02_DNN.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-03-15T08:32:17.030152Z",
     "start_time": "2020-03-15T08:32:16.041272Z"
    }
   },
   "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 torch.utils.data as data_utils \n",
    "\n",
    "\n",
    "from torchvision import transforms\n",
    "\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from matplotlib import pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## custom Batch Generator"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 1. load trn, val data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-03-15T08:32:18.754118Z",
     "start_time": "2020-03-15T08:32:18.740945Z"
    }
   },
   "outputs": [],
   "source": [
    "trn = pd.read_csv(\"./dnn_data/trn.tsv\", sep='\\t')\n",
    "val = pd.read_csv(\"./dnn_data/val.tsv\", sep='\\t')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-03-15T08:32:19.572010Z",
     "start_time": "2020-03-15T08:32:19.554613Z"
    }
   },
   "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>feature_1</th>\n",
       "      <th>feature_2</th>\n",
       "      <th>feature_3</th>\n",
       "      <th>feature_4</th>\n",
       "      <th>feature_5</th>\n",
       "      <th>feature_6</th>\n",
       "      <th>feature_7</th>\n",
       "      <th>feature_8</th>\n",
       "      <th>y</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.770747</td>\n",
       "      <td>1.777481</td>\n",
       "      <td>1.119744</td>\n",
       "      <td>10</td>\n",
       "      <td>0.904246</td>\n",
       "      <td>0.750227</td>\n",
       "      <td>6</td>\n",
       "      <td>7</td>\n",
       "      <td>74.559605</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.517279</td>\n",
       "      <td>2.102868</td>\n",
       "      <td>1.433156</td>\n",
       "      <td>10</td>\n",
       "      <td>0.835582</td>\n",
       "      <td>0.982529</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "      <td>68.654484</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.528888</td>\n",
       "      <td>2.287112</td>\n",
       "      <td>1.181075</td>\n",
       "      <td>1</td>\n",
       "      <td>0.944968</td>\n",
       "      <td>0.961392</td>\n",
       "      <td>7</td>\n",
       "      <td>2</td>\n",
       "      <td>81.007162</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.398252</td>\n",
       "      <td>1.285644</td>\n",
       "      <td>0.968844</td>\n",
       "      <td>1</td>\n",
       "      <td>0.946535</td>\n",
       "      <td>0.748881</td>\n",
       "      <td>7</td>\n",
       "      <td>7</td>\n",
       "      <td>82.157630</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.483614</td>\n",
       "      <td>2.503184</td>\n",
       "      <td>1.563709</td>\n",
       "      <td>10</td>\n",
       "      <td>0.936454</td>\n",
       "      <td>0.945359</td>\n",
       "      <td>7</td>\n",
       "      <td>10</td>\n",
       "      <td>86.360000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   feature_1  feature_2  feature_3  feature_4  feature_5  feature_6  \\\n",
       "0   0.770747   1.777481   1.119744         10   0.904246   0.750227   \n",
       "1   0.517279   2.102868   1.433156         10   0.835582   0.982529   \n",
       "2   0.528888   2.287112   1.181075          1   0.944968   0.961392   \n",
       "3   0.398252   1.285644   0.968844          1   0.946535   0.748881   \n",
       "4   0.483614   2.503184   1.563709         10   0.936454   0.945359   \n",
       "\n",
       "   feature_7  feature_8          y  \n",
       "0          6          7  74.559605  \n",
       "1          4          1  68.654484  \n",
       "2          7          2  81.007162  \n",
       "3          7          7  82.157630  \n",
       "4          7         10  86.360000  "
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "trn.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-03-15T08:32:20.563290Z",
     "start_time": "2020-03-15T08:32:20.560593Z"
    }
   },
   "outputs": [],
   "source": [
    "X_features = [\"feature_1\", \"feature_2\", \"feature_3\", \"feature_4\", \"feature_5\", \"feature_6\", \"feature_7\", \"feature_8\"]\n",
    "y_feature = [\"y\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-03-15T08:32:21.140391Z",
     "start_time": "2020-03-15T08:32:21.132660Z"
    }
   },
   "outputs": [],
   "source": [
