06-full-model.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.compat.v1 import ConfigProto\n",
    "from tensorflow.compat.v1 import InteractiveSession\n",
    "config = ConfigProto()\n",
    "config.gpu_options.allow_growth = True\n",
    "session = InteractiveSession(config=config)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import datetime\n",
    "from pandas_summary import DataFrameSummary"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.utils import to_categorical"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.read_feather('train_normalized_data.fth')\n",
    "df_test = pd.read_feather('test_normalized_data.fth')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "cat_vars = ['Store', 'DayOfWeek', 'Year', 'Month', 'Day', 'StateHoliday', 'CompetitionMonthsOpen', 'Promo2Weeks', \n",
    "            'StoreType', 'Assortment', 'PromoInterval', 'CompetitionOpenSinceYear', 'Promo2SinceYear', 'State', \n",
    "            'Week', 'Events', 'Promo_fw', 'Promo_bw', 'StateHoliday_bool_fw', 'StateHoliday_bool_bw', 'SchoolHoliday_fw', 'SchoolHoliday_bw']\n",
    "\n",
    "cat_vars = ['Store', 'DayOfWeek']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "contin_vars = ['CompetitionDistance', \n",
    "   'Max_TemperatureC', 'Mean_TemperatureC', 'Min_TemperatureC', 'Precipitationmm',\n",
    "   'Max_Humidity', 'Mean_Humidity', 'Min_Humidity', 'Max_Wind_SpeedKm_h', \n",
    "   'Mean_Wind_SpeedKm_h', 'CloudCover', 'trend', 'trend_DE',\n",
    "   'AfterStateHoliday_bool', 'BeforeStateHoliday_bool', 'Promo', 'SchoolHoliday', 'StateHoliday_bool']\n",
    "contin_vars = ['BeforeStateHoliday_bool', 'Max_TemperatureC']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "uniques = DataFrameSummary(df[cat_vars]).summary().loc[['uniques']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "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>uniques</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Store</th>\n",
       "      <td>1115</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>DayOfWeek</th>\n",
       "      <td>7</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          uniques\n",
       "Store        1115\n",
       "DayOfWeek       7"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "uniques.T"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Asignación de dimensión de embeddings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "cat_var_dict = {'Store': 50, 'DayOfWeek': 2, 'Year': 2, 'Month': 2,\n",
    "'Day': 10, 'StateHoliday': 2, 'CompetitionMonthsOpen': 2,\n",
    "'Promo2Weeks': 1, 'StoreType': 2, 'Assortment': 3, 'PromoInterval': 3,\n",
    "'CompetitionOpenSinceYear': 4, 'Promo2SinceYear': 4, 'State': 6,\n",
    "'Week': 25, 'Events': 4, 'Promo_fw': 1,\n",
    "'Promo_bw': 1, 'StateHoliday_bool_fw': 1,\n",
    "'StateHoliday_bool_bw': 1, 'SchoolHoliday_fw': 1,\n",
    "'SchoolHoliday_bw': 1}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Store 50 1115 [   0    1    2 ... 1112 1113 1114]\n",
      "\n",
      "DayOfWeek 2 7 [0 1 2 3 4 5 6]\n",
      "\n"
     ]
    }
   ],
   "source": [
    "for v in cat_vars:\n",
    "    uniques_ = df[v].unique()\n",
    "    uniques_.sort()\n",
    "    print(v, cat_var_dict[v], len(uniques_), uniques_)\n",
    "    print()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Definición de modelo"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "add_customers = True\n",
    "log_output = False\n",
    "output_activation = 'linear'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.keras.callbacks import ModelCheckpoint\n",
    "from tensorflow.keras.optimizers import Adam\n",
    "from tensorflow.keras.layers import Embedding, Input, Flatten, Concatenate, Dense, BatchNormalization, Activation, LeakyReLU, Dropout\n",
    "from tensorflow.keras.regularizers import l2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_cat_vars_model(cat_vars, uniques, cat_var_dict):\n",
    "    cat_vars_embed_outs = []\n",
    "    cat_var_inputs = []\n",
    "    for cat_var in cat_vars:\n",
    "        cat_var_in = Input(shape=(1,), name=f\"{cat_var}_input\")\n",
    "        cat_var_inputs.append(cat_var_in)\n",
    "        embed_out = Embedding(uniques[cat_var][0], cat_var_dict[cat_var], name=f'{cat_var}_Embed')(cat_var_in)\n",
    "        flatten_out = Flatten(name=f\"{cat_var}_flat\")(embed_out)\n",
    "        cat_vars_embed_outs.append(flatten_out)\n",
    "    return cat_var_inputs, cat_vars_embed_outs\n",
    "\n",
    "def get_cont_vars_input(contin_vars, dense_layer=False):\n",
    "    cont_vars_inputs = []\n",
    "    cont_vars_outputs = []\n",
    "    for cont_var in contin_vars:\n",
    "        cont_var_in = Input(shape=(1,), name=f\"{cont_var}_input\")\n",
    "        cont_vars_inputs.append(cont_var_in)\n",
    "        if dense_layer:\n",
    "            cont_var_out = Dense(1, name=f\"{cont_var}_input\", activation = 'linear')(cont_var_in)\n",
    "            cont_vars_outputs.append(cont_var_out)\n",
    "        else:\n",
    "            cont_vars_outputs.append(cont_var_in)\n",
    "    return cont_vars_inputs, cont_vars_outputs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "cat_var_inputs, cat_vars_embed_outs = get_cat_vars_model(cat_vars, uniques, cat_var_dict)\n",
    "cont_vars_inputs,  cont_vars_outs= get_cont_vars_input(contin_vars)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Descomentar para ver resultados\n",
    "# cat_vars_embed_outs\n",
    "# cat_var_inputs\n",
    "# cont_vars_inputs\n",
    "# cont_vars_outs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "first_hidden_units = 1000\n",
    "second_hidden_units = 500\n",
    "l2_lambda = 1e-3\n",
    "merged = Concatenate(name='All_Concatenate')(cat_vars_embed_outs + cont_vars_inputs)\n",
    "x = Dense(first_hidden_units, kernel_initializer=\"uniform\", kernel_regularizer=l2(l2_lambda))(merged)\n",
    "# x = BatchNormalization()(x)\n",
    "x = Activation('relu')(x)\n",
    "# x = LeakyReLU()(x)\n",
    "x = Dense(second_hidden_units, kernel_initializer=\"uniform\", kernel_regularizer=l2(l2_lambda))(x)\n",
    "# x = BatchNormalization()(x)\n",
    "x = Activation('relu')(x)\n",
    "# x = LeakyReLU()(x)\n",
    "\n",
    "output_1 = Dense(1, name='Sales', activation=output_activation)(x)\n",
    "output_2 = Dense(1, name='Customers', activation=output_activation)(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "if add_customers:\n",
    "    model = Model(cat_var_inputs + cont_vars_inputs, [output_1, output_2])\n",
    "else: \n",
    "    model = Model(cat_var_inputs + cont_vars_inputs, [output_1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_1\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "Store_input (InputLayer)        [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "DayOfWeek_input (InputLayer)    [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Store_Embed (Embedding)         (None, 1, 50)        55750       Store_input[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "DayOfWeek_Embed (Embedding)     (None, 1, 2)         14          DayOfWeek_input[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "Store_flat (Flatten)            (None, 50)           0           Store_Embed[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "DayOfWeek_flat (Flatten)        (None, 2)            0           DayOfWeek_Embed[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "BeforeStateHoliday_bool_input ( [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Max_TemperatureC_input (InputLa [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "All_Concatenate (Concatenate)   (None, 54)           0           Store_flat[0][0]                 \n",
