SAUBER Neckartor GRU Test-Copy1.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<function get_engine at 0x7f31e5298f28>\n"
     ]
    }
   ],
   "source": [
    "#Import database setup\n",
    "import setup_env"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from datetime import datetime\n",
    "from math import sqrt\n",
    "from numpy import concatenate\n",
    "from matplotlib import pyplot\n",
    "import pandas as pd\n",
    "from pandas import read_csv\n",
    "from pandas import DataFrame\n",
    "from pandas import concat\n",
    "from sklearn.preprocessing import MinMaxScaler\n",
    "from sklearn.preprocessing import RobustScaler\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from sklearn.metrics import mean_squared_error\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense\n",
    "from keras.layers import LSTM\n",
    "from keras.layers import GRU"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "from math import isnan"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "rows = \"*\"\n",
    "sel_link_id = 589239594\n",
    "sel_min_confidence = 0\n",
    "sel_max_weekday = 8\n",
    "sel_func_classes = ('4','3')\n",
    "sel_dir_travel = 'F'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "sql_query = f\"\"\"\n",
    "    SELECT tr.*  \n",
    "    FROM here_traffic.stuttgart_traffic tr\n",
    "    JOIN here_streets.fc_streets_all_2018q3 st on tr.link_id = st.link_id\n",
    "    WHERE tr.link_id = {sel_link_id}\n",
    "    AND tr.confidence > {sel_min_confidence}\n",
    "    AND tr.weekday_n < {sel_max_weekday}\n",
    "    --AND st.func_class in {sel_func_classes}\n",
    "    AND tr.dir_travel = 'F'\n",
    "    LIMIT 100000\n",
    "\"\"\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "DB Engine läuft\n",
      "Engine(postgresql://postgres:***@192.168.10.184:5430/here)\n"
     ]
    }
   ],
   "source": [
    "engine = setup_env.get_database()\n",
    "print(engine)\n",
    "try:\n",
    "    con = engine.raw_connection()\n",
    "    con.cursor().execute(\"SET SCHEMA '{}'\".format(\"here_traffic\"))\n",
    "except:\n",
    "    print(\"Con: DB Verbindung prüfen!\") \n",
    "    exit"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 457 ms, sys: 96.5 ms, total: 554 ms\n",
      "Wall time: 37.5 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "    pd_read = pd.read_sql_query(sql_query, con)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.DataFrame(\n",
    "    pd_read,\n",
    "    columns=[\n",
    "        \"id_pk\",\n",
    "        \"link_id\",\n",
    "        \"dir_travel\",\n",
    "        \"mean_kmh\",     #2\n",
    "        \"datum_zeit\",   #5\n",
    "        \"weekday_n\",    #6\n",
    "        \"epoch_60\",     #7\n",
    "        \"confidence\",\n",
    "        \"count_n\",\n",
    "        \"holiday\"\n",
    "    ],\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "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>id_pk</th>\n",
       "      <th>link_id</th>\n",
       "      <th>dir_travel</th>\n",
       "      <th>mean_kmh</th>\n",
       "      <th>datum_zeit</th>\n",
       "      <th>weekday_n</th>\n",
       "      <th>epoch_60</th>\n",
       "      <th>confidence</th>\n",
       "      <th>count_n</th>\n",
       "      <th>holiday</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>116979619</td>\n",
       "      <td>589239594</td>\n",
       "      <td>F</td>\n",
       "      <td>46.9</td>\n",
       "      <td>2019-03-14 23:00:00</td>\n",
       "      <td>4</td>\n",
       "      <td>23</td>\n",
       "      <td>40</td>\n",
       "      <td>44</td>\n",
       "      <td>-99</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>116979618</td>\n",
       "      <td>589239594</td>\n",
       "      <td>F</td>\n",
       "      <td>47.0</td>\n",
       "      <td>2019-03-14 22:00:00</td>\n",
       "      <td>4</td>\n",
       "      <td>22</td>\n",
       "      <td>40</td>\n",
       "      <td>53</td>\n",
       "      <td>-99</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>116979617</td>\n",
       "      <td>589239594</td>\n",
       "      <td>F</td>\n",
       "      <td>45.2</td>\n",
       "      <td>2019-03-14 21:00:00</td>\n",
       "      <td>4</td>\n",
       "      <td>21</td>\n",