    "trn_X_pd, trn_y_pd = trn[X_features], trn[y_feature]\n",
    "val_X_pd, val_y_pd = val[X_features], val[y_feature]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-03-15T08:34:27.150152Z",
     "start_time": "2020-03-15T08:34:27.144450Z"
    }
   },
   "outputs": [],
   "source": [
    "trn_X = torch.from_numpy(trn_X_pd.astype(float).values)\n",
    "trn_y = torch.from_numpy(trn_y_pd.astype(float).values)\n",
    "\n",
    "val_X = torch.from_numpy(val_X_pd.astype(float).values)\n",
    "val_y = torch.from_numpy(val_y_pd.astype(float).values)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-03-15T08:32:21.814597Z",
     "start_time": "2020-03-15T08:32:21.812374Z"
    }
   },
   "outputs": [],
   "source": [
    "batch_size=64"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-03-15T08:34:29.852332Z",
     "start_time": "2020-03-15T08:34:29.848521Z"
    }
   },
   "outputs": [],
   "source": [
    "trn = data_utils.TensorDataset(trn_X, trn_y)\n",
    "trn_loader = data_utils.DataLoader(trn, batch_size=batch_size, shuffle=True)\n",
    "\n",
    "val = data_utils.TensorDataset(val_X, val_y)\n",
    "val_loader = data_utils.DataLoader(val, batch_size=batch_size, shuffle=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-03-15T08:34:30.033160Z",
     "start_time": "2020-03-15T08:34:30.029691Z"
    }
   },
   "outputs": [],
   "source": [
    "tmp = next(iter(trn_loader))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-03-15T08:34:30.369292Z",
     "start_time": "2020-03-15T08:34:30.357785Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[tensor([[5.8083e-01, 2.3504e+00, 7.2618e-01, 1.0000e+00, 1.0000e+00, 9.8835e-01,\n",
       "          5.0000e+00, 8.0000e+00],\n",
       "         [6.8490e-01, 2.6601e+00, 7.0480e-01, 1.0000e+00, 7.8875e-01, 9.5704e-01,\n",
       "          7.0000e+00, 9.0000e+00],\n",
       "         [5.9617e-01, 2.1074e+00, 6.8805e-01, 1.0000e+01, 6.5303e-01, 9.5500e-01,\n",
       "          4.0000e+00, 2.0000e+00],\n",
       "         [4.6618e-01, 2.4170e+00, 9.9786e-01, 1.0000e+00, 6.2376e-01, 9.7757e-01,\n",
       "          6.0000e+00, 2.0000e+00],\n",
       "         [5.0665e-01, 1.7576e+00, 9.9688e-01, 1.0000e+00, 1.0000e+00, 8.4077e-01,\n",
       "          7.0000e+00, 9.0000e+00],\n",
       "         [4.9069e-01, 2.5487e+00, 2.1666e+00, 1.0000e+00, 9.9189e-01, 9.9980e-01,\n",
       "          6.0000e+00, 1.0000e+01],\n",
       "         [7.1390e-01, 2.5845e+00, 1.1495e+00, 1.0000e+00, 8.5633e-01, 9.9519e-01,\n",
       "          6.0000e+00, 8.0000e+00],\n",
       "         [6.7145e-01, 2.5900e+00, 1.6690e+00, 1.0000e+00, 1.0000e+00, 9.6371e-01,\n",
       "          6.0000e+00, 1.0000e+00],\n",
       "         [4.7936e-01, 2.1349e+00, 1.8340e-01, 1.0000e+01, 6.0059e-01, 9.6727e-01,\n",
       "          6.0000e+00, 1.2000e+01],\n",
       "         [6.9422e-01, 2.0190e+00, 1.1865e+00, 1.0000e+01, 9.0047e-01, 7.5412e-01,\n",
       "          7.0000e+00, 3.0000e+00],\n",
       "         [6.5434e-01, 2.4105e+00, 1.8607e-01, 1.0000e+01, 7.8434e-01, 7.8652e-01,\n",
       "          7.0000e+00, 1.7000e+01],\n",
       "         [6.1322e-01, 1.8914e+00, 2.1471e+00, 1.0000e+01, 9.5641e-01, 9.9756e-01,\n",
       "          4.0000e+00, 2.0000e+00],\n",
       "         [5.3491e-01, 1.9500e+00, 2.3280e+00, 1.0000e+01, 9.2731e-01, 7.6218e-01,\n",
       "          6.0000e+00, 1.1000e+01],\n",
       "         [4.9404e-01, 2.2730e+00, 1.2880e+00, 1.0000e+00, 9.1440e-01, 7.9941e-01,\n",
       "          7.0000e+00, 1.1000e+01],\n",
       "         [7.6481e-01, 2.1519e+00, 5.2053e-01, 1.0000e+00, 6.9189e-01, 9.7789e-01,\n",
       "          5.0000e+00, 4.0000e+00],\n",
       "         [1.7387e-01, 1.0715e+00, 1.3056e+00, 1.0000e+00, 9.5078e-01, 9.5962e-01,\n",
       "          5.0000e+00, 1.0000e+00],\n",
       "         [7.4908e-01, 2.2742e+00, 8.3655e-02, 1.0000e+01, 5.5100e-01, 9.9737e-01,\n",
       "          5.0000e+00, 3.0000e+00],\n",
       "         [5.3266e-01, 1.7257e+00, 4.5679e+00, 1.0000e+01, 1.0000e+00, 8.4669e-01,\n",
       "          5.0000e+00, 1.0000e+00],\n",
       "         [5.2159e-01, 2.5074e+00, 2.6114e+00, 1.0000e+00, 7.3089e-01, 9.5588e-01,\n",
       "          7.0000e+00, 1.0000e+00],\n",
       "         [3.8983e-01, 1.9615e+00, 1.8327e-02, 1.0000e+00, 6.0940e-01, 9.7037e-01,\n",
       "          6.0000e+00, 9.0000e+00],\n",