      "                                                                 DayOfWeek_flat[0][0]             \n",
      "                                                                 BeforeStateHoliday_bool_input[0][\n",
      "                                                                 Max_TemperatureC_input[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "dense_2 (Dense)                 (None, 1000)         55000       All_Concatenate[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "activation_2 (Activation)       (None, 1000)         0           dense_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_3 (Dense)                 (None, 500)          500500      activation_2[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "activation_3 (Activation)       (None, 500)          0           dense_3[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "Sales (Dense)                   (None, 1)            501         activation_3[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "Customers (Dense)               (None, 1)            501         activation_3[0][0]               \n",
      "==================================================================================================\n",
      "Total params: 612,266\n",
      "Trainable params: 612,266\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# Descomentar para ver\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save_weights('initial_weights.hdf5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cantidad en val: 30188, porcentaje: 0.9642465458145908\n"
     ]
    }
   ],
   "source": [
    "df_train = df[df.Date < datetime.datetime(2015, 7, 1)]  \n",
    "df_val = df[df.Date >= datetime.datetime(2015, 7, 1)]\n",
    "print(f'Cantidad en val: {len(df_val)}, porcentaje: {len(df_train)/(len(df_train) + len(df_val))}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "from matplotlib import pyplot as plt\n",
    "# plt.figure(figsize=(20,5))\n",
    "# plt.plot((df_train[df_train['Store']==1]['Sales'].values - df_train[df_train['Store']==1]['Sales'].mean())/df_train[df_train['Store']==1]['Sales'].std())\n",
    "# plt.show()\n",
    "# plt.figure(figsize=(20,5))\n",
    "# plt.plot(np.log(df_train[df_train['Store']==1]['Sales'].values)/np.max(np.log(df_train[df_train['Store']==1]['Sales'].values)))\n",
    "# plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 133,
   "metadata": {},
   "outputs": [],
   "source": [
    "stores_mean = {}\n",
    "for store, g_df in df_train.groupby('Store'):\n",
    "    stores_mean[store] = g_df[g_df['Sales'] > 0]['Sales'].mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 138,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_train.loc[:, 'mean_by_store'] = df_train['Store'].apply(stores_mean.get)\n",
    "df_val.loc[:, 'mean_by_store'] = df_val['Store'].apply(stores_mean.get)\n",
    "\n",
    "df_train.loc[:, 'Sales_store'] = df_train['Sales'] - df_train['mean_by_store']\n",
    "df_val.loc[:, 'Sales_store'] = df_val['Sales'] - df_val['mean_by_store']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 139,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras import backend as K"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 140,
   "metadata": {},
   "outputs": [],
   "source": [
    "def rmspe(y_true, y_pred):\n",
    "    return K.sqrt(K.mean(K.square((y_true - y_pred)/y_true)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 141,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_metric(df, sales_):\n",
    "    return np.sqrt((((df['Sales'] - sales_)/df['Sales'])**2).mean())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 142,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.0943562401279468"
      ]
     },
     "execution_count": 142,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "get_metric(df_val, df_val['mean_by_store'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 143,
   "metadata": {},
   "outputs": [],
   "source": [
    "all_vars = cat_vars + contin_vars\n",
    "X_train = np.hsplit(df_train[all_vars].values, len(all_vars))\n",
    "X_val = np.hsplit(df_val[all_vars].values, len(all_vars))\n",
    "X_test = np.hsplit(df_test[all_vars].values, len(all_vars))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 144,
   "metadata": {},
   "outputs": [],
   "source": [
    "if add_customers:\n",
    "    y_out_columns = ['Sales', 'Customers']\n",
    "else:\n",
    "    y_out_columns = ['Sales_store']\n",
    "    \n",
    "if log_output:\n",
    "    # Escala logaritmica\n",
    "    max_log_y = np.max(np.log(df[y_out_columns])).values\n",
    "    y_train = np.log(df_train[y_out_columns].values)/max_log_y\n",
    "    y_val = np.log(df_val[y_out_columns].values)/max_log_y\n",
    "else:\n",
    "    # Normalización\n",
    "#     y_mean = df_train[y_out_columns].mean().values\n",
    "#     y_std = df_train[y_out_columns].std().values\n",
    "#     y_train = (df_train[y_out_columns].values - y_mean)/y_std\n",
    "#     y_val = (df_val[y_out_columns].values - y_mean)/y_std\n",
    "    y_max = df_train[y_out_columns].max().values\n",
    "    y_train = df_train[y_out_columns].values/y_max\n",
    "    y_val = df_val[y_out_columns].values/y_max\n",
    "y_train = np.hsplit(y_train, y_train.shape[1])\n",
    "y_val = np.hsplit(y_val, y_val.shape[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 145,
   "metadata": {},
   "outputs": [],
   "source": [
    "lr = 0.001\n",
    "model.compile(optimizer=Adam(lr=lr), metrics=['mse', rmspe], loss='mse')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 146,
   "metadata": {},
   "outputs": [],
   "source": [
    "if add_customers:\n",
    "    checkpoint = ModelCheckpoint('bestmodel.hdf5', monitor='val_Sales_mse', verbose=1, save_best_only=True)\n",
    "else:\n",
    "    checkpoint = ModelCheckpoint('bestmodel.hdf5', monitor='val_loss', verbose=1, save_best_only=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 147,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n",
      "\n",
      "Epoch 00001: val_loss improved from inf to 0.00241, saving model to bestmodel.hdf5\n",
      "3181/3181 - 6s - loss: 0.0032 - mse: 0.0032 - rmspe: 28584.9297 - val_loss: 0.0024 - val_mse: 0.0024 - val_rmspe: 6987.0015\n",
      "Epoch 2/20\n",
      "\n",
      "Epoch 00002: val_loss did not improve from 0.00241\n",
      "3181/3181 - 6s - loss: 0.0032 - mse: 0.0031 - rmspe: 30246.5664 - val_loss: 0.0024 - val_mse: 0.0024 - val_rmspe: 5996.3315\n",
      "Epoch 3/20\n",
      "\n",
      "Epoch 00003: val_loss did not improve from 0.00241\n",
      "3181/3181 - 6s - loss: 0.0032 - mse: 0.0031 - rmspe: 26245.4902 - val_loss: 0.0024 - val_mse: 0.0024 - val_rmspe: 4579.7686\n",
      "Epoch 4/20\n",
      "\n",
      "Epoch 00004: val_loss improved from 0.00241 to 0.00238, saving model to bestmodel.hdf5\n",
      "3181/3181 - 5s - loss: 0.0032 - mse: 0.0031 - rmspe: 32159.5176 - val_loss: 0.0024 - val_mse: 0.0023 - val_rmspe: 12744.7197\n",
      "Epoch 5/20\n",
      "\n",
      "Epoch 00005: val_loss did not improve from 0.00238\n",
      "3181/3181 - 5s - loss: 0.0032 - mse: 0.0031 - rmspe: 36987.4336 - val_loss: 0.0024 - val_mse: 0.0024 - val_rmspe: 3571.9629\n",
      "Epoch 6/20\n",
      "\n",
      "Epoch 00006: val_loss improved from 0.00238 to 0.00237, saving model to bestmodel.hdf5\n",
      "3181/3181 - 6s - loss: 0.0032 - mse: 0.0031 - rmspe: 30227.8379 - val_loss: 0.0024 - val_mse: 0.0023 - val_rmspe: 3602.8274\n",
      "Epoch 7/20\n",
      "\n",
      "Epoch 00007: val_loss did not improve from 0.00237\n",