       "      <td>40</td>\n",
       "      <td>74</td>\n",
       "      <td>-99</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>116979616</td>\n",
       "      <td>589239594</td>\n",
       "      <td>F</td>\n",
       "      <td>45.3</td>\n",
       "      <td>2019-03-14 20:00:00</td>\n",
       "      <td>4</td>\n",
       "      <td>20</td>\n",
       "      <td>40</td>\n",
       "      <td>96</td>\n",
       "      <td>-99</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>116979615</td>\n",
       "      <td>589239594</td>\n",
       "      <td>F</td>\n",
       "      <td>41.7</td>\n",
       "      <td>2019-03-14 19:00:00</td>\n",
       "      <td>4</td>\n",
       "      <td>19</td>\n",
       "      <td>40</td>\n",
       "      <td>139</td>\n",
       "      <td>-99</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16857</th>\n",
       "      <td>116980600</td>\n",
       "      <td>589239594</td>\n",
       "      <td>F</td>\n",
       "      <td>47.0</td>\n",
       "      <td>2019-04-29 05:00:00</td>\n",
       "      <td>1</td>\n",
       "      <td>5</td>\n",
       "      <td>40</td>\n",
       "      <td>30</td>\n",
       "      <td>-99</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16858</th>\n",
       "      <td>116980601</td>\n",
       "      <td>589239594</td>\n",
       "      <td>F</td>\n",
       "      <td>45.1</td>\n",
       "      <td>2019-04-29 06:00:00</td>\n",
       "      <td>1</td>\n",
       "      <td>6</td>\n",
       "      <td>40</td>\n",
       "      <td>116</td>\n",
       "      <td>-99</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16859</th>\n",
       "      <td>116980602</td>\n",
       "      <td>589239594</td>\n",
       "      <td>F</td>\n",
       "      <td>43.9</td>\n",
       "      <td>2019-04-29 07:00:00</td>\n",
       "      <td>1</td>\n",
       "      <td>7</td>\n",
       "      <td>40</td>\n",
       "      <td>181</td>\n",
       "      <td>-99</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16860</th>\n",
       "      <td>116980603</td>\n",
       "      <td>589239594</td>\n",
       "      <td>F</td>\n",
       "      <td>44.0</td>\n",
       "      <td>2019-04-29 08:00:00</td>\n",
       "      <td>1</td>\n",
       "      <td>8</td>\n",
       "      <td>40</td>\n",
       "      <td>141</td>\n",
       "      <td>-99</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16861</th>\n",
       "      <td>116980604</td>\n",
       "      <td>589239594</td>\n",
       "      <td>F</td>\n",
       "      <td>45.6</td>\n",
       "      <td>2019-04-29 09:00:00</td>\n",
       "      <td>1</td>\n",
       "      <td>9</td>\n",
       "      <td>40</td>\n",
       "      <td>80</td>\n",
       "      <td>-99</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>16862 rows × 10 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "           id_pk    link_id dir_travel  mean_kmh          datum_zeit  \\\n",
       "0      116979619  589239594          F      46.9 2019-03-14 23:00:00   \n",
       "1      116979618  589239594          F      47.0 2019-03-14 22:00:00   \n",
       "2      116979617  589239594          F      45.2 2019-03-14 21:00:00   \n",
       "3      116979616  589239594          F      45.3 2019-03-14 20:00:00   \n",
       "4      116979615  589239594          F      41.7 2019-03-14 19:00:00   \n",
       "...          ...        ...        ...       ...                 ...   \n",
       "16857  116980600  589239594          F      47.0 2019-04-29 05:00:00   \n",
       "16858  116980601  589239594          F      45.1 2019-04-29 06:00:00   \n",
       "16859  116980602  589239594          F      43.9 2019-04-29 07:00:00   \n",
       "16860  116980603  589239594          F      44.0 2019-04-29 08:00:00   \n",
       "16861  116980604  589239594          F      45.6 2019-04-29 09:00:00   \n",
       "\n",
       "       weekday_n  epoch_60  confidence  count_n  holiday  \n",
       "0              4        23          40       44      -99  \n",
       "1              4        22          40       53      -99  \n",
       "2              4        21          40       74      -99  \n",
       "3              4        20          40       96      -99  \n",
       "4              4        19          40      139      -99  \n",