       "         [5.4257e-01, 2.8584e+00, 1.1436e+00, 1.0000e+01, 8.2703e-01, 8.7309e-01,\n",
       "          8.0000e+00, 1.5000e+01],\n",
       "         [5.5000e-01, 1.5558e+00, 1.2287e-01, 1.0000e+01, 5.3869e-01, 9.5936e-01,\n",
       "          3.0000e+00, 1.2000e+01],\n",
       "         [6.3670e-01, 2.0816e+00, 6.4883e-01, 1.0000e+00, 7.7033e-01, 8.1180e-01,\n",
       "          8.0000e+00, 5.0000e+00],\n",
       "         [5.6567e-01, 2.7780e+00, 1.3717e+00, 1.0000e+00, 9.4450e-01, 9.6558e-01,\n",
       "          6.0000e+00, 8.0000e+00],\n",
       "         [6.3295e-01, 1.7330e+00, 9.7164e-02, 1.0000e+00, 6.7353e-01, 9.6534e-01,\n",
       "          5.0000e+00, 1.3000e+01],\n",
       "         [5.1308e-01, 2.1519e+00, 1.5189e+00, 1.0000e+00, 9.2906e-01, 8.1393e-01,\n",
       "          7.0000e+00, 1.6000e+01],\n",
       "         [3.5981e-01, 1.4648e+00, 1.2028e+00, 1.0000e+00, 9.5513e-01, 6.7593e-01,\n",
       "          7.0000e+00, 1.1000e+01],\n",
       "         [6.3103e-01, 1.8882e+00, 1.8230e-01, 1.0000e+01, 7.6466e-01, 9.2701e-01,\n",
       "          6.0000e+00, 5.0000e+00],\n",
       "         [5.2467e-01, 2.3464e+00, 1.2002e+00, 1.0000e+00, 1.0000e+00, 9.8206e-01,\n",
       "          5.0000e+00, 8.0000e+00],\n",
       "         [4.5827e-01, 1.3232e+00, 1.3976e+00, 1.0000e+01, 4.4027e-01, 7.6332e-01,\n",
       "          6.0000e+00, 2.0000e+00],\n",
       "         [5.8893e-01, 2.2464e+00, 2.8408e+00, 1.0000e+01, 8.2303e-01, 9.6377e-01,\n",
       "          6.0000e+00, 2.0000e+00],\n",
       "         [7.0140e-01, 3.0751e+00, 1.0140e+00, 1.0000e+00, 1.0000e+00, 1.0000e+00,\n",
       "          6.0000e+00, 1.0000e+00],\n",
       "         [5.5109e-01, 2.0368e+00, 2.2546e+00, 1.0000e+01, 9.5670e-01, 9.1136e-01,\n",
       "          6.0000e+00, 1.0000e+00],\n",
       "         [6.0515e-01, 3.0312e+00, 5.6021e-02, 1.0000e+00, 5.8575e-01, 9.0167e-01,\n",
       "          8.0000e+00, 1.6000e+01],\n",
       "         [6.5826e-01, 2.0542e+00, 1.2918e+00, 1.0000e+01, 9.7427e-01, 9.6317e-01,\n",
       "          6.0000e+00, 3.0000e+00],\n",
       "         [5.5450e-01, 2.8933e+00, 3.1785e-01, 1.0000e+00, 5.4126e-01, 8.9462e-01,\n",
       "          8.0000e+00, 7.0000e+00],\n",
       "         [5.7654e-01, 9.6501e-01, 9.5407e-01, 1.0000e+01, 9.8278e-01, 6.1954e-01,\n",
       "          7.0000e+00, 3.0000e+00],\n",
       "         [5.8548e-01, 2.3874e+00, 0.0000e+00, 1.0000e+00, 7.1169e-01, 8.9495e-01,\n",
       "          7.0000e+00, 2.0000e+00],\n",
       "         [3.8817e-01, 1.3529e+00, 1.5127e+00, 1.0000e+00, 9.8690e-01, 8.3133e-01,\n",
       "          7.0000e+00, 6.0000e+00],\n",
       "         [6.6274e-01, 1.8707e+00, 4.1016e-02, 1.0000e+01, 6.8815e-01, 9.1305e-01,\n",
       "          5.0000e+00, 8.0000e+00],\n",
       "         [5.8948e-01, 2.1177e+00, 5.8082e-02, 1.0000e+01, 4.5796e-01, 9.9961e-01,\n",
       "          5.0000e+00, 5.0000e+00],\n",
       "         [6.1700e-01, 2.6987e+00, 1.6310e+00, 4.0000e+00, 8.3021e-01, 9.6272e-01,\n",
       "          6.0000e+00, 5.0000e+00],\n",
       "         [5.9065e-01, 1.7060e+00, 1.0409e-01, 1.0000e+01, 6.1832e-01, 8.1384e-01,\n",
       "          5.0000e+00, 1.0000e+01],\n",
       "         [5.4431e-01, 1.3458e+00, 9.5058e-01, 1.0000e+01, 1.0000e+00, 8.2057e-01,\n",
       "          3.0000e+00, 8.0000e+00],\n",
       "         [5.0553e-01, 2.3250e+00, 1.2340e+00, 2.0000e+00, 6.3008e-01, 9.6528e-01,\n",
       "          6.0000e+00, 3.0000e+00],\n",
       "         [5.7433e-01, 2.9258e+00, 1.1961e+00, 1.0000e+00, 8.7632e-01, 9.5697e-01,\n",
       "          8.0000e+00, 5.0000e+00],\n",
       "         [5.1226e-01, 2.6103e+00, 6.5030e-01, 1.0000e+00, 6.8003e-01, 9.9312e-01,\n",
       "          6.0000e+00, 2.0000e+00],\n",
       "         [5.7277e-01, 3.2785e+00, 1.0482e+00, 1.0000e+00, 8.0932e-01, 9.0043e-01,\n",
       "          9.0000e+00, 2.1000e+01],\n",
       "         [4.1876e-01, 2.5684e+00, 8.0847e-01, 3.0000e+00, 8.8807e-01, 8.7155e-01,\n",
       "          9.0000e+00, 2.1000e+01],\n",
       "         [2.0930e-01, 6.2427e-01, 5.8344e-03, 1.0000e+01, 7.1593e-01, 8.7902e-01,\n",
       "          5.0000e+00, 1.0000e+00],\n",
       "         [6.4842e-01, 2.4245e+00, 1.0071e+00, 1.0000e+01, 7.0255e-01, 9.9222e-01,\n",
       "          4.0000e+00, 1.0000e+00],\n",