      "3181/3181 - 6s - loss: 0.0032 - mse: 0.0031 - rmspe: 24390.4727 - val_loss: 0.0024 - val_mse: 0.0024 - val_rmspe: 6782.6631\n",
      "Epoch 8/20\n",
      "\n",
      "Epoch 00008: val_loss improved from 0.00237 to 0.00235, saving model to bestmodel.hdf5\n",
      "3181/3181 - 6s - loss: 0.0032 - mse: 0.0031 - rmspe: 24205.0605 - val_loss: 0.0023 - val_mse: 0.0023 - val_rmspe: 4676.0630\n",
      "Epoch 9/20\n",
      "\n",
      "Epoch 00009: val_loss did not improve from 0.00235\n",
      "3181/3181 - 5s - loss: 0.0032 - mse: 0.0031 - rmspe: 27878.5020 - val_loss: 0.0024 - val_mse: 0.0023 - val_rmspe: 9577.1494\n",
      "Epoch 10/20\n",
      "\n",
      "Epoch 00010: val_loss did not improve from 0.00235\n",
      "3181/3181 - 6s - loss: 0.0032 - mse: 0.0031 - rmspe: 25539.4805 - val_loss: 0.0024 - val_mse: 0.0023 - val_rmspe: 8914.3955\n",
      "Epoch 11/20\n",
      "\n",
      "Epoch 00011: val_loss did not improve from 0.00235\n",
      "3181/3181 - 5s - loss: 0.0032 - mse: 0.0031 - rmspe: 36763.8320 - val_loss: 0.0024 - val_mse: 0.0024 - val_rmspe: 7638.3477\n",
      "Epoch 12/20\n",
      "\n",
      "Epoch 00012: val_loss did not improve from 0.00235\n",
      "3181/3181 - 6s - loss: 0.0032 - mse: 0.0031 - rmspe: 21779.1484 - val_loss: 0.0024 - val_mse: 0.0023 - val_rmspe: 6412.5991\n",
      "Epoch 13/20\n",
      "\n",
      "Epoch 00013: val_loss did not improve from 0.00235\n",
      "3181/3181 - 5s - loss: 0.0032 - mse: 0.0031 - rmspe: 20546.5156 - val_loss: 0.0024 - val_mse: 0.0023 - val_rmspe: 5955.6079\n",
      "Epoch 14/20\n",
      "\n",
      "Epoch 00014: val_loss improved from 0.00235 to 0.00234, saving model to bestmodel.hdf5\n",
      "3181/3181 - 6s - loss: 0.0031 - mse: 0.0031 - rmspe: 26011.4004 - val_loss: 0.0023 - val_mse: 0.0023 - val_rmspe: 9517.8896\n",
      "Epoch 15/20\n",
      "\n",
      "Epoch 00015: val_loss did not improve from 0.00234\n",
      "3181/3181 - 5s - loss: 0.0031 - mse: 0.0031 - rmspe: 28036.6250 - val_loss: 0.0025 - val_mse: 0.0024 - val_rmspe: 13017.4307\n",
      "Epoch 16/20\n",
      "\n",
      "Epoch 00016: val_loss did not improve from 0.00234\n",
      "3181/3181 - 6s - loss: 0.0031 - mse: 0.0031 - rmspe: 29693.8574 - val_loss: 0.0024 - val_mse: 0.0024 - val_rmspe: 7549.5166\n",
      "Epoch 17/20\n",
      "\n",
      "Epoch 00017: val_loss did not improve from 0.00234\n",
      "3181/3181 - 6s - loss: 0.0031 - mse: 0.0031 - rmspe: 30479.8262 - val_loss: 0.0025 - val_mse: 0.0025 - val_rmspe: 4913.4141\n",
      "Epoch 18/20\n",
      "\n",
      "Epoch 00018: val_loss did not improve from 0.00234\n",
      "3181/3181 - 6s - loss: 0.0031 - mse: 0.0031 - rmspe: 30230.7910 - val_loss: 0.0024 - val_mse: 0.0024 - val_rmspe: 3213.0254\n",
      "Epoch 19/20\n",
      "\n",
      "Epoch 00019: val_loss did not improve from 0.00234\n",
      "3181/3181 - 5s - loss: 0.0031 - mse: 0.0031 - rmspe: 22614.4336 - val_loss: 0.0024 - val_mse: 0.0023 - val_rmspe: 9185.4287\n",
      "Epoch 20/20\n",
      "\n",
      "Epoch 00020: val_loss did not improve from 0.00234\n",
      "3181/3181 - 6s - loss: 0.0031 - mse: 0.0031 - rmspe: 29076.2285 - val_loss: 0.0024 - val_mse: 0.0024 - val_rmspe: 6890.3833\n"
     ]
    }
   ],
   "source": [
    "epochs = 20\n",
    "batch_size = 256\n",
    "history = model.fit(X_train, y_train, validation_data=(X_val, y_val),  epochs=epochs, batch_size=batch_size, callbacks=[checkpoint], verbose=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 148,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7efb883cad68>]"
      ]
     },
     "execution_count": 148,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAY0AAAD4CAYAAAAQP7oXAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjMsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+AADFEAAAgAElEQVR4nO3deXxV9Z3/8dcnGwkQAllYTAhrhIIgSMSlotYN7K8tbcdWbGudVsb2V22ny0y1099PWzv+fnWm27RuZdRfrQ+nyNjaYRxGxSKCC2BwQbZA2CSyJAQICUsgyef3xznBa8xyk9zkhtz38/G4j9x7zvfc+zmHm7w553zP95i7IyIiEo2keBcgIiJnDoWGiIhETaEhIiJRU2iIiEjUFBoiIhK1lHgX0J1yc3N99OjR8S5DROSMsnbt2gPuntfSvD4dGqNHj6akpCTeZYiInFHMbFdr83R4SkREoqbQEBGRqCk0REQkagoNERGJmkJDRESiFlVomNkcMys1szIzu6OF+f3M7Mlw/mozGx0x7wfh9FIzmx1OSzezNWb2tpltMLMfR7R/Imy73sweNbPUZp91vpk1mNl1nV1pERHpnHZDw8ySgfuBa4FJwA1mNqlZs5uBQ+4+HvglcG+47CRgHjAZmAM8EL5fHXCFu58LTAPmmNmF4Xs9AUwEpgAZwPxmtdwLPNeptRURkS6JZk9jJlDm7tvd/SSwEJjbrM1c4LHw+VPAlWZm4fSF7l7n7juAMmCmB2rD9qnhwwHcfUk434E1QEHE53wT+CNQ0dEV7YjN+45w77ObqT5+qjs/RkTkjBNNaOQDuyNel4fTWmzj7vVANZDT1rJmlmxmbxEEwFJ3Xx35huFhqRuBZ8PX+cBngIfaKtbMbjGzEjMrqaysjGL1PuzdqmM8uHwbOw8c7dTyIiJ9VTShYS1Ma37nptbatLqsuze4+zSCPYmZZnZOs3YPACvcfWX4+lfA7e7e0Fax7r7A3YvdvTgvr8Wr4NtVmNMfgHcPHuvU8iIifVU0w4iUAyMjXhcAe1ppU25mKUAWcDCaZd39sJktJzjnsR7AzO4C8oCvRTQtBhYGR73IBT5uZvXu/uco1qFDRg5RaIiItCSaPY3XgSIzG2NmaQQnthc3a7MYuCl8fh2wLDwnsRiYF/auGgMUAWvMLM/MBgOYWQZwFbA5fD0fmA3c4O6NTR/g7mPcfbS7jyY4b/KN7ggMgAH9Usgd2I/dCg0RkQ9od0/D3evN7DaCHkvJwKPuvsHM7gZK3H0x8AjwuJmVEexhzAuX3WBmi4CNQD1wq7s3mNkI4LGwN1QSsMjdnwk/8iFgF/BauFfxJ3e/O4brHJXC7Ax2VSk0REQiRTXKrbsvAZY0m3ZnxPMTwOdaWfYe4J5m09YB01tpH02Q/XW7RXdRYXZ/Xt95qLs/RkTkjKIrwltRmN2fvdXHOVnf2H5jEZEEodBoxcjs/jQ67Dl8PN6liIj0GgqNVhRmqweViEhzCo1W6FoNEZEPU2i0YlhmOmkpSep2KyISQaHRiqQkY+SQDO1piIhEUGi0oTC7v67VEBGJoNBoQ2F2f3YfPEZwcbuIiCg02jAyuz81dfUcPqYh0kVEQKHRJnW7FRH5IIVGG9TtVkTkgxQabdAQ6SIiH6TQaIOGSBcR+SCFRjsKs3WthohIE4VGO3SthojI+xQa7dAQ6SIi71NotENDpIuIvE+h0Q5dqyEi8j6FRjt0rYaIyPsUGu3QEOkiIu9TaLRDQ6SLiLxPoRGFwuz+Cg0RERQaUSnM7s+7VRoiXUREoREFDZEuIhJQaERB3W5FRAIKjSio262ISEChEQUNkS4iElBoRCEYIj1N12qISMJTaERJ3W5FRBQaUVNoiIgoNKJWmN2fPYc1RLqIJDaFRpQ0RLqISJShYWZzzKzUzMrM7I4W5vczsyfD+avNbHTEvB+E00vNbHY4Ld3M1pjZ22a2wcx+HNH+ibDtejN71MxSw+lfNLN14eNVMzu3qyvfEbpWQ0QkitAws2TgfuBaYBJwg5lNatbsZuCQu48HfgncGy47CZgHTAbmAA+E71cHXOHu5wLTgDlmdmH4Xk8AE4EpQAYwP5y+A7jM3acCPwEWdGqNO0nXaoiIRLenMRMoc/ft7n4SWAjMbdZmLvBY+Pwp4Eozs3D6Qnevc/cdQBkw0wO1YfvU8OEA7r4knO/AGqAgnP6qux8Kl1nVNL2nDMtMJy1ZQ6SLSGKLJjTygd0Rr8vDaS22cfd6oBrIaWtZM0s2s7eACmCpu6+OfMPwsNSNwLMt1HQz8N8tFWtmt5hZiZmVVFZWRrF60UlKMgqyNUS6iCS2aELDWpjWfLjX1tq0uqy7N7j7NII9hplmdk6zdg8AK9x95Qc+yOxjBKFxe0vFuvsCdy929+K8vLyWmnTaKHW7FZEEF01olAMjI14XAHtaa2NmKUAWcDCaZd39MLCc4JwH4XvcBeQB341sa2ZTgYeBue5eFUXtMaUh0kUk0UUTGq8DRWY2xszSCE5sL27WZjFwU/j8OmBZeE5iMTAv7F01BigC1phZnpkNBjCzDOAqYHP4ej4wG7jB3U9fFGFmhcCfgBvdfUvnVrdrNES6iCS6lPYauHu9md0GPAckA4+6+wYzuxsocffFwCPA42ZWRrCHMS9cdoOZLQI2AvXAre7eYGYjgMfCnlRJwCJ3fyb8yIeAXcBrwbl0/uTudwN3EpwneSCcXu/uxbHZDNGJ7HY7ZEBaT360iEiv0G5oQNCjCVjSbNqdEc9PAJ9rZdl7gHuaTVsHTG+lfYs1uft83u9+GxeR3W7PHTk4nqWIiMSFrgjvAA2RLiKJTqHRARoiXUQSnUKjgzTarYgkMoVGByk0RCSRKTQ6qGmI9FMNGiJdRBKPQqODmoZIf++QhkgXkcSj0OggDZEuIolModFBGiJdRBKZQqODNES6iCQyhUYHaYh0EUlkCo1O0BDpIpKoFBqdoCHSRSRRKTQ6oWmI9OrjGiJdRBKLQqMTmrrd7qrSISoRSSwKjU5Qt1sRSVQKjU7QEOkikqgUGp2gIdJFJFEpNDpppLrdikgCUmh0kq7VEJFEpNDoJA2RLiKJSKHRSRoiXUQSkUKjkzREuogkIoVGJ+laDRFJRAqNTtIQ6SKSiBQanaQh0kUkESk0uqBQ3W5FJMEoNLpglIZIF5EEo9DoAg2RLiKJRqHRBRoiXUQSjUKjC9TtVkQSjUKjCzREuogkmqhCw8zmmFmpmZWZ2R0tzO9nZk+G81eb2eiIeT8Ip5ea2exwWrqZrTGzt81sg5n9OKL9E2Hb9Wb2qJmlhtPNzH4dvtc6MzuvqyvfVRoiXUQSTbuhYWbJwP3AtcAk4AYzm9Ss2c3AIXcfD/wSuDdcdhIwD5gMzAEeCN+vDrjC3c8FpgFzzOzC8L2eACYCU4AMYH44/VqgKHzcAjzYmRWONQ2RLiKJJJo9jZlAmbtvd/eTwEJgbrM2c4HHwudPAVeamYXTF7p7nbvvAMqAmR6oDdunhg8HcPcl4XwH1gAFEZ/x+3DWKmCwmY3ozErHkoZIF5FEEk1o5AO7I16Xh9NabOPu9UA1kNPWsmaWbGZvARXAUndfHfmG4WGpG4FnO1AHZnaLmZWYWUllZWUUq9c1GiJdRBJJNKFhLUxrfjVba21aXdbdG9x9GsGexEwzO6dZuweAFe6+sgN14O4L3L3Y3Yvz8vJaWCS2moZI33NYQ6SLSN8XTWiUAyMjXhcAe1prY2YpQBZwMJpl3f0wsJzgnAfhe9wF5AHf7WAdPU7XaohIIokmNF4HisxsjJmlEZzYXtyszWLgpvD5dcCy8JzEYmBe2LtqDMFJ7DVmlmdmgwHMLAO4Ctgcvp4PzAZucPfGZp/x5bAX1YVAtbvv7cQ6x5Su1RCRRJLSXgN3rzez24DngGTgUXffYGZ3AyXuvhh4BHjczMoI9jDmhctuMLNFwEagHrjV3RvCE9iPhT2pkoBF7v5M+JEPAbuA14