       "...          ...       ...         ...      ...      ...  \n",
       "16857          1         5          40       30      -99  \n",
       "16858          1         6          40      116      -99  \n",
       "16859          1         7          40      181      -99  \n",
       "16860          1         8          40      141      -99  \n",
       "16861          1         9          40       80      -99  \n",
       "\n",
       "[16862 rows x 10 columns]"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "df.drop_duplicates(inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "df.set_index('datum_zeit', inplace=True, drop=True)\n",
    "df.sort_index(inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "fill_index = pd.date_range('2018-05-01 00:00:00', '2019-04-30 23:00:00', freq='1H')\n",
    "df = df.reindex(fill_index)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "df['link_id']=sel_link_id\n",
    "df['dir_travel']=df['dir_travel'].fillna(method='backfill')\n",
    "df['epoch_60']=df.index.hour\n",
    "df['weekday_n']=df.index.dayofweek+1\n",
    "\n",
    "df[\"hourweek\"] = (df[\"weekday_n\"]-1)*24+df[\"epoch_60\"]\n",
    "df[\"hourweek\"] = pd.to_numeric(df.hourweek, errors='coerce')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "count_mean = df['count_n'].groupby(df['hourweek']).mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "for index, row in df.iterrows():\n",
    "    m_idx = row['hourweek']\n",
    "    if isnan(row['count_n']):\n",
    "        row['count_n'] = count_mean[m_idx]\n",
    "        df['count_n'].at[index] = count_mean[m_idx].round()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "remain_nan = df['count_n'].isna().sum()\n",
    "if remain_nan > 0:\n",
    "    print(\"%i Remaining cols without value. Aborting.\" % (remain_nan))\n",
    "    exit(1)   "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "df['count_diff'] = df['count_n'].diff()\n",
    "df['count_diff'].fillna(0,inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "median = df['count_n'].median()\n",
    "std = df['count_n'].std()\n",
    "outliers = (df['count_n'] - median).abs() > 3*std\n",
    "df[outliers] = np.nan\n",
    "df['count_n'].fillna(median, inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "# pyplot.figure(figsize=(25, 4))\n",
    "# pyplot.plot(df['count_n'].diff().diff(24)['2018-05-01' :'2019-05-01'], color=\"blue\", label=\"df\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "# load dataset\n",
    "#dataset = df['count_n']\n",
    "dataset = df[['count_n','epoch_60','weekday_n','holiday']]\n",
    "values = dataset.values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(8760, 4)"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "def series_to_supervised(data, n_in=1, n_out=1, dropnan=True):\n",
    "\tn_vars = 1 if type(data) is list else data.shape[1]\n",
    "\tdf = DataFrame(data)\n",
    "\tcols, names = list(), list()\n",
    "\t# input sequence (t-n, ... t-1)\n",
    "\tfor i in range(n_in, 0, -1):\n",
    "\t\tcols.append(df.shift(i))\n",
    "\t\tnames += [('var%d(t-%d)' % (j+1, i)) for j in range(n_vars)]\n",
    "\t# forecast sequence (t, t+1, ... t+n)\n",
    "\tfor i in range(0, n_out):\n",
    "\t\tcols.append(df.shift(-i))\n",
    "\t\tif i == 0:\n",
    "\t\t\tnames += [('var%d(t)' % (j+1)) for j in range(n_vars)]\n",
    "\t\telse:\n",
    "\t\t\tnames += [('var%d(t+%d)' % (j+1, i)) for j in range(n_vars)]\n",
    "\t# put it all together\n",
    "\tagg = concat(cols, axis=1)\n",
    "\tagg.columns = names\n",
    "\t# drop rows with NaN values\n",
    "\tif dropnan:\n",
    "\t\tagg.dropna(inplace=True)\n",
    "\treturn agg"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "# load dataset\n",
    "#dataset = df['count_n']\n",
    "dataset = df[['count_n','epoch_60','weekday_n','holiday']]\n",
    "values = dataset.values\n",
    "# ensure all data is float\n",
    "values = values.astype('float32')\n",
    "# normalize features\n",
    "scaler = MinMaxScaler(feature_range=(0, 1))\n",
    "#scaled = scaler.fit_transform(values.reshape(-1,1))\n",
    "scaled = scaler.fit_transform(values)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_hours = 72\n",