       "         [4.9532e-01, 2.5596e+00, 2.8411e+00, 1.0000e+00, 8.1758e-01, 1.0000e+00,\n",
       "          6.0000e+00, 1.0000e+01],\n",
       "         [5.5252e-01, 2.5977e+00, 1.2119e+00, 2.0000e+00, 8.8340e-01, 8.6782e-01,\n",
       "          7.0000e+00, 1.0000e+01],\n",
       "         [5.8590e-01, 2.8119e+00, 9.4075e-01, 1.0000e+00, 7.1421e-01, 9.8999e-01,\n",
       "          7.0000e+00, 1.5000e+01],\n",
       "         [6.2868e-01, 2.2278e+00, 9.0017e-02, 1.0000e+00, 6.2752e-01, 9.8826e-01,\n",
       "          6.0000e+00, 1.0000e+00],\n",
       "         [6.6013e-01, 2.0515e+00, 1.7314e+00, 1.0000e+01, 8.9174e-01, 7.6785e-01,\n",
       "          7.0000e+00, 1.1000e+01],\n",
       "         [6.8025e-01, 2.1515e+00, 1.3077e+00, 7.0000e+00, 9.4149e-01, 7.5254e-01,\n",
       "          7.0000e+00, 1.8000e+01],\n",
       "         [6.3777e-01, 2.8235e+00, 1.5360e+00, 1.0000e+00, 8.9634e-01, 9.6523e-01,\n",
       "          7.0000e+00, 1.7000e+01],\n",
       "         [6.7968e-01, 1.7016e+00, 1.9264e+00, 1.0000e+01, 8.4209e-01, 9.6003e-01,\n",
       "          4.0000e+00, 2.0000e+00],\n",
       "         [5.3229e-01, 2.2402e+00, 1.3076e+00, 1.0000e+00, 9.8436e-01, 8.5355e-01,\n",
       "          7.0000e+00, 8.0000e+00],\n",
       "         [5.8020e-01, 3.2291e+00, 5.1772e-01, 1.0000e+00, 6.8024e-01, 9.6299e-01,\n",
       "          8.0000e+00, 1.2000e+01],\n",
       "         [4.6125e-01, 2.3484e+00, 0.0000e+00, 1.0000e+00, 6.3093e-01, 9.8154e-01,\n",
       "          7.0000e+00, 2.0000e+00],\n",
       "         [6.1350e-01, 1.4058e+00, 3.8186e-01, 1.0000e+01, 7.6631e-01, 7.1432e-01,\n",
       "          6.0000e+00, 1.0000e+00],\n",
       "         [5.4960e-01, 1.4363e+00, 3.8706e-01, 1.0000e+01, 9.3367e-01, 7.2803e-01,\n",
       "          5.0000e+00, 3.0000e+00]], dtype=torch.float64), tensor([[81.1564],\n",
       "         [84.4100],\n",
       "         [84.0905],\n",
       "         [89.1601],\n",
       "         [80.6688],\n",
       "         [84.1000],\n",
       "         [86.6488],\n",
       "         [77.9078],\n",
       "         [81.1666],\n",
       "         [79.7700],\n",
       "         [79.6444],\n",
       "         [81.1784],\n",
       "         [76.9791],\n",
       "         [83.1937],\n",
       "         [88.2702],\n",
       "         [83.1661],\n",
       "         [88.7547],\n",
       "         [73.8955],\n",
       "         [80.2087],\n",
       "         [85.1370],\n",
       "         [82.0000],\n",
       "         [82.3458],\n",
       "         [81.3796],\n",
       "         [88.7547],\n",
       "         [84.1543],\n",
       "         [87.3097],\n",
       "         [77.9773],\n",
       "         [74.1732],\n",
       "         [79.5376],\n",
       "         [77.3440],\n",
       "         [79.1957],\n",
       "         [82.1400],\n",
       "         [80.1900],\n",
       "         [89.2529],\n",
       "         [81.4722],\n",
       "         [89.2529],\n",
       "         [80.5459],\n",
       "         [74.1678],\n",
       "         [71.4474],\n",
       "         [82.1971],\n",
       "         [78.3500],\n",
       "         [83.1300],\n",
       "         [78.1729],\n",
       "         [81.1800],\n",
       "         [78.5000],\n",
       "         [79.5457],\n",
       "         [71.7582],\n",
       "         [87.4581],\n",
       "         [80.7759],\n",
       "         [77.8998],\n",
       "         [82.6955],\n",
       "         [86.2480],\n",
       "         [82.5329],\n",
       "         [82.4573],\n",
       "         [84.2700],\n",
       "         [78.5900],\n",
       "         [74.0890],\n",
       "         [88.5978],\n",
       "         [80.2000],\n",
       "         [83.1321],\n",
       "         [83.1101],\n",
       "         [76.0311],\n",
       "         [78.6100],\n",
       "         [86.2319]], dtype=torch.float64)]"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tmp"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-03-15T08:34:32.586349Z",
     "start_time": "2020-03-15T08:34:32.582102Z"
    }
   },
   "outputs": [],
   "source": [
    "# for dictionary batch\n",
    "class Dataset(data_utils.Dataset):\n",
    "   \n",
    "    def __init__(self, X, y):\n",
    "        self.X = X\n",
    "        self.y = y\n",
    "   \n",
    "    def __getitem__(self, idx):\n",