Jz6fzJ3e8GlgAfJziZfgz4Smw2QddoiHQRSSTthgYEPZoI/mhHTrsz4vkJ4HOtLHsPcE+zaeuA6a20b7GmcM/l1mjq7UkaIl1EEomuCI8BDZEuIolCoREDGiJdRBKFQiMGNES6iCQKhUYMNHW71SEqEenrFBox0NTtVtdqiEhfp9CIAQ2RLiKJQqERAxoiXUQShUIjRjREuogkAoVGjOhaDRFJBAqNGBmlIdJFJAEoNGJEQ6SLSCJQaMSIrtUQkUSg0IgRXashIolAoREjGiJdRBKBQiNGNES6iCQChUYMqdutiPR1Co0YKtQQ6SLSxyk0YqhQQ6SLSB+n0IghdbsVkb5OoRFDTd1uFRoi0lcpNGKoaYh0XashIn2VQiOGNES6iPR1Co0Y0xDpItKXKTRiTNdqiEhfptCIsUINkS4ifZhCI8YKNUS6iPRhCo0Y07UaItKXKTRiTNdqiEhfptCIsaYh0t/VtRoi0gcpNGJMQ6SLSF+m0OgG6nYrIn2VQqMbaIh0EemrogoNM5tjZqVmVmZmd7Qwv5+ZPRnOX21moyPm/SCcXmpms8Np6Wa2xszeNrMNZvbjiPa3he3dzHIjpmeZ2X9GLPOVrqx4d2oaIr3q6Ml4lyIiElPthoaZJQP3A9cCk4AbzGxSs2Y3A4fcfTzwS+DecNlJwDxgMjAHeCB8vzrgCnc/F5gGzDGzC8P3egW4CtjV7DNuBTaGy1wO/NzM0jq2uj1jSn4WAFf/4iX+75JN7Ko6GueKRERiI5o9jZlAmbtvd/eTwEJgbrM2c4HHwudPAVeamYXTF7p7nbvvAMqAmR6oDdunhg8HcPc33X1nC3U4kBm+70DgIFAf5Xr2qAvG5vBvf3MBF47N4eGXd3D5z5Zz06NreGHjfhoadchKRM5cKVG0yQd2R7wuBy5orY2715tZNZATTl/VbNl8OL0HsxYYD9zv7qvbqeM+YDGwB8gErnf3D43VYWa3ALcAFBYWRrF63ePicblcPC6XfdUn+MOad1n4+rvM/30J+YMz+MIFhXy+eCR5mf3iVp+ISGdEs6dhLUxr/t/l1tq0uqy7N7j7NKAAmGlm57RTx2zgLeAsgkNa95nZoA+9ufsCdy929+K8vLx23rL7Dc9K5ztXn83Lt1/Bg188j1E5/fnn50q5+Kd/4Vt/eJM1Ow7qhLmInDGi2dMoB0ZGvC4g+N9+S23KzSwFyCI4fNTusu5+2MyWE5zzWN9GHV8BfurBX9gyM9sBTATWRLEOcZeanMS1U0Zw7ZQRlFXU8sTqXTy1tpzFb+9hwrBMvnTRKD4zPZ+B/aL5JxERiY9o9jReB4rMbEx44nkewWGiSIuBm8Ln1wHLwj/ui4F5Ye+qMUARsMbM8sxsMICZZRCc+N7cTh3vAleGywwDJgDbo6i/1xk/dCB3fXIyq//hSn762SmkJBv/+8/rueCeF/jff15P6b6aeJcoItKidv9bG56juA14DkgGHnX3DWZ2N1Di7ouBR4DHzayMYA9jXrjsBjNbBGwkOGl9q7s3mNkI4LHwvEYSsMjdnwEws28B3weGA+vMbIm7zwd+AvzOzN4hOOx1u7sfiOG26HH901KYN7OQ688fyVu7D/P4ql08WbKbx1ft4sKx2fz6hukMzUyPd5kiIqdZXz6eXlxc7CUlJfEuo0MOHT3JopLd/PKFLUzJz+KJ+ReSlqJrMEWk55jZWncvbmme/hr1MkMGpPG1y8Zx719N5fWdh7jnvzbGuyQRkdN01rWXmjstn3fKq3n45R1MLRjMX80oiHdJIiLa0+jN7rh2IheNzeEfnn6H9e9Vx7scERGFRm+WkpzEfV+YTs6ANL72+FoOaiwrEYkzhUYvlzOwHw/dOIPK2jpu+7c3qG/40EXwIiI9RqFxBphaMJh//PQ5vLqtin96rjTe5YhIAtOJ8DPE54tHsq78MAtWbGdKfhafPPeseJckIglIexpnkDs/MZkZo4bw/afWsXnfkXiXIyIJSKFxBklLSeLBL55HZnoKX3t8LdXHTsW7JBFJMAqNM8zQQek8+KXz2HP4ON9+8k0adX8OEelBCo0z0IxR2dz5ycm8WFrJr17YEu9yRCSBKDTOUF+6oJDPzSjg18vKeG7DvniXIyIJQqFxhjIzfvLpc5hakMX3Fr1NWUVt+wuJiHSRQuMMlp6azINfmkFaShJfe7yEmhM6MS4i3UuhcYbLH5zBfV+Yzs6qY/zdv7+tE+Mi0q0UGn3AxeNy+cG1E3luw34efGlbvMsRkT5ModFH3HzJGOZOO4ufPV/K8tKKeJcjIn2UQqOPMDN++tmpTBiWybf+8Ca7qo7GuyQR6YM09lQfkpGWzIIbi/nkfS8z/7ESPj09nwFpyQzol8LAfikMCB/B8+TT01KT9X8HEYmOQqOPKczpz29umM6t//YG/xzliLhpKUmng2RAWgqZ6SlkD0hj/qyxnD86u5srFpEzibn33d42xcXFXlJSEu8y4sLdqatv5GhdPUfrGqitq+foyfrgZ/iorWuIeP7BaWWVtVTW1PHZ6fnc8fGJDM1Mj/cqiUgPMbO17l7c0jztafRRZkZ6ajLpqcnkDOz48sdO1nP/i2X864odLN24n+9cfTZfvmgUKTqUJZLQ9BdAWtQ/LYW/nz2RZ789i+mjhnD3Mxv5xG9eZs2Og/EuTUTiSKEhbRqbN5DHvnI+D31pBjUn6vn8b1/jO0++RUXNiXiXJiJxoNCQdpkZc84ZzgvfvYzbPjae/1q3lyt/9hKPvLxD9ywXSTAKDYlaRloyfzd7As9951LOGzWEn4SHrFZvr4p3aSLSQxQa0mFjcgfwu6+cz29vDA5ZXb9gFd9e+CYVR3TISqSvU2hIp5gZsycHh6y+ecV4lryzjyt+/hIPr9zOKR2yEumzFBrSJRlpyXzvmuCQVfHoIfzjf23iE79+mVXbq+jL1wCJJCpd3Ccx4+4s3bifH//nRt47fJyCIRnMKspjVlEuF4/LYXD/tHiXKCJRaOviPoWGxNzxkw089UY5K7ZU8tq2Kmrr6kkymFIwmFnjc5lVlMv0wiGkpWhHV6Q3UmhI3JxqaOTt3YdZufUAK7dW8nZ5NQ2NTv+0ZC4am8MlRUGIjMsbiJnFu1wRIQahYWZzgH8BkoGH3f2nzeb3A34PzACqgOvdfWc47wfAzUAD8C13f87M0oEVQD+CoUyecve7wva3Ad8GxgF57n4g4nMuB34FpAIH3P2ytupWaPQ+1cdP8dq2Kl4uq+TlrQfYWXUMgBFZ6VwyPpdZZ+fx0XE55AzsF+dKRRJXl0LDzJKBLcDVQDnwOnCDu2+MaPMNYKq7f93M5gGfcffrzWwS8AdgJnAW8AJwNtAIDHD3WjNLBV4G/tbdV5nZdOAQsBwobgoNMxsMvArMcfd3zWyou7d5tyGFRu+3++AxVm49cDpEjpyoB2BKfhbXnz+SvzqvgIy05DhXKZJYujpg4UygzN23h2+2EJgLbIxoMxf4Ufj8KeA+C441zAUWunsdsMPMyoCZ7v4aUBu2Tw0fDuDub4af07yOLwB/cvd3w3a6PV0fMDK7P1+4oJAvXFBIQ6PzznvVrNxSyXMb9/G//ryenz1fypcuGMWXLxrF0EEaaVck3qI5E5kP7I54XR5Oa7GNu9cD1UBOW8uaWbKZvQVUAEvdfXU7dZwNDDGz5Wa21sy+3FIjM7vFzErMrKSysjKK1ZPeIjnJmDZyMN+8soj/vO0SFn3tIi4Yk839y8v46L3L+N6it9m450i8yxRJaNHsabR0drL5Ma3W2rS6rLs3ANPCw05Pm9k57r6+nVpnAFcCGcBrZrbK3bd84M3dFwALIDg81cb7SS9mZswck83MMdnsqjrK/3tlJ4tKdvPHN8q5eFwO82eN4fKzh5KUpJPnIj0pmj2NcmBkxOsCYE9rbcwsBcgCDkazrLsfJjh/MSeKOp5196PheY4VwLlR1C9nuFE5A/jRpybz2h1Xcse1E9leeZSv/q6Eq375Ek+s3sXxkw3xLlEkYUQTGq8DRWY2xszSgHnA4mZtFgM3hc+vA5Z5cIZ9MTDPzPqZ2RigCFhjZnnhHgZmlgFcBWxup47/AGaZWYqZ9QcuADZFUb/0EVn9U/n6ZeNYefvH+Jd50xiQlsIPn17PxT/9Cz9/vlTDtYv0gHYPT7l7fdgN9jmCLrePuvsGM7sbKHH3xcAjwOPhie6DBMFC2G4RwUnzeuBWd28wsxHAY2HPrCRgkbs/A2Bm3wK+DwwH1pnZEnef7+6bzOxZYB1B76uH2zmcJX1UanISc6fl86lzz+L1nYd4eOV27nuxjN++tJ1PTTuLmy8Zw0dGDIp3mSJ9ki7ukz5h54Gj/O7V4LzHsZMNfHR8DnfM+QhTCrLiXZrIGUdXhEvCqD52ij+8/i4Pr9zBwaN1/PXFY/jeNWczoF80fT5EBNoODQ3+I31K03mPv3zvMm6YWcijr+zg6l+8xNKN++NdmkifoNCQPikrI5V7PjOFP/7Pi8hMT+Vvfl/C1x9fy75qnSwX6QqFhvRpM0Zl88y3LuH7cybwYmkFV/3iJR57dScNjX33sKxId1JoSJ+XmpzENy4fz/PfuZTphYO5a/EGPvvgq2zYUx3v0kTOOAoNSRijcgbw+6/O5F/mTeO9Q8f41H2v8H+WbOLYyfp4lyZyxlBoSEIxM+ZOy+eF717G54sLWLBiO1f/YgUvbo7N+JcNjc6OA0dZXlrB9spaGnUYTPoYdbmVhLZmx0H+4el3KKuo5X9MHcFdn5gU1Wi6jY3Oe4ePs2V/DVv214Y/ayirqKWuvvF0u8x+KUzOH8TUgsGck5/F1PwsRuX0j/kNp46cOEVZRS1l+2vZWlHDrqpjjB86kIvH5VI8egjpqRpeXqKn6zRE2nCyvpHfvrSN37xYRr+UJG6fM5EvzCwkKclwd/YdOUHpvhq2RoTD1opajkWMeTUiK52zh2Vy9rCBFA3LZFR2f3YdPMY75dWse6+aTXuPcDIMk8z0FKbkZzGlIIsp+VlMzR/MyOyMqIKk+tgptlYEn98UUlv317LvyPu9wvqlJJE/JIN3q45R3+ikpSQxo3AIF4/L4eLxuUwtyCI1WQcZpHUKDZEo7DhwlB8+/Q6vbqti8lmDSE9NZsv+GmpOvH/OI3dgPyYMH0jR0EwmDA9CYvzQTLIyUtt871MNjWzZX8M75dW8817w2Ly3hpMNQZBkZaR+IEgmDM/kQE0dWypqKQtDamtFLZU1daffMyM1mfFDB1I0LKinKHxeMKQ/yUlGbV09r+84yKvbDvBKWRUb9wbDyg/sl8LMMdlBiIzLZeLwTI0WLB+g0BCJkrvz9Jvv8dBL2xjcP40J4d5DsBeRyZABaTH7rJP1QZCsOx0khyndV8Ophg/+Tg5IS2b8sEzOjgiI8UMHkj84o0N/7A8ePcmq7VW8UnaA17ZVsf3AUQCyB6Rx0dgcLh4fhMjobjh8JmcWhYbIGaKuvoHSfTWU7qth6KB0ioYOZERWerf8Ed9bfZxXy6p4ZdsBXi2rOn2I66ysdC4al8ulZ+dyaVFeTINSzgwKDRFpk3vQ6+vVbVW8uu0Ar26r4vCxUyQZnDtyMJedncflE4YyNT9Lh7ISgEJDRDqkodFZV36Y5aWVvLSlkrfLD+MeHMq6tCiXyycM5dKz88jWXkifpNAQkS45ePQkK7dWsry0khVbKqk6ehIzmFrQtBeSx7kFg0nWXkifoNAQkZhpbHTeea+a5aWVLN9SwVu7g72QIf1TmVUUBMilZ+eRO7BfvEvtNHdP6M4ACg0R6TaHjp5kxdbgMNaKLZUcqD0JBIeyhg9KZ0RWOsOzgp/DBqUzIivj9OvedJ+T3QeP8WJpBcs2V7BqexVT8rP4m1ljueojwxLuPI5CQ0R6RGOjs2HPEV4uO8DuQ8fYV32CvdUn2H/kBAePnvxQ+8z0FIYPej9UhmdlnA6a8UMHUjAkuoseO6O+oZG1uw6xrLSCFzdXsGV/LQCjcvpz4ZgcXi47wHuHjzM2dwDzZ43ls+flJ8yV9QoNEYm7E6ca2H/k/RDZW30iDJXj7DtSx77q41TU1BH5JymzXwoTR2TykRGDTj8mDMskI61zf7wPHj3JS1sq+MumClZsqeTIiXpSkoyZY7K5YuJQPjZxKGNzB2Bm1Dc0smT9Phas2Mb6946QMyCNL180mhsvGtXnOwAoNETkjHCqoZHKmjr2HD7Olv21bNp7hE17j7B5Xw21dcGV+UkGo3MH8JHhg/hIRKC0dD2Lu7Npbw3LNu9n2eYK3gzPv+QOTOPyCUO5cuJQLinKJTO99Sv63Z1V2w/yryu3s2xzBf1SkvhccQE3XzKWMbkDunV7xItCQ0TOaI2NTvmh42wMQ2TT3iNs2neE3QePn26TlZHKR0ZkMnH4IMYPHciGPUdYXlrB3vBujVMLsvjYhKFcMXEoUzp5vcnW/TU8vHIHT7/5HqcaG7lm0jBuuXQsM0Zlx2xdewOFhoj0STUnTlG6r4ZNe4+wcW/ws3RfDcdPNTAgLZlZRXlcMXEol0/Ii2r04mhV1Jzg8dd28fiqXRw+dorphYO5ZdZYrpk8vE90O1ZoiEjCaGh09hw+zrBB6aSldO9ovsdO1vPU2nIeXrmDdw8eY1ROf26+ZAzXzSigf1rv6RnWUQoNEZFu1NDoPL9hHwtWbufNdw8zuH8qn5x6FldPGsaFY3O6PbxiTaEhItJD1u46yKMv72TZ5gqOn2ogs18Kl03I4+pJw7h8wtB2h9HvDdoKjTN3/0lEpBeaMSqbGaOyOXGqgVfKDrB0435e2LSfZ9btJSXJuHBsDldPGsbVk4Zx1uCMmH1ufUMjuw4eY2t4N8mmgSZjTXsaIiLdrLHReXP3YZZu3M/zG/exvTK4l8nkswadDpBJIwZFdSFjQ6Ozq+ooW/bXBgFREfzcXnn09E29AG792Dj+fvbETtWrw1MiIr3Itspalm7cz9KN+3nj3UO4Q/7gDK6eNIxrJg3j/DHZJJmx++Cx07cXbrof/bbK2tO3DoZguaYbhRUNa7qb5MAunYhXaIiI9FKVNXUs2xwEyMqtB6irbySzXwqnGhs5cer9cDgrK/10KBSFd5IcP3QgA7th/C6d045xOYIAAAcuSURBVBAR6aXyMvtx/fmFXH9+IcdO1rNy6wFe2lJJRmry6YAoGjqwzavWe5JCQ0Skl+iflsLsycOZPXl4vEtpVVSdh81sjpmVmlmZmd3Rwvx+ZvZkOH+1mY2OmPeDcHqpmc0Op6Wb2Roze9vMNpjZjyPa3xa2dzPLbeGzzjezBjO7rjMrLCIindduaJhZMnA/cC0wCbjBzCY1a3YzcMjdxwO/BO4Nl50EzAMmA3OAB8L3qwOucPdzgWnAHDO7MHyvV4CrgF2t1HIv8FwH11NERGIgmj2NmUCZu29395PAQmBuszZzgcfC508BV1rQd2wusNDd69x9B1AGzPRAbdg+NXw4gLu/6e47W6nlm8AfgYqo1k5ERGIqmtDIB3ZHvC4Pp7XYxt3rgWogp61lzSzZzN4iCICl7r66rSLMLB/4DPBQO+1uMbMSMyuprKxsZ9VERKQjogmNlq42ad5Pt7U2rS7r7g3uPg0oAGaa2Tnt1PEr4HZ3b2irkbsvcPdidy/Oy4v91ZAiIoksmt5T5cDIiNcFwJ5W2pSbWQqQBRyMZll3P2xmywnOeaxvo45iYGF4xWQu8HEzq3f3P0exDiIiEgPR7Gm8DhSZ2RgzSyM4sb24WZvFwE3h8+uAZR5cNbgYmBf2rhoDFAFrzCzPzAYDmFkGwYnvzW0V4e5j3H20u48mOG/yDQWGiEjPajc0wnMUtxH0WNoELHL3DWZ2t5l9Kmz2CJBjZmXAd4E7wmU3AIuAjcCzwK3h4aURwItmto4glJa6+zMAZvYtMysn2CtZZ2YPx251RUSkK/r0MCJmVkkLXXejlAsciGE5sdbb64PeX6Pq6xrV1zW9ub5R7t7iSeE+HRpdYWYlrY290hv09vqg99eo+rpG9XVNb6+vNWfW7aRERCSuFBoiIhI1hUbrFsS7gHb09vqg99eo+rpG9XVNb6+vRTqnISIiUdOehoiIRE2hISIiUUv40OjKvUJ6oLaRZvaimW0K7zvyty20udzMqs3srfBxZ0/VF37+TjN7J/zsD91b1wK/DrffOjM7rwdrmxCxXd4ysyNm9u1mbXp8+5nZo2ZWYWbrI6Zlm9lSM9sa/hzSyrI3hW22mtlNLbXppvr+2cw2h/+GTzeN6NDCsm1+H7qxvh+Z2XsR/44fb2XZNn/fu7G+JyNq2xkO1trSst2+/brM3RP2ASQD24CxQBrwNjCpWZtvAA+Fz+cBT/ZgfSOA88LnmcCWFuq7HHgmjttwJ5DbxvyPA/9NMHjlhcDqOP5b7yO4aCmu2w+4FDgPWB8x7Z+AO8LndwD3trBcNrA9/DkkfD6kh+q7BkgJn9/bUn3RfB+6sb4fAX8XxXegzd/37qqv2fyfA3fGa/t19ZHoexpduVdIt3P3ve7+Rvi8hmAYl+bD0vd2c4Hfe2AVMNjMRsShjiuBbe7e2RECYsbdVxAM6Bkp8nv2GPDpFhadTTDkzkF3PwQsJRjos9vrc/fnPRhSCGAVwTA/cdHK9otGNL/vXdZWfeHfjs8Df4j15/aURA+NrtwrpEeFh8WmAy3dd+QiC26d+99mNrlHCwuGun/ezNaa2S0tzI9mG/eEebT+ixrP7ddkmLvvheA/C8DQFtr0lm35VYK9x5a0933oTreFh88ebeXwXm/YfrOA/e6+tZX58dx+UUn00OjKvUJ6jJkNJLhj4bfd/Uiz2W8QHHI5F/gN0NMj/37U3c8juB3wrWZ2abP5vWH7pQGfAv69hdnx3n4d0Ru25Q+BeuCJVpq0933oLg8C4whuH72X4BBQc3HffsANtL2XEa/tF7VED42O3CsE++C9QnqEmaUSBMYT7v6n5vPd/YiHt8519yVAqpnl9lR97r4n/FkBPE1wCCBSNNu4u10LvOHu+5vPiPf2i7C/6bBd+LOlWxrHdVuGJ94/AXzRwwPwzUXxfegW7r7fgxu7NQL/2srnxnv7pQCfBZ5srU28tl9HJHpodOVeId0uPP75CLDJ3X/RSpvhTedYzGwmwb9pVQ/VN8DMMpueE5wsbX4jrcXAl8NeVBcC1U2HYXpQq/+7i+f2aybye3YT8B8ttHkOuMbMhoSHX64Jp3U7M5sD3A58yt2PtdImmu9Dd9UXeZ7sM618bjS/793pKmCzu5e3NDOe269D4n0mPt4Pgt49Wwh6VfwwnHY3wS8HQDrBYY0yYA0wtgdru4Rg93kd8Fb4+DjwdeDrYZvbgA0EPUFWARf3YH1jw899O6yhaftF1mfA/eH2fQco7uF/3/4EIZAVMS2u248gwPYCpwj+93szwXmyvwBbw5/ZYdti4OGIZb8afhfLgK/0YH1lBOcDmr6HTT0KzwKWtPV96KH6Hg+/X+sIgmBE8/rC1x/6fe+J+sLpv2v63kW07fHt19WHhhEREZGoJfrhKRER6QCFhoiIRE2hISIiUVNoiIhI1BQaIiISNYWGiIhETaEhIiJR+/8ZsjnvmxU/twAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history.history['val_loss'])\n",
    "# plt.plot(history.history['val_Customers_mse'])\n",
    "# plt.plot(history.history['val_Sales_mse'])\n",
    "plt.show()\n",
    "plt.plot(history.history['loss'])\n",
    "# plt.plot(history.history['Customers_mse'])\n",
    "# plt.plot(history.history['Sales_mse'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Métrica"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "$$\n",
    "\\textrm{RMSE} = \\sqrt{\\frac{1}{n} \\sum_{i=1}^{n} \\left(\\frac{\\hat{y}_i - y_i}{y_i}\\right)^2}\n",
    "$$"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 149,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "944/944 [==============================] - 1s 870us/step - loss: 0.0024 - mse: 0.0024 - rmspe: 6890.3286\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.0024249653797596693, 0.0023766581434756517, 6890.32861328125]"
      ]
     },
     "execution_count": 149,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.evaluate(X_val, y_val)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 150,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "944/944 [==============================] - 1s 871us/step - loss: 0.0023 - mse: 0.0023 - rmspe: 9517.7100\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.0023449291475117207, 0.0022967790719121695, 9517.7099609375]"
      ]
     },
     "execution_count": 150,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.load_weights('bestmodel.hdf5')\n",
    "model.evaluate(X_val, y_val)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 151,
   "metadata": {},
   "outputs": [],
   "source": [
    "# model.load_weights('bestmodel.hdf5')\n",
    "# model.evaluate(X_val, y_val)\n",
    "# 30188/30188 [==============================] - 5s 172us/step\n",
    "# [0.12197033089921382,\n",
    "#  0.07211007360268763,\n",
    "#  0.037183713050426136,\n",
    "#  0.07211007360268763,\n",
    "#  0.037183713050426136]\n",
    "\n",
    "# [0.12932546436786652,\n",
    "#  0.07751645147800446,\n",
    "#  0.039259567856788635,\n",
    "#  0.07751645147800446,\n",
    "#  0.039259567856788635]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 152,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "944/944 [==============================] - 1s 578us/step\n",
      "1284/1284 [==============================] - 1s 669us/step\n"
     ]
    }
   ],
   "source": [
    "if log_output:\n",
    "    if add_customers:\n",
    "        y_pred = np.exp(model.predict(X_val, verbose=1)[0][:, 0]*max_log_y[0])\n",
    "        y_pred_test = np.exp(model.predict(X_test, verbose=1)[0][:, 0]*max_log_y[0])\n",
    "    else:\n",
    "        y_pred = np.exp(model.predict(X_val, verbose=1)*max_log_y)[:,0]\n",
    "        y_pred_test = np.exp(model.predict(X_test, verbose=1)*max_log_y)[:,0]\n",
    "else:\n",
    "    if add_customers:\n",
    "        y_pred = (model.predict(X_val, verbose=1)[0]*y_std[0] + y_mean[0])[:,0]\n",
    "        y_pred_test = (model.predict(X_test, verbose=1)[0]*y_std[0] + y_mean[0])[:,0]\n",
    "    else:\n",
    "#         y_pred = model.predict(X_val, verbose=1)[:,0]*y_std + y_mean\n",
    "#         y_pred_test = model.predict(X_test, verbose=1)[:,0]*y_std + y_mean\n",
    "        y_pred = model.predict(X_val, verbose=1)[:,0]*y_max\n",
    "        y_pred_test = model.predict(X_test, verbose=1)[:,0]*y_max\n",
    "y_pred_test[df_test['Open'] == 0] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 153,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1.0362932044369766"
      ]
     },
     "execution_count": 153,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.sqrt((((df_val['Sales'].values - y_pred)/df_val['Sales'].values)**2).sum()/len(y_pred))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 154,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ -43.74442669,  194.92447543,  -78.59563165,  197.4268742 ,\n",
       "       -235.0385011 ,  -42.97024203,  177.68631682, -109.93191853,\n",
       "        -55.80277873,   76.66592689])"
      ]
     },
     "execution_count": 154,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_pred_test[:10]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Baseline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "sample_csv = pd.read_csv('dataset/rossmann/sample_submission.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [],
   "source": [
    "stores_mean = {}\n",
    "for store, g_df in df.groupby('Store'):\n",
    "    stores_mean[store] = g_df[g_df['Sales'] > 0]['Sales'].mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_test['Sales'] = df_test['Store'].apply(stores_mean.get)\n",
    "df_test.loc[df_test['Open'] == 0, 'Sales'] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "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>Store</th>\n",
       "      <th>Sales</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>4759.096031</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>6942.568678</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>6</td>\n",
       "      <td>8817.050891</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>7</td>\n",
       "      <td>5539.358418</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>8</td>\n",
       "      <td>6562.337612</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>9</td>\n",
       "      <td>5568.420918</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>10</td>\n",
       "      <td>8030.977041</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>11</td>\n",
       "      <td>7589.598214</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>12</td>\n",
       "      <td>5034.747182</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>13</td>\n",