    "n_features = dataset.shape[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "reframed = series_to_supervised(scaled, n_hours, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(720, 292)\n",
      "     var1(t-72)  var2(t-72)  var3(t-72)  var4(t-72)  var1(t-71)  var2(t-71)  \\\n",
      "268    0.004082    0.173913    0.333333        0.98    0.102041    0.217391   \n",
      "269    0.102041    0.217391    0.333333        0.98    0.248980    0.260870   \n",
      "270    0.248980    0.260870    0.333333        0.98    0.489796    0.304348   \n",
      "271    0.489796    0.304348    0.333333        0.98    0.404082    0.347826   \n",
      "272    0.404082    0.347826    0.333333        0.98    0.338775    0.391304   \n",
      "\n",
      "     var3(t-71)  var4(t-71)  var1(t-70)  var2(t-70)  ...  var3(t-2)  \\\n",
      "268    0.333333        0.98    0.248980    0.260870  ...   0.833333   \n",
      "269    0.333333        0.98    0.489796    0.304348  ...   0.833333   \n",
      "270    0.333333        0.98    0.404082    0.347826  ...   0.833333   \n",
      "271    0.333333        0.98    0.338775    0.391304  ...   0.833333   \n",
      "272    0.333333        0.98    0.346939    0.434783  ...   0.833333   \n",
      "\n",
      "     var4(t-2)  var1(t-1)  var2(t-1)  var3(t-1)  var4(t-1)   var1(t)  \\\n",
      "268        0.0   0.000000   0.130435   0.833333        0.0  0.024490   \n",
      "269        0.0   0.024490   0.173913   0.833333        0.0  0.020408   \n",
      "270        0.0   0.020408   0.217391   0.833333        0.0  0.008163   \n",
      "271        0.0   0.008163   0.260870   0.833333        0.0  0.073469   \n",
      "272        0.0   0.073469   0.304348   0.833333        0.0  0.151020   \n",
      "\n",
      "      var2(t)   var3(t)  var4(t)  \n",
      "268  0.173913  0.833333      0.0  \n",
      "269  0.217391  0.833333      0.0  \n",
      "270  0.260870  0.833333      0.0  \n",
      "271  0.304348  0.833333      0.0  \n",
      "272  0.347826  0.833333      0.0  \n",
      "\n",
      "[5 rows x 292 columns]\n"
     ]
    }
   ],
   "source": [
    "# frame as supervised learning\n",
    "reframed = series_to_supervised(scaled, n_hours, 1)\n",
    "print(reframed.shape)\n",
    "print(reframed.head())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(576, 288) 576 (576,)\n",
      "(576, 72, 4) (576,) (144, 72, 4) (144,)\n"
     ]
    }
   ],
   "source": [
    "values = reframed.values\n",
    "n_train_hours = 24 * 24\n",
    "train = values[:n_train_hours, :]\n",
    "test = values[n_train_hours:, :]\n",
    "# split into input and outputs\n",
    "n_obs = n_hours * n_features\n",
    "train_X, train_y = train[:, :n_obs], train[:, -n_features]\n",
    "test_X, test_y = test[:, :n_obs], test[:, -n_features]\n",
    "print(train_X.shape, len(train_X), train_y.shape)\n",
    "# reshape input to be 3D [samples, timesteps, features]\n",
    "train_X = train_X.reshape((train_X.shape[0], n_hours, n_features))\n",
    "test_X = test_X.reshape((test_X.shape[0], n_hours, n_features))\n",
    "print(train_X.shape, train_y.shape, test_X.shape, test_y.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "def rmse(y_true, y_pred):\n",
    "        return K.sqrt(K.mean(K.square(y_pred - y_true))) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 576 samples, validate on 144 samples\n",
      "Epoch 1/50\n",
      " - 2s - loss: 0.2187 - val_loss: 0.2581\n",
      "Epoch 2/50\n",
      " - 2s - loss: 0.2128 - val_loss: 0.2466\n",
      "Epoch 3/50\n",
      " - 2s - loss: 0.1757 - val_loss: 0.2134\n",
      "Epoch 4/50\n",
      " - 3s - loss: 0.1550 - val_loss: 0.1894\n",
      "Epoch 5/50\n",
      " - 2s - loss: 0.1455 - val_loss: 0.1794\n",
      "Epoch 6/50\n",
      " - 2s - loss: 0.1284 - val_loss: 0.1539\n",
      "Epoch 7/50\n",
      " - 2s - loss: 0.1146 - val_loss: 0.1325\n",
      "Epoch 8/50\n",
      " - 2s - loss: 0.1013 - val_loss: 0.1123\n",
      "Epoch 9/50\n",
      " - 2s - loss: 0.0980 - val_loss: 0.1120\n",
      "Epoch 10/50\n",
      " - 3s - loss: 0.1012 - val_loss: 0.1094\n",
      "Epoch 11/50\n",
      " - 2s - loss: 0.0995 - val_loss: 0.1126\n",
      "Epoch 12/50\n",
      " - 1s - loss: 0.0961 - val_loss: 0.1149\n",
      "Epoch 13/50\n",
      " - 1s - loss: 0.0912 - val_loss: 0.1076\n",