    "        return {'X': self.X[idx], 'y': self.y[idx]}\n",
    "   \n",
    "    def __len__(self):\n",
    "        return len(self.X)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-03-15T08:34:32.823228Z",
     "start_time": "2020-03-15T08:34:32.819929Z"
    }
   },
   "outputs": [],
   "source": [
    "trn = Dataset(trn_X, trn_y)\n",
    "trn_loader = data_utils.DataLoader(trn, batch_size=batch_size, shuffle=True)\n",
    "\n",
    "val = Dataset(val_X, val_y)\n",
    "val_loader = data_utils.DataLoader(val, batch_size=batch_size, shuffle=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-03-15T08:34:33.223267Z",
     "start_time": "2020-03-15T08:34:33.219784Z"
    }
   },
   "outputs": [],
   "source": [
    "tmp = next(iter(trn_loader))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-03-15T08:34:33.710508Z",
     "start_time": "2020-03-15T08:34:33.698495Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'X': tensor([[6.8642e-01, 3.0662e+00, 1.0430e+00, 1.0000e+00, 7.3301e-01, 9.8043e-01,\n",
       "          7.0000e+00, 5.0000e+00],\n",
       "         [6.4069e-01, 3.3031e+00, 8.0163e-02, 9.0000e+00, 7.1114e-01, 9.6430e-01,\n",
       "          9.0000e+00, 1.2000e+01],\n",
       "         [6.1958e-01, 2.6973e+00, 1.0070e+00, 1.0000e+01, 5.4546e-01, 9.9144e-01,\n",
       "          7.0000e+00, 3.0000e+00],\n",
       "         [5.0175e-01, 1.5722e+00, 1.2750e+00, 1.0000e+00, 6.7294e-01, 6.9731e-01,\n",
       "          7.0000e+00, 1.3000e+01],\n",
       "         [5.6134e-01, 1.6326e+00, 3.3743e+00, 1.0000e+01, 9.8828e-01, 7.1937e-01,\n",
       "          6.0000e+00, 1.1000e+01],\n",
       "         [3.9965e-01, 1.5086e+00, 5.6467e-01, 1.0000e+01, 6.1882e-01, 7.5318e-01,\n",
       "          7.0000e+00, 1.0000e+00],\n",
       "         [4.9889e-01, 2.1329e+00, 4.6555e-01, 1.0000e+01, 7.2297e-01, 7.6876e-01,\n",
       "          7.0000e+00, 1.2000e+01],\n",
       "         [4.9348e-01, 2.5340e+00, 7.9817e-01, 7.0000e+00, 9.9741e-01, 9.5803e-01,\n",
       "          7.0000e+00, 3.0000e+00],\n",
       "         [5.6068e-01, 2.7336e+00, 1.1058e+00, 1.0000e+00, 6.6741e-01, 9.7421e-01,\n",
       "          6.0000e+00, 2.0000e+00],\n",
       "         [6.8664e-01, 3.1593e+00, 6.7231e-01, 1.0000e+00, 8.1791e-01, 9.9979e-01,\n",
       "          7.0000e+00, 7.0000e+00],\n",
       "         [3.9328e-01, 1.9788e+00, 1.1576e+00, 1.0000e+00, 9.3969e-01, 9.5979e-01,\n",
       "          6.0000e+00, 1.0000e+00],\n",
       "         [6.9814e-01, 2.6854e+00, 2.9404e-01, 1.0000e+01, 6.4895e-01, 1.0000e+00,\n",
       "          5.0000e+00, 4.0000e+00],\n",
       "         [4.3430e-01, 1.8448e+00, 1.8355e-02, 1.0000e+01, 6.1963e-01, 7.7998e-01,\n",
       "          7.0000e+00, 1.5000e+01],\n",
       "         [5.9991e-01, 2.1737e+00, 8.4708e-01, 1.0000e+00, 8.7779e-01, 8.4858e-01,\n",
       "          6.0000e+00, 1.7000e+01],\n",
       "         [4.4356e-01, 1.3238e+00, 7.0156e-01, 1.0000e+01, 5.9544e-01, 7.8175e-01,\n",
       "          4.0000e+00, 3.0000e+00],\n",
       "         [5.6070e-01, 2.2511e+00, 9.2327e-01, 1.0000e+00, 8.0837e-01, 8.5665e-01,\n",
       "          7.0000e+00, 1.4000e+01],\n",
       "         [5.7849e-01, 1.6029e+00, 1.7907e+00, 7.0000e+00, 9.2099e-01, 6.7282e-01,\n",
       "          7.0000e+00, 1.6000e+01],\n",
       "         [4.6559e-01, 3.0100e+00, 8.0727e-01, 1.0000e+00, 6.4756e-01, 1.0000e+00,\n",
       "          7.0000e+00, 1.2000e+01],\n",
       "         [7.3499e-01, 2.2644e+00, 1.8066e+00, 1.0000e+00, 9.5080e-01, 8.9771e-01,\n",
       "          6.0000e+00, 1.0000e+00],\n",
       "         [6.3248e-01, 3.2314e+00, 9.7484e-03, 1.0000e+00, 6.0323e-01, 8.9470e-01,\n",
       "          1.0000e+01, 1.1000e+01],\n",
       "         [6.4673e-01, 2.6737e+00, 1.0133e-02, 1.0000e+01, 5.7965e-01, 9.1145e-01,\n",
       "          7.0000e+00, 1.6000e+01],\n",
       "         [5.6597e-01, 2.7407e+00, 8.9787e-01, 1.0000e+00, 8.6448e-01, 9.6669e-01,\n",
       "          8.0000e+00, 1.2000e+01],\n",
       "         [5.8877e-01, 1.9434e+00, 2.4360e-01, 1.0000e+01, 5.9813e-01, 9.6676e-01,\n",
       "          5.0000e+00, 4.0000e+00],\n",
       "         [5.0490e-01, 2.5791e+00, 1.0665e+00, 1.0000e+01, 8.7686e-01, 9.5105e-01,\n",
       "          7.0000e+00, 1.4000e+01],\n",
       "         [5.2736e-01, 1.4089e+00, 4.9686e-01, 1.0000e+00, 6.9005e-01, 7.3738e-01,\n",