       "      <td>5508.567394</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Store        Sales\n",
       "0      0  4759.096031\n",
       "1      2  6942.568678\n",
       "2      6  8817.050891\n",
       "3      7  5539.358418\n",
       "4      8  6562.337612\n",
       "5      9  5568.420918\n",
       "6     10  8030.977041\n",
       "7     11  7589.598214\n",
       "8     12  5034.747182\n",
       "9     13  5508.567394"
      ]
     },
     "execution_count": 89,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_test[['Store', 'Sales']].head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "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>Store</th>\n",
       "      <th>Sales</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>543</th>\n",
       "      <td>702</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>676</th>\n",
       "      <td>878</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>840</th>\n",
       "      <td>1096</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1399</th>\n",
       "      <td>702</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1532</th>\n",
       "      <td>878</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      Store  Sales\n",
       "543     702    0.0\n",
       "676     878    0.0\n",
       "840    1096    0.0\n",
       "1399    702    0.0\n",
       "1532    878    0.0"
      ]
     },
     "execution_count": 90,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_test[df_test['Open'] == 0][['Store', 'Sales']].head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 91,
   "metadata": {},
   "outputs": [],
   "source": [
    "sample_csv['Sales'] = df_test['Sales']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {},
   "outputs": [],
   "source": [
    "sample_csv.to_csv(f'submision_baseline.csv', index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "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</th>\n",
       "      <th>Sales</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>4759.096031</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>6942.568678</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3</td>\n",
       "      <td>8817.050891</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>5539.358418</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5</td>\n",
       "      <td>6562.337612</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Id        Sales\n",
       "0   1  4759.096031\n",
       "1   2  6942.568678\n",
       "2   3  8817.050891\n",
       "3   4  5539.358418\n",
       "4   5  6562.337612"
      ]
     },
     "execution_count": 93,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sample_csv.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Sumbit a la competición"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "sample_csv = pd.read_csv('dataset/rossmann/sample_submission.csv')\n",
    "sample_csv['Sales'] = y_pred_test\n",
    "sample_csv.head()\n",
    "\n",
    "sample_csv.to_csv(f'submision_{add_customers}-{log_output}-{output_activation}-{l2_lambda}-{first_hidden_units}-{epochs}-{batch_size}-{lr}.csv', index=False)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Analisis de embedings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_embed(layer_name, cat_names):\n",
    "    Y = model.get_layer(layer_name).get_weights()[0]\n",
    "    print(Y.shape)\n",
    "    plt.figure(figsize=(8,8))\n",
    "    plt.scatter(-Y[:, 0], -Y[:, 1])\n",
    "    for i, txt in enumerate(cat_names):\n",
    "        plt.annotate(txt, (-Y[i, 0],-Y[i, 1]), xytext = (-5, 8), textcoords = 'offset points')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(7, 2)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "model.load_weights('bestmodel.hdf5')\n",
    "plot_embed('DayOfWeek_Embed', ['Mon', 'Tue', 'Wed', 'Thu', 'Fri', 'Sat','Sun'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(12, 2)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_embed('Month_Embed', list(range(12)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.manifold import TSNE\n",
    "from sklearn.decomposition import PCA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(7, 2)"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "day_of_week_embedding.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAegAAAHSCAYAAAAnsVjHAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjMsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+AADFEAAAeS0lEQVR4nO3df3DV9Z3v8de7J5hGsUZMbE1Qg1VzVyBN9OAvpk5GWmKrRbrV0bbX2kLLOP4Y1xa2jZ2p2Clud9Jpe53dbcvYXtgpI9uLgBa7N7hW7LSgNJEURExhuqyaoBJqunB7UALv+0dO0kADGnPO+b6T83zMOOT7Oefk++Y7yJPv93w5mLsLAADE8p6kBwAAAH+NQAMAEBCBBgAgIAINAEBABBoAgIAINAAAAZUkPcBQFRUVXlNTk/QYAAAUTHt7e4+7Vx67HirQNTU1amtrS3oMAAAKxsz+a7h1LnEDABAQgQYAIKBQl7gBFLd9+/Zp1qxZkqRXX31VqVRKlZX9b81t3rxZJ510UpLjAQVFoAGEccYZZ6ijo0OStHjxYk2cOFELFy5MeCogGVziBhDerl27VF9fP7j97W9/W9/61rckSTt37lRTU5MuueQSXXXVVfr973+f1JhATnEGDWBMW7BggR566CF98IMf1G9+8xvdeeedWr9+fdJjAaNGoAGMWb29vXrmmWf0qU99anCtr68vwYmA3CHQAMIrKSnRkSNHBrcPHjyokpISubsqKioG37cGxhPegwYQ3gc+8AF1d3frjTfe0MGDB/X4449Lkk4//XSdddZZWrNmjSTpyJEj+t3vfpfkqEDOEGgA4b33ve/VvffeqxkzZmjOnDm66KKLBh9buXKlfvjDH+pDH/qQpk6dqnXr1iU4KZA75u5JzzAonU47H/UJACgmZtbu7ulj1zmDBgAgIAINAEBABBoAgIAINAAAAfH3oAGEtHZLl1paO9Xdm1FVeZkWNdVqbkN10mMBBUOgAYSzdkuXmldvU+bQYUlSV29Gzau3SRKRRtHgEjeAcFpaOwfjPCBz6LBaWjsTmggoPAINIJzu3syI1oHxiEADCKeqvGxE68B4RKABhLOoqVZlE1JHrZVNSGlRU21CEwGFx01iAMIZuBGMu7hRzAg0gJDmNlQTZBQ1LnEDABAQgQYAICACDQBAQAQaAICACDQAAAERaAAAAiLQAAAERKABAAiIQAMAEBCBBgAgIAINAEBABBoAgIAINAAAARFoAAACItAAAAREoAEACIhAAwAQEIEGACAgAg0AQEAEGgCAgAg0AAABEWgAAAIi0AAABESgAQAIiEADABAQgQYAICACDQBAQAQaAICACDQAAAHlJNBmVm5mq8zsRTPbYWZXmNkkM3vCzHZmfzw9F/sCAKAY5OoM+n9J+r/u/j8kfUjSDklfk/Sku18g6cnsNgAAeAdGHWgze5+kqyT9WJLc/S1375V0vaTl2actlzR3tPsCAKBY5OIM+jxJeyX9bzPbYmYPmdkpkt7v7nskKfvjmTnYFwAARSEXgS6RdLGkH7h7g6T/pxFczjazBWbWZmZte/fuzcE4AACMfbkI9CuSXnH3Z7Pbq9Qf7NfM7CxJyv74+nAvdvel7p5293RlZWUOxgEAYOwbdaDd/VVJL5tZbXZplqQXJD0m6dbs2q2SHh3tvgAAKBYlOfo+d0laYWYnSfqDpC+oP/4/M7P5kl6SdGOO9gUAwLiXk0C7e4ek9DAPzcrF9wcAoNjwSWIAAAREoAEACIhAAwAQEIEGACAgAg0AQEAEGgCAgAg0AAABEWgAAAIi0AAABESgC2jJkiWaOnWq6urqVF9fr2efffa4z122bJm6u7sLOB0AIJJcfRY33samTZu0bt06PffccyotLVVPT4/eeuut4z5/2bJlmjZtmqqqqgo4JQAgCs6gC2TPnj2qqKhQaWmpJKmiokJVVVX65je/qRkzZmjatGlasGCB3F2rVq1SW1ubPvvZz6q+vl6ZTCbh6QEAhUagC2T27Nl6+eWXdeGFF+r222/X008/LUm688479dvf/lbPP/+8MpmM1q1bpxtuuEHpdForVqxQR0eHysrKEp4eAFBoBLpAJk6cqPb2di1dulSVlZW66aabtGzZMj311FO67LLLNH36dP3yl7/U9u3bkx4VABAA70EXUCqVUmNjoxobGzV9+nT96Ec/0tatW9XW1qazzz5bixcv1sGDB5MeEwAQAGfQBdLZ2amdO3cObnd0dKi2tlZS//vRBw4c0KpVqwYfP/XUU7V///6CzwkAiIEz6AI5cOCA7rrrLvX29qqkpETnn3++li5dqvLyck2fPl01NTWaMWPG4PM///nP67bbblNZWZk2bdrE+9AAUGTM3ZOeYVA6nfa2trakxwAAoGDMrN3d08euc4kbAICACDQAAAERaAAAAuImsQJbu6VLLa2d6u7NqKq8TIuaajW3oTrpsQAAwRDoAlq7pUvNq7cpc+iwJKmrN6Pm1dskiUgDAI7CJe4CamntHIzzgMyhw2pp7UxoIgBAVAS6gLp7h/9HL463DgAoXgS6gKrKh/+wkeOtAwCKF4EuoEVNtSqbkDpqrWxCSouaahOaCAAQFTeJFdDAjWDcxQ0AeDsEusDmNlQTZADA2+ISNwAAARFoAAACItAAAAREoAEACIhAAwAQEIEGACAgAg0AQEAEGgCAgAg0AAABEWgAAAIi0AAABESgAQAIiEADABAQgQYAICACDQBAQAQaAICACDQAAAERaAAAAiLQAAAERKABAAiIQAMAEBCBBgAgIAINAEBABBoAgIAINAAAARFoAAACItAAAAREoAEACIhAAwAQEIEGACAgAg0AQEAEGgCAgAg0AAABEWgAAAIi0AAABESgAQAIiEADABAQgQYAICACDQBAQAQaAICACDQAAAERaAAAAiLQAACcwJIlSzR16lTV1dWpvr5ezz77bEH2W5Krb2RmKUltkrrc/TozmyJppaRJkp6TdIu7v5Wr/QEAkG+bNm3SunXr9Nxzz6m0tFQ9PT16663CpCyXZ9B3S9oxZPsfJX3P3S+Q9Iak+TncFwAAebdnzx5VVFSotLRUklRRUaGqqirV1NSop6dHktTW1qbGxkZJ0uLFizVv3jw1NjbqvPPO04MPPviu952TQJvZZEnXSnoou22Srpa0KvuU5ZLm5mJfAAAUyuzZs/Xyyy/rwgsv1O23366nn376bV/z4osvqrW1VZs3b9b999+vQ4cOvat95+oM+vuS/l7Skez2GZJ63b0vu/2KpOoc7QsAgIKYOHGi2tvbtXTpUlVWVuqmm27SsmXLTviaa6+9VqWlpaqoqNCZZ56p11577V3te9TvQZvZdZJed/d2M2scWB7mqX6c1y+QtECSzjnnnNGOAwBATqVSKTU2NqqxsVHTp0/X8uXLVVJSoiNH+s9JDx48eNTzBy6HD7y2r69P70YuzqBnSppjZrvVf1PY1eo/oy43s4E/AEyW1D3ci919qbun3T1dWVmZg3EAAMiNzs5O7dy5c3C7o6ND5557rmpqatTe3i5JeuSRR/Ky71GfQbt7s6RmScqeQS9098+a2f+RdIP6o32rpEdHuy8AAArpwIEDuuuuu9Tb26uSkhKdf/75Wrp0qXbs2KH58+frgQce0GWXXZaXfZv7sFee3903+0ugrzOz8/SXv2a1RdL/dPc3T/T6dDrtbW1tOZsHAIDozKzd3dPHrufs70FLkrtvkLQh+/UfJF2ay+8PAECx4JPEAAAIiEADABAQgQYAIKCcvgcNAMB4tHZLl1paO9Xdm1FVeZkWNdVqbkN+P3+LQAMAcAJrt3SpefU2ZQ4dliR19WbUvHqbJOU10lziBgDgBFpaOwfjPCBz6LBaWjvzul8CDQDACXT3Zka0nisEGgCAE6gqLxvReq4QaAAATmBRU63KJqSOWiubkNKiptq87pebxAAAOIGBG8G4ixsAgGDmNlTnPcjH4hI3AAABEWgAAAIi0AAABESgAQAIiEADABAQgQYAICACDQBAQAQaAICACDQAAAERaAAAAiLQAAAERKABAAiIQAMAEBCBBgAgIAINAEBABBoAgIAINAAAARFoAAACItAAAAREoAEACIhAAwAQEIEGACAgAg0AQEAEGgCAgAg0AAABEWgAAAIi0AAABESgAQAIiEADABAQgQYAICACDQBAQAQaAICACDQAAAERaAAAAiLQAAAERKABAAiIQAMAEBCBBgAgIAINAEBABBoAgIAINAAAARFoAAACItAAAAREoAEACIhAAwAQEIEGACAgAg0AQEAEGihCZqZbbrllcLuvr0+VlZW67rrrEpwKwFAEGihCp5xyip5//nllMhlJ0hNPPKHq6uqEpwIwFIEGitTHPvYxPf7445Kkhx9+WJ/+9KcHH/vjH/+ouXPnqq6uTpdffrm2bt0qSVq8eLHmzZunxsZGnXfeeXrwwQcTmR0oBgQaKFI333yzVq5cqYMHD2rr1q267LLLBh+777771NDQoK1bt+qBBx7Q5z73ucHHXnzxRbW2tmrz5s26//77dejQoSTGB8Y9Ag0Uqbq6Ou3evVsPP/ywPv7xjx/12K9//evB96ivvvpq7du3T3/6058kSddee61KS0tVUVGhM888U6+99lrBZweKQUnSAwBIzpw5c7Rw4UJt2LBB+/btG1x39796rplJkkpLSwfXUqmU+vr68j8oUIQ4gwaK2Lx58/SNb3xD06dPP2r9qquu0ooVKyRJGzZsUEVFhd73vvclMSJQtDiDBorY5MmTdffdd//V+uLFi/WFL3xBdXV1Ovnkk7V8+fIEpgOKmw13KSsp6XTa29rakh4DAICCMbN2d08fu84lbgAAAiLQAAAERKABAAiIm8SAIrZ2S5daWjvV3ZtRVXmZFjXVam4DH/kJRECggSK1dkuXmldvU+bQYUlSV29Gzau3SRKRBgIY9SVuMzvbzJ4ysx1mtt3M7s6uTzKzJ8xsZ/bH00c/LoBcaWntHIzzgMyhw2pp7UxoIgBD5eI96D5JX3H3v5F0uaQ7zOwiSV+T9KS7XyDpyew2gCC6ezMjWgdQWKMOtLvvcffnsl/vl7RDUrWk6yUNfLrBcklzR7svALlTVV42onUAhZXTu7jNrEZSg6RnJb3f3fdI/RGXdOZxXrPAzNrMrG3v3r25HAfACSxqqlXZhNRRa2UTUlrUVJvQRACGylmgzWyipEck/Z27//c7fZ27L3X3tLunKysrczUOgLcxt6Fa//C301VdXiaTVF1epn/42+ncIAYEkZO7uM1sgvrjvMLdV2eXXzOzs9x9j5mdJen1XOwLQO7MbagmyEBQubiL2yT9WNIOd//ukIcek3Rr9utbJT062n0BAFAscnEGPVPSLZK2mVlHdu1eSd+W9DMzmy/pJUk35mBfAAAUhVEH2t1/LcmO8/Cs0X5/AACKEZ/FDQBAQAQaAICACDQAAAERaAAAAiLQAAAERKABAAiIQAMAEBCBBgAgIAINAEBABBoAgIAINAAAARFoAAACItAAAAREoAEACIhAAwAQEIEGACAgAg0AQEAEGgCAgAg0AAABEWgAAAIi0AAABESgAQAIiEADABAQgQYAICACDQBAQAQaAICACDQAAAERaAAAAiLQAAAERKABAAiIQAMAEBCBBgAgIAINAEBABBoAgIAINAAAARFoAAACItAAAAREoAEACIhAAwAQEIEGACAgAg0AQEAEGgCAgAg0AAABEWgAAAIad4FOpVKqr68f/G/37t1/9Zzu7m7dcMMNhR8OAIB3qCTpAXKtrKxMHR0dx328r69PVVVVWrVqVQGnAgBgZMbdGfRwli1bphtvvFGf+MQnNHv2bO3evVvTpk1LeiwAAI5r3J1BZzIZ1dfXS5KmTJmiNWvWSJI2bdqkrVu3atKkScNe9gYAIJJxF+jjXeL+6Ec/qkmTJiUwEQAAI1cUl7gl6ZRTTkl6BAAA3rGiCTQAAGMJgQYAICBz96RnGJROp72trS3pMQAAKBgza3f39LHrnEEDABAQgQYAICACDQBAQAQaAICAxt0HlQxYu6VLLa2d6u7NqKq8TIuaajW3oTrpsQAAeEfGZaDXbulS8+ptyhw6LEnq6s2oefU2SSLSAIAxYVxe4m5p7RyM84DMocNqae1MaCIAAEZmXAa6uzczonUAAKIZl4GuKi8b0ToAANGMy0AvaqpV2YTUUWtlE1Ja1FSb0EQAAIzMuLxJbOBGMO7iBgCMVeMy0FJ/pAkyAGCsGpeXuAEAGOsINAAAARFoAAACItAAAAREoAEACIhAAwAQUN4DbWbXmFmnme0ys6/le38AAIwHeQ20maUk/bOkj0m6SNKnzeyifO4TAIDxIN9n0JdK2uXuf3D3tyStlHR9nvcJAMCYl+9AV0t6ecj2K9k1AABwAvkOtA2z5kc9wWyBmbWZWdvevXvzPA4AAGNDvgP9iqSzh2xPltQ99AnuvtTd0+6erqyszPM4AACMDfkO9G8lXWBmU8zsJEk3S3osz/sEAGDMy+u/ZuXufWZ2p6RWSSlJP3H37fncJwAA40He/7lJd/+FpF/kez8AAIwnfJIYAAABEWgAAAIi0AAABESgAQAIiEADABAQgQYAICACDQBAQAQaAICACDQAAAERaAAAAiLQAAAERKABAAiIQAMAEBCBBgAgIAINAEBABBoAgIAINAAAARFoAAACItAAAAREoAEACIhAAwAQEIEGACAgAg0AQEAEGgCAgAg0AAABEWgAAAIi0AAABESgAQCJuueee/T9739/cLupqUlf/OIXB7e/8pWv6Lvf/e47+l6LFy/Wd77znZzPmAQCDQBI1JVXXqmNGzdKko4cOaKenh5t37598PGNGzdq5syZSY2XGAINAEjUzJkzBwO9fft2TZs2TaeeeqreeOMNvfnmm9qxY4caGhrU0tKiGTNmqK6uTvfdd9/g65csWaLa2lp95CMfUWdnZ1I/jZwrSXoAAEBxq6qqUklJiV566SVt3LhRV1xxhbq6urRp0yaddtppqqur04YNG7Rz505t3rxZ7q45c+boV7/6lU455RStXLlSW7ZsUV9fny6++GJdcsklSf+UcoJAAwASN3AWvXHjRn35y19WV1eXNm7cqNNOO01XXnml1q9fr/Xr16uhoUGSdODAAe3cuVP79+/XJz/5SZ188smSpDlz5iT508gpLnEDABI38D70tm3bNG3aNF1++eXatGnT4PvP7q7m5mZ1dHSoo6NDu3bt0vz58yVJZpbw9PlBoAEAiZs5c6bWrVunSZMmKZVKadKkSert7dWmTZt0xRVXqKmpST/5yU904MABSVJXV5def/11XXXVVVqzZo0ymYz279+vn//85wn/THKHS9wAgMRNnz5dPT09+sxnPnPU2oEDB1RRUaHZs2drx44duuKKKyRJEydO1E9/+lNdfPHFuummm1RfX69zzz1XH/7wh5P6KeScuXvSMwxKp9Pe1taW9BgAABSMmbW7e/rYdS5xAwAQEIEGACAgAg0AQEDcJAYACGPtli61tHaquzejqvIyLWqq1dyG6qTHSgSBBgCEsHZLl5pXb1Pm0GFJUldvRs2rt0lSUUaaS9wAgBBaWjsH4zwgc+iwWlrHz+drjwSBBgCE0N2bGdH6eEegAQAhVJWXjWh9vCPQAIAQFjXVqmxC6qi1sgkpLWqqTWiiZHGTGAAghIEbwbiLux+BBgCEMbehumiDfCwucQMAEBCBBgAgIAINAEBABBoAgIAINAAAARFoAAACItAAAAREoAEACIhAAwAQEIEGACAgAg0AQEAEGgCAgAg0AAABEWgAAAIi0AAABESgAQAIiEADABAQgQYAICACDQBAQAQaAICACDQAAAERaAAAAiLQAAAERKABAAiIQAMAENCoAm1mLWb2opltNbM1ZlY+5LFmM9tlZp1m1jT6UQEAKB6jPYN+QtI0d6+T9HtJzZJkZhdJulnSVEnXSPoXM0uNcl8AABSNUQXa3de7e1928xlJk7NfXy9ppbu/6e7/KWmXpEtHsy8AAIpJLt+Dnifp37NfV0t6echjr2TXAADAO1Dydk8ws/+Q9IFhHvq6uz+afc7XJfVJWjHwsmGe78f5/gskLZCkc8455x2MDADA+Pe2gXb3j5zocTO7VdJ1kma5+0CEX5F09pCnTZbUfZzvv1TSUklKp9PDRhwAgGIz2ru4r5H0VUlz3P3PQx56TNLNZlZqZlMkXSBp82j2BQBAMXnbM+i38U+SSiU9YWaS9Iy73+bu283sZ5JeUP+l7zvc/fAo9wUAQNEYVaDd/fwTPLZE0pLRfH8AAIoVnyQGAEBABBoAgIAINAAAARFoAAACItAAAAREoAEACIhAAwAQEIEGACAgAg0AQEAEGgCAgAg0AAABEWgAAAIi0AAABDTaf24SwHHs27dPs2bNkiS9+uqrSqVSqqys1O7du1VVVaUXXngh4QkBRMYZNJAnZ5xxhjo6OtTR0aHbbrtN99xzz+D2e97D/3oATozfJYAEHD58WF/60pc0depUzZ49W5lMRpLU2NiotrY2SVJPT49qamoSnBJAkgg0kICdO3fqjjvu0Pbt21VeXq5HHnkk6ZEABEOggQRMmTJF9fX1kqRLLrlEu3fvTnYgAOEQaCABpaWlg1+nUin19fVJkkpKSnTkyBFJ0sGDBxOZDUAMBBoIpKamRu3t7ZKkVatWJTwNgCQRaCCQhQsX6gc/+IGuvPJK9fT0JD0OgASZuyc9w6B0Ou0Dd7ACAFAMzKzd3dPHrnMGDQBAQAQaAICACDQAAAERaAAAAuIfywAKYO2WLrW0dqq7N6Oq8jItaqrV3IbqpMcCEBiBBvJs7ZYuNa/epsyhw5Kkrt6MmldvkyQiDeC4uMQN5FlLa+dgnAdkDh1WS2tnQhMBGAsINJBn3b2ZEa0DgESggbyrKi8b0ToASAQayLtFTbUqm5A6aq1sQkqLmmoTmgjAWMBNYkCeDdwIxl3cAEaCQAMFMLehmiADGBEucQMAEBCBBgAgIAINAEBABBoAgIAINAAAARFoAAACItAAAAREoAEACIhAAwAQEIEGACAgAg0AQEAEGgCAgAg0AAABEWgAAAIi0AAABGTunvQMg8xsr6T/SnqOEaqQ1JP0EOMUxzY/OK75w7HNj/F+XM9198pjF0MFeiwyszZ3Tyc9x3jEsc0Pjmv+cGzzo1iPK5e4AQAIiEADABAQgR69pUkPMI5xbPOD45o/HNv8KMrjynvQAAAExBk0AAABEehRMrOFZuZmVpHdNjN70Mx2mdlWM7s46RnHGjNrMbMXs8dvjZmVD3msOXtsO82sKck5xyIzuyZ77HaZ2deSnmesMrOzzewpM9thZtvN7O7s+iQze8LMdmZ/PD3pWccqM0uZ2RYzW5fdnmJmz2aP7b+Z2UlJz5hvBHoUzOxsSR+V9NKQ5Y9JuiD73wJJP0hgtLHuCUnT3L1O0u8lNUuSmV0k6WZJUyVdI+lfzCyV2JRjTPZY/bP6f41eJOnT2WOKkeuT9BV3/xtJl0u6I3ssvybpSXe/QNKT2W28O3dL2jFk+x8lfS97bN+QND+RqQqIQI/O9yT9vaShb+RfL+lfvd8zksrN7KxEphuj3H29u/dlN5+RNDn79fWSVrr7m+7+n5J2Sbo0iRnHqEsl7XL3P7j7W5JWqv+YYoTcfY+7P5f9er/6Q1Kt/uO5PPu05ZLmJjPh2GZmkyVdK+mh7LZJulrSquxTiuLYEuh3yczmSOpy998d81C1pJeHbL+SXcO7M0/Sv2e/5tiODscvD8ysRlKDpGclvd/d90j9EZd0ZnKTjWnfV//Jz5Hs9hmSeof8wb0ofu2WJD1AZGb2H5I+MMxDX5d0r6TZw71smDVulT/GiY6tuz+afc7X1X8pccXAy4Z5Psf2neP45ZiZTZT0iKS/c/f/7j/Rw2iY2XWSXnf3djNrHFge5qnj/tcugT4Bd//IcOtmNl3SFEm/y/4POVnSc2Z2qfr/ZHf2kKdPltSd51HHnOMd2wFmdquk6yTN8r/8XUCO7ehw/HLIzCaoP84r3H11dvk1MzvL3fdk39p6PbkJx6yZkuaY2cclvVfS+9R/Rl1uZiXZs+ii+LXLJe53wd23ufuZ7l7j7jXq/43vYnd/VdJjkj6XvZv7ckl/GrjkhXfGzK6R9FVJc9z9z0MeekzSzWZWamZT1H8j3uYkZhyjfivpguzdsCep/4a7xxKeaUzKvif6Y0k73P27Qx56TNKt2a9vlfRooWcb69y92d0nZ39vvVnSL939s5KeknRD9mlFcWw5g869X0j6uPpvYPqzpC8kO86Y9E+SSiU9kb1C8Yy73+bu283sZ5JeUP+l7zvc/XCCc44p7t5nZndKapWUkvQTd9+e8Fhj1UxJt0jaZmYd2bV7JX1b0s/MbL76/3bHjQnNNx59VdJKM/uWpC3q/wPSuMYniQEAEBCXuAEACIhAAwAQEIEGACAgAg0AQEAEGgCAgAg0AAABEWgAAAIi0AAABPT/AV99XgrFt5jlAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 576x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "tsne = TSNE(n_components=2, random_state=1, learning_rate=10, n_iter=10000)\n",
    "Y = tsne.fit_transform(day_of_week_embedding)\n",
    "names = ['Mon', 'Tue', 'Wed', 'Thu', 'Fri', 'Sat','Sun']\n",
    "plt.figure(figsize=(8,8))\n",
    "plt.scatter(-Y[:, 0], -Y[:, 1])\n",
    "for i, txt in enumerate(names):\n",