      "Epoch 14/50\n",
      " - 2s - loss: 0.0925 - val_loss: 0.1067\n",
      "Epoch 15/50\n",
      " - 2s - loss: 0.0942 - val_loss: 0.1084\n",
      "Epoch 16/50\n",
      " - 2s - loss: 0.0916 - val_loss: 0.1106\n",
      "Epoch 17/50\n",
      " - 3s - loss: 0.0890 - val_loss: 0.1055\n",
      "Epoch 18/50\n",
      " - 2s - loss: 0.0906 - val_loss: 0.1055\n",
      "Epoch 19/50\n",
      " - 3s - loss: 0.0899 - val_loss: 0.1069\n",
      "Epoch 20/50\n",
      " - 1s - loss: 0.0887 - val_loss: 0.1073\n",
      "Epoch 21/50\n",
      " - 1s - loss: 0.0867 - val_loss: 0.1030\n",
      "Epoch 22/50\n",
      " - 1s - loss: 0.0885 - val_loss: 0.1035\n",
      "Epoch 23/50\n",
      " - 2s - loss: 0.0897 - val_loss: 0.1060\n",
      "Epoch 24/50\n",
      " - 2s - loss: 0.0877 - val_loss: 0.1078\n",
      "Epoch 25/50\n",
      " - 1s - loss: 0.0857 - val_loss: 0.1025\n",
      "Epoch 26/50\n",
      " - 1s - loss: 0.0894 - val_loss: 0.1020\n",
      "Epoch 27/50\n",
      " - 2s - loss: 0.0874 - val_loss: 0.1079\n",
      "Epoch 28/50\n",
      " - 1s - loss: 0.0849 - val_loss: 0.1012\n",
      "Epoch 29/50\n",
      " - 1s - loss: 0.0878 - val_loss: 0.1024\n",
      "Epoch 30/50\n",
      " - 1s - loss: 0.0918 - val_loss: 0.1152\n",
      "Epoch 31/50\n",
      " - 2s - loss: 0.0882 - val_loss: 0.1027\n",
      "Epoch 32/50\n",
      " - 2s - loss: 0.0947 - val_loss: 0.1028\n",
      "Epoch 33/50\n",
      " - 1s - loss: 0.0912 - val_loss: 0.1139\n",
      "Epoch 34/50\n",
      " - 1s - loss: 0.0850 - val_loss: 0.0987\n",
      "Epoch 35/50\n",
      " - 1s - loss: 0.0867 - val_loss: 0.0994\n",
      "Epoch 36/50\n",
      " - 1s - loss: 0.0864 - val_loss: 0.1082\n",
      "Epoch 37/50\n",
      " - 1s - loss: 0.0834 - val_loss: 0.0985\n",
      "Epoch 38/50\n",
      " - 1s - loss: 0.0849 - val_loss: 0.0997\n",
      "Epoch 39/50\n",
      " - 1s - loss: 0.0842 - val_loss: 0.1055\n",
      "Epoch 40/50\n",
      " - 1s - loss: 0.0821 - val_loss: 0.0981\n",
      "Epoch 41/50\n",
      " - 1s - loss: 0.0834 - val_loss: 0.0985\n",
      "Epoch 42/50\n",
      " - 2s - loss: 0.0834 - val_loss: 0.1035\n",
      "Epoch 43/50\n",
      " - 2s - loss: 0.0815 - val_loss: 0.0985\n",
      "Epoch 44/50\n",
      " - 2s - loss: 0.0827 - val_loss: 0.0992\n",
      "Epoch 45/50\n",
      " - 1s - loss: 0.0833 - val_loss: 0.1031\n",
      "Epoch 46/50\n",
      " - 2s - loss: 0.0813 - val_loss: 0.0982\n",
      "Epoch 47/50\n",
      " - 3s - loss: 0.0826 - val_loss: 0.0996\n",
      "Epoch 48/50\n",
      " - 3s - loss: 0.0847 - val_loss: 0.1049\n",
      "Epoch 49/50\n",
      " - 3s - loss: 0.0821 - val_loss: 0.0980\n",
      "Epoch 50/50\n",
      " - 2s - loss: 0.0845 - val_loss: 0.0999\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# design network\n",
    "model = Sequential()\n",
    "model.add(LSTM(50, input_shape=(train_X.shape[1], train_X.shape[2])))\n",
    "model.add(Dense(1))\n",
    "model.compile(loss='mae', optimizer='adam')\n",
    "# fit network\n",
    "history = model.fit(train_X, train_y, epochs=50, batch_size=72, validation_data=(test_X, test_y), verbose=2, shuffle=False)\n",
    "# plot history\n",
    "pyplot.plot(history.history['loss'], label='train')\n",
    "pyplot.plot(history.history['val_loss'], label='test')\n",
    "pyplot.legend()\n",
    "pyplot.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "ename": "ValueError",
     "evalue": "cannot reshape array of size 41472 into shape (144,4)",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m\u001b[0m",
      "\u001b[0;31mValueError\u001b[0mTraceback (most recent call last)",
      "\u001b[0;32m<ipython-input-31-2243fa13119e>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0;31m# make a prediction\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      2\u001b[0m \u001b[0myhat\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtest_X\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 3\u001b[0;31m \u001b[0mtest_X\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtest_X\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mreshape\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtest_X\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshape\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtest_X\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshape\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m2\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      