       "          7.0000e+00, 1.1000e+01],\n",
       "         [6.1017e-01, 2.3279e+00, 5.6309e-01, 1.0000e+01, 1.0000e+00, 9.5165e-01,\n",
       "          6.0000e+00, 1.0000e+00],\n",
       "         [5.2097e-01, 2.5022e+00, 1.2543e-02, 1.0000e+01, 6.5364e-01, 9.8130e-01,\n",
       "          6.0000e+00, 7.0000e+00],\n",
       "         [4.8532e-01, 2.3855e+00, 1.2214e+00, 1.0000e+00, 8.5681e-01, 9.4972e-01,\n",
       "          7.0000e+00, 1.3000e+01],\n",
       "         [6.0319e-01, 2.6227e+00, 9.5187e-01, 1.0000e+00, 1.0000e+00, 9.9838e-01,\n",
       "          7.0000e+00, 9.0000e+00],\n",
       "         [6.6506e-01, 3.4450e+00, 1.1966e+00, 1.0000e+00, 6.9277e-01, 9.0120e-01,\n",
       "          9.0000e+00, 2.1000e+01],\n",
       "         [6.9856e-01, 3.4765e+00, 1.1275e+00, 1.0000e+00, 9.1145e-01, 9.9653e-01,\n",
       "          7.0000e+00, 1.0000e+00],\n",
       "         [6.2475e-01, 3.4704e+00, 9.9335e-01, 1.0000e+00, 8.8233e-01, 9.8920e-01,\n",
       "          7.0000e+00, 1.3000e+01],\n",
       "         [4.3487e-01, 2.7007e+00, 1.1470e+00, 1.0000e+00, 9.6607e-01, 9.5539e-01,\n",
       "          7.0000e+00, 1.0000e+01],\n",
       "         [6.2387e-01, 1.8758e+00, 3.3498e-01, 1.0000e+01, 6.5946e-01, 7.2967e-01,\n",
       "          7.0000e+00, 1.3000e+01],\n",
       "         [6.0511e-01, 1.4761e+00, 3.4509e+00, 1.0000e+01, 9.8105e-01, 7.4718e-01,\n",
       "          6.0000e+00, 1.0000e+00],\n",
       "         [6.9797e-01, 2.2249e+00, 6.1652e-01, 1.0000e+00, 6.0577e-01, 9.0423e-01,\n",
       "          6.0000e+00, 6.0000e+00],\n",
       "         [5.2244e-01, 1.7817e+00, 4.9571e-02, 1.0000e+01, 1.0000e+00, 1.0000e+00,\n",
       "          4.0000e+00, 2.0000e+00],\n",
       "         [6.3342e-01, 3.2925e+00, 7.2138e-01, 1.0000e+00, 8.9349e-01, 9.6248e-01,\n",
       "          8.0000e+00, 1.6000e+01],\n",
       "         [5.2985e-01, 2.4009e+00, 1.4964e+00, 1.0000e+00, 7.7015e-01, 9.7767e-01,\n",
       "          5.0000e+00, 5.0000e+00],\n",
       "         [5.0448e-01, 2.7041e+00, 2.5717e-01, 1.0000e+00, 5.0690e-01, 1.0000e+00,\n",
       "          7.0000e+00, 1.3000e+01],\n",
       "         [8.1887e-01, 2.0664e+00, 2.8674e-01, 1.0000e+00, 7.2687e-01, 8.9451e-01,\n",
       "          5.0000e+00, 5.0000e+00],\n",
       "         [5.9259e-01, 2.8561e+00, 7.8048e-01, 1.0000e+00, 7.9606e-01, 1.0000e+00,\n",
       "          8.0000e+00, 4.0000e+00],\n",
       "         [4.2435e-01, 1.9302e+00, 5.6034e-01, 1.0000e+00, 7.0610e-01, 9.8949e-01,\n",
       "          5.0000e+00, 1.0000e+00],\n",
       "         [4.8559e-01, 1.7265e+00, 4.4234e+00, 1.0000e+01, 9.9372e-01, 7.4595e-01,\n",
       "          5.0000e+00, 6.0000e+00],\n",
       "         [6.2542e-01, 8.8630e-01, 6.1246e-01, 2.0000e+00, 7.7694e-01, 6.2570e-01,\n",
       "          6.0000e+00, 5.0000e+00],\n",
       "         [4.5494e-01, 1.7327e+00, 1.8310e+00, 1.0000e+01, 9.5780e-01, 9.0821e-01,\n",
       "          4.0000e+00, 1.0000e+00],\n",
       "         [5.7861e-01, 2.4130e+00, 1.1891e+00, 1.0000e+00, 7.1704e-01, 9.5076e-01,\n",
       "          6.0000e+00, 5.0000e+00],\n",
       "         [4.7249e-01, 2.4034e+00, 5.6079e-01, 5.0000e+00, 6.2290e-01, 9.6755e-01,\n",
       "          7.0000e+00, 2.0000e+00],\n",
       "         [4.8085e-01, 1.5492e+00, 1.3715e+00, 1.0000e+00, 1.0000e+00, 7.6411e-01,\n",
       "          7.0000e+00, 1.2000e+01],\n",
       "         [7.1390e-01, 3.9390e+00, 1.3013e+00, 1.0000e+00, 9.6328e-01, 9.9635e-01,\n",
       "          9.0000e+00, 1.7000e+01],\n",
       "         [4.8488e-01, 1.5106e+00, 4.5087e-01, 1.0000e+00, 7.5158e-01, 7.5484e-01,\n",
       "          7.0000e+00, 1.2000e+01],\n",
       "         [4.6226e-01, 2.3009e+00, 0.0000e+00, 1.0000e+00, 0.0000e+00, 9.6483e-01,\n",
       "          6.0000e+00, 1.0000e+00],\n",
       "         [6.9186e-01, 2.0232e+00, 8.6420e-01, 1.0000e+01, 6.4953e-01, 7.2438e-01,\n",
       "          7.0000e+00, 1.8000e+01],\n",
       "         [3.3702e-01, 1.0882e+00, 9.1506e-02, 1.0000e+01, 5.3977e-01, 7.7750e-01,\n",
       "          4.0000e+00, 6.0000e+00],\n",
       "         [5.8638e-01, 2.6035e+00, 9.1341e-01, 1.0000e+00, 4.9635e-01, 9.6703e-01,\n",
       "          8.0000e+00, 1.4000e+01],\n",
       "         [5.3799e-01, 3.0242e+00, 6.3566e-01, 1.0000e+00, 8.1984e-01, 9.7777e-01,\n",
