0PKyav1jS6Ve3lapcCyhMflpmj82V8XZKTLGyCo5W8jndingTe6DFIrEtbe2VXtrW1VW1axoPKF4eynVGomrMRT9QIF0JCletzwuI2Mkl8soPeBRfppf+el+edwu7ahs1q7qlo4iSZLS/R6l+NxKrrxL5jt1coG+MKNIs0ZmKxxL6OWtlXr67QPaXdOqGcOzNGd0tmaNyHnPrKuA16Ux+WnKCfoOm81aq/LaNpXXtSozxausVK9ygj6W6wEAgEGnK5tKe5TcVPo0JYug1ZIusdZuOszYeyU9y6bSAAA4z1rbLRtYJxLJJXF1LREdbAzpYENIh9pnHPk8Lvk9biWsVVskruZwTG3RuGJxq4RNvjW2RVXVHFZVUzhZUuUFNb4wXaPzgorFE6prP9WtLRKXMUYuI9W2RPTytkqFogkVZaWooS2q5nBMeWk+TRqSobf31aspFPvQzNmpydlaQzMDKkgPKC/Np51VLVq5s0YHGkIfGF+UlaKpRRmaOixTqX6PDja0qaIhpKZQTCNyUjUmP6jReUF53S61RuJqjcSUsFapPo/S/B4F/R6l+d1K9SXfjyes6lsjqm+LqjUcV4rPpVSfR6k+t0LRhJpC0WR+Iw3JCGhoZkCZKclle9G4VTgWV6B903MAAICPottOGWu/s88qeay8W9I91tofG2N+KKnUWvu0MWa2pCclZUsKSTporT3mSPdJIQQAAI6kORzTi5sO6rkNFcoJ+nTWsUWaNyZHHrdL8YTVtoNNentfvTwu07FRd2s0rh2VzdpZ3aLd1S061BhSZVNYTaGYcoM+zRuTq3ljcjS2IE2NbTHVt0ZU3RzWloNN2nygUbuqWyRJfo9LQzIDSvN7tLe29YjlU3fxuV2KJRLqvD95btCngvYleoXpARVkJGdatUXiqmlJZg9F4/K6XR2zvvLT/RqSGdCQjIDCsYT217XqQENIdS0Redwu+dxGHrdLoWiywGsOxeRxGw3LSlFRVoqGZqYoxeeSr33mWaJ9n6pQe9HndkkuY+RxGwV9HqUFPMoIeBVonxnmdhsZqaMgbO34M6aWcHID9swUrzLaZ2gNzQy8Z4ZWLJ5QeV2bDjaElOb3dHxtM1I8Hyg4rbWqbYkolrAds9hSvO7DjgWAnvT8hgo9tLpc/3XmZI0vTP/A7d31nzRAV3RrIdQTKIQAAEBvCUXj8ntcR/3HeHM4pmgsoaxUb8dYa62qmyPaXdOiRCI5KyjV75bLGLWEY8m3SEzN4XjHZbfLKDvVp8xUr4I+j0LReEch4ve6lB7wKiPgUcJKhxpDqmgIqaopLI/LKOB1yedJzkQ61BhWZXupdagxpOrmcEdh5Pe4lJfmV8DrUjRuFYklNwpvaIt+4PPyeVzKDfoUS1hF4wlFYwml+NztM5s8isQS2l/fptZIvNuf+67ITvVqWFaKovGEdle3KhL/4MG1KV63huckN1cPeN3aVd2iPTUtajlM5nef/9ygTzmd3lL9bgU8bgW8brmMOvbAisQTCkfj7X8mlJHi1fCcVI3MSVVOmi9ZHDZFVNsaUarPrdygX3lpyWWJ5XVtKq9tVVVzWMXZKRpfkK5xBWmKJxLaU9OqPTWtqmkJdzxuwOtSwiY3do8lrNzGJL8OgeQss3dLtjS/p2OWmNed3ES+LZJcMtrW/v30z/f/eX00nlBumk8F6QEVpPtVkB54T0GWSFhVNYdV0RBSitet7FSvMlO98nvcn+hrWNMcVmVTuOM5TVjbsSwzK8WnaCKhplBMjW1RJaxVbtCvnKBPPk9yFpy1VpF4Ql6Xq0v7kHWneCL58xPwHv13RFfF4ol/bv6PASWesHJ3+h5NJKz+7+/b9buXymSMlOb36NZLZ+rE8cml4Bv2Neg/nlyvXVUtuvz4UbrmxNHKS/Mf9THWlddp5c5aTRmWoVMm5PO9hI+EQggAAGCAeXeJXYrPraDPfdgXCKFoXJWNYR1sDMnvcakoO0W5Qd9RX0xYa1XfGtXBxpDCsX8WJG6Xkd/jlt/jkrd9dlbCJoul1khcTaGoGkMxhaPxjs3QrZVS28uNVJ9baX6PUtuX1Rlj1NgW7dj36UBDm/bXtWl/fZs8LpfGFgQ1Nj9NRVkpagnH1NA+9kB9SHtrW7WvrlWhaFyj8oIalRvUyNzU95QZrZGYalsiqmuNqKY5otqW5FtNS0Rt0bgisfeWTcYkyzW/Jzkjyud2dSxR7CpjpMwUr+pbP1jGSZLL6D0zv3qbz+NSQbpfPrdL++vb3rN/17s6ZzRGSvX+szD0eVwds8ICHrdy03zKTfMpPeBVWWWzNu5vUMVhlmJ2RdDnVixhOzJ5XEYF7bPcMlK8qmuJqLr96+jzuJTePiMtI8XTXqx6lR5IZvS6jTwulyLxhBrbkt+XbZGYjDFym+S+Zu9+zzaFYmoKxZKz5Nq/1gGvS4UZ7UVaxj8LtYDXpf11bdpX16aKhjal+NzKTfMrP80vv8eltva93ZrDMVU0hFTR0KbKprAyU7yaMjRDk4dmaGhmILl0tjGsmpZIcm+19OSMv0TC6kBDSBX1yY9rjSTL5dZIXDlBn8bmBzUmP01DMgJK2OTPmCTlBH0dswHjCat9dW3aV9eqhraoirJSNDI3qBG5qYrEEjrUmFzq23m2ozFS0OdRensBmd7+XGYEvPJ7XR0/p42hqPwet7JSvR2HADSHY2oKJX9OjIxc7TMHXUbtS3+T77van/d3f+7f/VlMJGxHUZse8KiuNarKxpCqmsPyuV3thxcElJXqTX7O8X/ua9fcXrzHErbj6/7u7yefxyWPy6WDDSFtP9Sk7YeaFI4ldNrkAn1qYoGC/vfORrTSe5bmhmNxba1o0sYDDQpFE+3LopPPxcb9DVq/v0G7qls0aUiGTp2UrxPH5+uu13fp71sO6YKSYl1/yjhdf/8avVPZrP86c7L21bXpnmW7lJvm18wRWXpx8yH5PS5dNHuEjhuRpWFZKRqamZzNuauqRbuqW7TxQINe3V71nt8n4wvSdM2Jo3Xa5EI1tEXbf7eFVd2c/D1X0xKW22WU1/59mZHiVbz9PwAi8YSCPo+yg8mvnyQdbAzpUENIda1RFaT7VZydouKcVGUEPPK4XB2nnR6pnE0krJpCMbVGY8pO9SngTf4efnfm5o6qFtW2RFSQ4VdhRkD5af6OAhg9j0IIAAAAeJ9EIjkTJd7+YtJzmFkc1lrVtUa1pyb5giYn6FNeWnJGS2skrpqWsKqbIrKyGp6dqmFZKfJ5XGoKRVVW2ayyymZ53S6NyG2fZdQ+OysUjSsUTchlJE/7zJ94wqolHFdzONoxy+zdoiIcSy7Ti8aTM25SvG6ltJdsyffd73vfI7fLdMzWOdSYnHlW2ZScaRaOJVScnaIROakampmicCyhutaI6lsjCscSMpJkjKxNZmoJx9QciSnWaaP5lkg8+aK+OayGtqhG5QU1rShT04oyVZSVIr83udyw82bv9W0ReV2ujhLHZaSalohqm5P7bHlcyRf0Po9LLZG4DjWEdLC9vMgO+pQX9Ck76FMsnlBjKFlENLbF1Nhe7DSGoskZV/FkWeJ1t5/8mOJVitct2/51T+775e4oPjqXIAGvW7Ut4eSsvKaQKhuTz1vnsqg4O1VDMwMKReOqaY6oqjk5I6rz12BoZvIF/pDMgKqbI9pc0aitFY0KxxLyuV3KT09+HzWHY6puCqup/f5