4\u001b[0m \u001b[0;31m# invert scaling for forecast\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      5\u001b[0m \u001b[0minv_yhat\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mconcatenate\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0myhat\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtest_X\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maxis\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mValueError\u001b[0m: cannot reshape array of size 41472 into shape (144,4)"
     ]
    }
   ],
   "source": [
    "# make a prediction\n",
    "yhat = model.predict(test_X)\n",
    "test_X = test_X.reshape((test_X.shape[0], test_X.shape[2]))\n",
    "# invert scaling for forecast\n",
    "inv_yhat = concatenate((yhat, test_X[:, 1:]), axis=1)\n",
    "inv_yhat = scaler.inverse_transform(inv_yhat)\n",
    "inv_yhat = inv_yhat[:,0]\n",
    "# invert scaling for actual\n",
    "test_y = test_y.reshape((len(test_y), 1))\n",
    "inv_y = concatenate((test_y, test_X[:, 1:]), axis=1)\n",
    "inv_y = scaler.inverse_transform(inv_y)\n",
    "inv_y = inv_y[:,0]\n",
    "# calculate RMSE\n",
    "rmse = sqrt(mean_squared_error(inv_y, inv_yhat))\n",
    "print('Test RMSE: %.3f' % rmse)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'inv_y' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m\u001b[0m",
      "\u001b[0;31mNameError\u001b[0mTraceback (most recent call last)",
      "\u001b[0;32m<ipython-input-30-ffa24419a31f>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0mpyplot\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfigure\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfigsize\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m25\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m4\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m \u001b[0mpyplot\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mplot\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minv_y\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;36m180\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mcolor\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m\"green\"\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlabel\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m\"count_n\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      3\u001b[0m \u001b[0;31m#pyplot.plot(df['count_n'].diff()['2019-04-29' :'2019-05-01'], color=\"blue\", label=\"df\")\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m \u001b[0mpyplot\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mplot\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minv_yhat\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;36m180\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mcolor\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m\"red\"\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlabel\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m\"count_n predicted\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      5\u001b[0m \u001b[0mpyplot\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mlegend\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'inv_y' is not defined"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<Figure size 1800x288 with 0 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "pyplot.figure(figsize=(25, 4))\n",
    "pyplot.plot(inv_y[:-1,][-180:], color=\"green\", label=\"count_n\")\n",
    "#pyplot.plot(df['count_n'].diff()['2019-04-29' :'2019-05-01'], color=\"blue\", label=\"df\")\n",
    "pyplot.plot(inv_yhat[1:,][-180:], color=\"red\", label=\"count_n predicted\")\n",
    "pyplot.legend()\n",
    "pyplot.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(144, 288)"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_X.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "pyplot.figure(figsize=(25, 4))\n",
    "pyplot.plot(df['count_n'].diff().diff(24)['2019-04-01' :'2019-05-01'], color=\"blue\", label=\"df\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "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"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}