       "          8.0000e+00, 1.3000e+01],\n",
       "         [5.7842e-01, 2.1993e+00, 1.0324e+00, 1.0000e+01, 9.8509e-01, 9.4975e-01,\n",
       "          6.0000e+00, 8.0000e+00],\n",
       "         [6.0391e-01, 2.8563e+00, 3.2455e-02, 3.0000e+00, 5.0859e-01, 9.6858e-01,\n",
       "          7.0000e+00, 9.0000e+00],\n",
       "         [7.1624e-01, 2.7303e+00, 8.9233e-03, 1.0000e+01, 6.1531e-01, 9.5453e-01,\n",
       "          6.0000e+00, 1.3000e+01],\n",
       "         [6.5378e-01, 1.0646e+00, 1.0766e+00, 3.0000e+00, 9.4590e-01, 5.9713e-01,\n",
       "          7.0000e+00, 2.0000e+00],\n",
       "         [4.2918e-01, 2.6191e+00, 1.0141e+00, 2.0000e+00, 7.5628e-01, 9.6297e-01,\n",
       "          7.0000e+00, 1.0000e+01],\n",
       "         [5.5990e-01, 2.6260e+00, 9.5748e-01, 1.0000e+01, 5.9909e-01, 9.1967e-01,\n",
       "          7.0000e+00, 7.0000e+00],\n",
       "         [6.4338e-01, 2.6092e+00, 1.3254e+00, 1.0000e+00, 9.3005e-01, 8.8520e-01,\n",
       "          8.0000e+00, 1.9000e+01],\n",
       "         [6.7000e-01, 2.9012e+00, 1.0437e+00, 1.0000e+00, 7.6431e-01, 9.6411e-01,\n",
       "          7.0000e+00, 1.4000e+01]], dtype=torch.float64),\n",
       " 'y': tensor([[91.7607],\n",
       "         [77.2609],\n",
       "         [85.0600],\n",
       "         [82.9805],\n",
       "         [74.7000],\n",
       "         [79.2039],\n",
       "         [74.6998],\n",
       "         [81.1529],\n",
       "         [85.0600],\n",
       "         [79.2200],\n",
       "         [77.0925],\n",
       "         [81.1800],\n",
       "         [76.9305],\n",
       "         [84.5184],\n",
       "         [78.4148],\n",
       "         [85.5929],\n",
       "         [87.8498],\n",
       "         [89.4800],\n",
       "         [87.8400],\n",
       "         [89.0960],\n",
       "         [81.3300],\n",
       "         [78.2434],\n",
       "         [81.1562],\n",
       "         [82.1274],\n",
       "         [80.9007],\n",
       "         [71.5931],\n",
       "         [80.8500],\n",
       "         [84.7338],\n",
       "         [80.5284],\n",
       "         [89.2529],\n",
       "         [78.9100],\n",
       "         [84.2712],\n",
       "         [83.1101],\n",
       "         [74.6941],\n",
       "         [81.5877],\n",
       "         [84.7256],\n",
       "         [77.1068],\n",
       "         [73.5123],\n",
       "         [86.8058],\n",
       "         [79.7191],\n",
       "         [82.7719],\n",
       "         [83.1100],\n",
       "         [77.5613],\n",
       "         [75.6800],\n",
       "         [81.5204],\n",
       "         [70.6198],\n",
       "         [80.5284],\n",
       "         [71.5696],\n",
       "         [70.8400],\n",
       "         [80.2064],\n",
       "         [85.0166],\n",
       "         [80.8561],\n",
       "         [78.6242],\n",
       "         [72.6605],\n",
       "         [85.3250],\n",
       "         [89.4794],\n",
       "         [77.4425],\n",
       "         [32.3709],\n",
       "         [84.4100],\n",
       "         [83.4693],\n",
       "         [77.4500],\n",
       "         [81.1500],\n",
       "         [87.8415],\n",
       "         [79.1470]], dtype=torch.float64)}"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tmp"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_batches = len(trn_loader)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Build Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "use_cuda = torch.cuda.is_available()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "class MLPRegressor(nn.Module):\n",
    "    \n",
    "    def __init__(self):\n",
    "        super(MLPRegressor, self).__init__()\n",
    "        h1 = nn.Linear(len(X_features), 50)\n",
    "        h2 = nn.Linear(50, 35)\n",
    "        h3 = nn.Linear(35, 1)\n",
    "        self.hidden = nn.Sequential(\n",
    "            h1,\n",
    "            nn.Tanh(),\n",
    "            h2,\n",
    "            nn.Tanh(),\n",
    "            h3,\n",
    "        )\n",
    "        if use_cuda:\n",
    "            self.hidden = self.hidden.cuda()\n",
    "        \n",
    "    def forward(self, x):\n",
    "        o = self.hidden(x)\n",
    "        return o.view(-1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = MLPRegressor()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "criterion = nn.MSELoss()\n",
    "learning_rate = 1e-3\n",
    "optimizer = optim.Adam(model.parameters(), lr=learning_rate)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_epochs = 10\n",
    "num_batches = len(trn_loader)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "35"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "num_batches"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 1/10 | step: 15/35 | trn_loss: 80.4772 | val_loss: 79.9006\n",
      "epoch: 1/10 | step: 30/35 | trn_loss: 79.5698 | val_loss: 78.7983\n",
      "epoch: 2/10 | step: 15/35 | trn_loss: 77.8362 | val_loss: 76.9738\n",
      "epoch: 2/10 | step: 30/35 | trn_loss: 76.5909 | val_loss: 75.7759\n",
      "epoch: 3/10 | step: 15/35 | trn_loss: 75.1105 | val_loss: 74.2213\n",
      "epoch: 3/10 | step: 30/35 | trn_loss: 73.8596 | val_loss: 73.4308\n",
      "epoch: 4/10 | step: 15/35 | trn_loss: 72.9306 | val_loss: 72.4381\n",
      "epoch: 4/10 | step: 30/35 | trn_loss: 72.2745 | val_loss: 71.6900\n",
      "epoch: 5/10 | step: 15/35 | trn_loss: 71.4342 | val_loss: 70.8200\n",
      "epoch: 5/10 | step: 30/35 | trn_loss: 70.6700 | val_loss: 70.2096\n",
      "epoch: 6/10 | step: 15/35 | trn_loss: 69.7620 | val_loss: 69.4448\n",
      "epoch: 6/10 | step: 30/35 | trn_loss: 69.3763 | val_loss: 68.8077\n",
      "epoch: 7/10 | step: 15/35 | trn_loss: 68.6557 | val_loss: 68.0679\n",
      "epoch: 7/10 | step: 30/35 | trn_loss: 67.7106 | val_loss: 67.4442\n",
      "epoch: 8/10 | step: 15/35 | trn_loss: 66.9765 | val_loss: 66.6800\n",
      "epoch: 8/10 | step: 30/35 | trn_loss: 66.6197 | val_loss: 66.0855\n",
      "epoch: 9/10 | step: 15/35 | trn_loss: 65.7067 | val_loss: 65.3690\n",
      "epoch: 9/10 | step: 30/35 | trn_loss: 65.2980 | val_loss: 64.8316\n",
      "epoch: 10/10 | step: 15/35 | trn_loss: 64.4303 | val_loss: 64.1169\n",
      "epoch: 10/10 | step: 30/35 | trn_loss: 64.0514 | val_loss: 63.5649\n",
      "finish Training\n"
     ]
    }
   ],
   "source": [
    "trn_loss_list = []\n",
    "val_loss_list = []\n",
    "\n",
    "for epoch in range(num_epochs):\n",
    "    trn_loss_summary = 0.0\n",
    "    for i, trn in enumerate(trn_loader):\n",
    "        trn_X, trn_y = trn['X'], trn['y']\n",
    "        if use_cuda:\n",
    "            trn_X, trn_y = trn_X.cuda(), trn_y.cuda()\n",
    "        optimizer.zero_grad()\n",
    "        trn_pred = model(trn_X)\n",
    "        trn_loss = criterion(trn_pred, trn_y)\n",
    "        trn_loss.backward()\n",
    "        optimizer.step()\n",
    "        \n",
    "        trn_loss_summary += trn_loss\n",
    "        \n",
    "        if (i+1) % 15 == 0:\n",
    "            with torch.no_grad():\n",
    "                val_loss_summary = 0.0\n",
    "                for j, val in enumerate(val_loader):\n",
    "                    val_X, val_y = val['X'], val['y']\n",
    "                    if use_cuda:\n",
    "                        val_X, val_y = val_X.cuda(), val_y.cuda()\n",
    "                    val_pred = model(val_X)\n",
    "                    val_loss = criterion(val_pred, val_y)\n",
    "                    val_loss_summary += val_loss\n",
    "                \n",
    "            print(\"epoch: {}/{} | step: {}/{} | trn_loss: {:.4f} | val_loss: {:.4f}\".format(\n",
    "                epoch + 1, num_epochs, i+1, num_batches, (trn_loss_summary/15)**(1/2), (val_loss_summary/len(val_loader))**(1/2)\n",
    "            ))\n",
    "                \n",
    "            trn_loss_list.append((trn_loss_summary/15)**(1/2))\n",
    "            val_loss_list.append((val_loss_summary/len(val_loader))**(1/2))\n",
    "            trn_loss_summary = 0.0\n",
    "        \n",
    "print(\"finish Training\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Visualize"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0,0.5,'loss')"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x648 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(16,9))\n",
    "x_range = range(len(trn_loss_list))\n",
    "plt.plot(x_range, trn_loss_list, label=\"trn\")\n",
    "plt.plot(x_range, val_loss_list, label=\"val\")\n",
    "plt.legend()\n",
    "plt.xlabel(\"training steps\")\n",
    "plt.ylabel(\"loss\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.9"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}