zgr72Pc/8Sg94FfQnZ5RVNYW1o6pFu6qbFYp+sMjDkWWneuUyRjUtEfk9Li0Yl6doPKG9ta3aX9emWMIqI+BRXntZsaOqWdH44V8fD8kIaNr/b+9eY+QuqziO/87uXHa6u91tu0shWxQoJVi8VIIK3oKIWvBFNSEGTJQYEnwBURNfiEajJr7QF0pioiQSCXgDCUokBi8EqxgjV4tyacBWqLR02d12b8N2ZudyfPE8O0y3Oy2t2/3P9P/9JM3M/9Ltaff0dPf0PM9/w4DOGerVjpem9MSeyca00Nc++iZd++6zZGaaLVX0uTt3aPtz45KkT77rDfrS1vM1UMhq93hRt/x5t+7dsU+1Fl3i4f683rdpSJedf5ouOWed/vL8uG796wvauX+m5e9zIDaAjqeR/XqYSdmmBlF3d3idr9Y1W64e9sCJ/nxGQ/35WFOWbo4PxenF9avzGlyVUy428bpivTg4V9H0XFjevLBceOHvZ0+mS5nuLo3NlrRv8pBeniqpXK2pkMuoN9etfLZL5UpdpWpN5Updw/15bRzu07mn9WlNb04Hi+XGUuuJYng9UAxTnxuH+3TOcPjPiMs3r1dfvvMfNEFDCAAAAMBJ0457otTr3phIWQ6vxo3yX8+UXSvVWl3FclUDhewRHyNspq/GdEUr9bqrOF8N35THqY2J4rxGp8NG/91d0oY1qzQyWFB/T0b7p0vac2BOL03OKZ9ZmLjJa3VPVoohuKvRgJyNjbbZeFyu1rS657VvyEuVuqYPzWtyrqJKta7+nkxsXGXix3LVXY2HCXjjveKxq78nPNVxXW9OXV0WG4vzmi1VtLY315iYqlRdr8yGiaapuUpjUiXTZVrVtNR1YRpsPi77rMTX+WpoBpy3vl9DfTnVXXr8xYP63dOjeuj5cfX3ZMKS0DhdeDA2CQ7N13Te6f1668iA3jwyoNU9WZVrYaownwn7szWbnqvob7sndOaaVXrLhoEjPuc/e3iPLhgZ0DvOWnvE5/PVclUvTx1qTIblMl06eyg8vGBw1ZFPyXR3/X33Ae0cndW63jCht7B0dU1vrjHldGi+poliWTOlSlxuGprexXK10ahxl04fyMcJrJzGZ8vaOzmnvZOHGpNX1bikdWHqc/Fxpss0GPeCK+S6NTVX0fhsWRPFslYXsto43KeNw71a15vXRDE0p0dnSo1l0KMzJc2UKuFzVw175w0UsnEJa2j+LkyHzpaqKlVqjSnCgUJWI4MFjawpqJDtbjzoISwND43ZXHeXRmdK2j1WbDRdpTAJuTb+ub3W5K9q11hR/z04F3Llq5cfc0lfJ6AhBAAAAAAAOp67q1Lz41p25u4anw3TjOv68hosZFsugytXa9pzYE6bTutru0b3iXi9DaHOn4UCAAAAAACnLDNTLnN8jRozC/uBre455r35TLfOW+LpcKc6dnUCAAAAAABIGRpCAAAAAAAAKUNDCAAAAAAAIGVoCAEAAAAAAKQMDSEAAAAAAICUoSEEAAAAAACQMjSEAAAAAAAAUoaGEAAAAAAAQMrQEAIAAAAAAEgZGkIAAAAAAAApY+6ezC9sNi5pTyK/+PIbkjSRdBDAUZCj6ATkKToBeYp2R46iE5Cn6ASdnKdvdPfhY92UWEPoVGJmj7v7RUnHAbRCjqITkKfoBOQp2h05ik5AnqITpCFPWTIGAAAAAACQMjSEAAAAAAAAUoaG0PL4UdIBAMdAjqITkKfoBOQp2h05ik5AnqITnPJ5yh5CAAAAAAAAKcOEEAAAAAAAQMrQEAIAAAAAAEgZGkL/BzPbambPmdkuM7sp6XiABWb2opk9ZWZPmtnj8dxaM3vAzP4dX9ckHSfSxcxuM7MxM3u66dySeWnB92N9/ZeZXZhc5EiLFjn6DTPbF+vpk2Z2ZdO1L8ccfc7MPpJM1EgbMzvTzLab2bNm9oyZfT6ep56iLRwlR6mnaBtm1mNmj5rZP2OefjOeP9vMHon5+Eszy8Xz+Xi8K14/K8n4lwsNoRNkZt2SfiDpCkmbJV1jZpuTjQo4zAfcfYu7XxSPb5L0oLtvkvRgPAZW0u2Sti461yovr5C0Kf64XtItKxQj0u12HZmjknRzrKdb3P1+SYr/5l8t6YL4c34YvzYATraqpC+6+2ZJF0u6IeYj9RTtolWOStRTtI+ypMvc/W2StkjaamYXS/qOQp6eK2lS0nXx/uskTcbzN8f7Oh4NoRP3Tkm73P0/7j4v6S5J2xKOCTiabZLuiO/vkPSxBGNBCrn7Q5IOLjrdKi+3SfqJBw9LGjSzM1YmUqRVixxtZZuku9y97O4vSNql8LUBcFK5+353/0d8Pytpp6QRUU/RJo6So61QT7HiYk0sxsNs/OGSLpN0Tzy/uJYu1Nh7JH3QzGyFwj1paAiduBFJLzUd79XRCx2wklzSH83sCTO7Pp5b7+774/tRSeuTCQ04TKu8pMaindwYl9rc1rTclhxF4uKShbdLekTUU7ShRTkqUU/RRsys28yelDQm6QFJuyVNuXs13tKci408jdenJa1b2YiXHw0h4NT0Xne/UGFM/AYze3/zRXd3haYR0DbIS7SpWyRtVBgn3y/pu8mGAwRm1ifpV5K+4O4zzdeop2gHS+Qo9RRtxd1r7r5F0gaFqbTzEw5pxdEQOnH7JJ3ZdLwhngMS5+774uuYpHsVCtwrCyPi8XUsuQiBhlZ5SY1FW3D3V+IXjHVJt+q1ZQzkKBJjZlmFb7R/7u6/jqepp2gbS+Uo9RTtyt2nJG2XdInCstpMvNSci408jdcHJB1Y4VCXHQ2hE/eYpE1xF/KcwkZo9yUcEyAz6zWz/oX3kj4s6WmF/Lw23natpN8kEyFwmFZ5eZ+kT8en41wsabppKQSwYhbttfJxhXoqhRy9Oj515GyFDXsfXen4kD5xz4ofS9rp7t9rukQ9RVtolaPUU7QTMxs2s8H4viDpQwr7XW2XdFW8bTT76wAAAAEdSURBVHEtXaixV0n6U5zG7GiZY9+Cpbh71cxulPQHSd2SbnP3ZxIOC5DCngH3xj3OMpJ+4e6/N7PHJN1tZtdJ2iPpEwnGiBQyszslXSppyMz2Svq6pG9r6by8X9KVChtLzkn6zIoHjNRpkaOXmtkWheU3L0r6rCS5+zNmdrekZxWeqHODu9eSiBup8x5Jn5L0VNz7QpK+Iuop2kerHL2Geoo2coakO+IT7bok3e3uvzWzZyXdZWbfkrRDobmp+PpTM9ul8ACKq5MIernZKdDUAgAAAAAAwHFgyRgAAAAAAEDK0BACAAAAAABIGRpCAAAAAAAAKUNDCAAAAAAAIGVoCAEAAAAAAKQMDSEAAAAAAICUoSEEAAAAAACQMv8DrXKb2fnaM+cAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 1440x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(20,5))\n",
    "plt.plot(losses)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.921"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test()"
   ]
  }
 ],
 "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": 2
}