    "    plt.annotate(txt, (-Y[i, 0],-Y[i, 1]), xytext = (-5, 8), textcoords = 'offset points')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['Store',\n",
       " 'DayOfWeek',\n",
       " 'Year',\n",
       " 'Month',\n",
       " 'Day',\n",
       " 'StateHoliday',\n",
       " 'CompetitionMonthsOpen',\n",
       " 'Promo2Weeks',\n",
       " 'StoreType',\n",
       " 'Assortment',\n",
       " 'PromoInterval',\n",
       " 'CompetitionOpenSinceYear',\n",
       " 'Promo2SinceYear',\n",
       " 'State',\n",
       " 'Week',\n",
       " 'Events',\n",
       " 'Promo_fw',\n",
       " 'Promo_bw',\n",
       " 'StateHoliday_fw',\n",
       " 'StateHoliday_bw',\n",
       " 'SchoolHoliday_fw',\n",
       " 'SchoolHoliday_bw']"
      ]
     },
     "execution_count": 93,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cat_vars"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Store embeddings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [],
   "source": [
    "submodel = Model(cat_var_inputs + cont_vars_inputs, merged)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_3\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "Store_input (InputLayer)        [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "DayOfWeek_input (InputLayer)    [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Year_input (InputLayer)         [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Month_input (InputLayer)        [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Day_input (InputLayer)          [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "StateHoliday_input (InputLayer) [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "CompetitionMonthsOpen_input (In [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Promo2Weeks_input (InputLayer)  [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "StoreType_input (InputLayer)    [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Assortment_input (InputLayer)   [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "PromoInterval_input (InputLayer [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "CompetitionOpenSinceYear_input  [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Promo2SinceYear_input (InputLay [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "State_input (InputLayer)        [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Week_input (InputLayer)         [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Events_input (InputLayer)       [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Promo_fw_input (InputLayer)     [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Promo_bw_input (InputLayer)     [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "StateHoliday_fw_input (InputLay [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "StateHoliday_bw_input (InputLay [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "SchoolHoliday_fw_input (InputLa [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "SchoolHoliday_bw_input (InputLa [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Store_Embed (Embedding)         (None, 1, 50)        55750       Store_input[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "DayOfWeek_Embed (Embedding)     (None, 1, 2)         14          DayOfWeek_input[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "Year_Embed (Embedding)          (None, 1, 2)         6           Year_input[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "Month_Embed (Embedding)         (None, 1, 6)         72          Month_input[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "Day_Embed (Embedding)           (None, 1, 10)        310         Day_input[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "StateHoliday_Embed (Embedding)  (None, 1, 3)         6           StateHoliday_input[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "CompetitionMonthsOpen_Embed (Em (None, 1, 2)         50          CompetitionMonthsOpen_input[0][0]\n",
      "__________________________________________________________________________________________________\n",
      "Promo2Weeks_Embed (Embedding)   (None, 1, 1)         26          Promo2Weeks_input[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "StoreType_Embed (Embedding)     (None, 1, 2)         8           StoreType_input[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "Assortment_Embed (Embedding)    (None, 1, 3)         9           Assortment_input[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "PromoInterval_Embed (Embedding) (None, 1, 3)         12          PromoInterval_input[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "CompetitionOpenSinceYear_Embed  (None, 1, 4)         84          CompetitionOpenSinceYear_input[0]\n",
      "__________________________________________________________________________________________________\n",
      "Promo2SinceYear_Embed (Embeddin (None, 1, 4)         32          Promo2SinceYear_input[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "State_Embed (Embedding)         (None, 1, 6)         72          State_input[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "Week_Embed (Embedding)          (None, 1, 2)         104         Week_input[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "Events_Embed (Embedding)        (None, 1, 4)         88          Events_input[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "Promo_fw_Embed (Embedding)      (None, 1, 1)         6           Promo_fw_input[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "Promo_bw_Embed (Embedding)      (None, 1, 1)         6           Promo_bw_input[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "StateHoliday_fw_Embed (Embeddin (None, 1, 1)         3           StateHoliday_fw_input[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "StateHoliday_bw_Embed (Embeddin (None, 1, 1)         3           StateHoliday_bw_input[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "SchoolHoliday_fw_Embed (Embeddi (None, 1, 1)         8           SchoolHoliday_fw_input[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "SchoolHoliday_bw_Embed (Embeddi (None, 1, 1)         8           SchoolHoliday_bw_input[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "Store_flat (Flatten)            (None, 50)           0           Store_Embed[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "DayOfWeek_flat (Flatten)        (None, 2)            0           DayOfWeek_Embed[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "Year_flat (Flatten)             (None, 2)            0           Year_Embed[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "Month_flat (Flatten)            (None, 6)            0           Month_Embed[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "Day_flat (Flatten)              (None, 10)           0           Day_Embed[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "StateHoliday_flat (Flatten)     (None, 3)            0           StateHoliday_Embed[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "CompetitionMonthsOpen_flat (Fla (None, 2)            0           CompetitionMonthsOpen_Embed[0][0]\n",
      "__________________________________________________________________________________________________\n",
      "Promo2Weeks_flat (Flatten)      (None, 1)            0           Promo2Weeks_Embed[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "StoreType_flat (Flatten)        (None, 2)            0           StoreType_Embed[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "Assortment_flat (Flatten)       (None, 3)            0           Assortment_Embed[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "PromoInterval_flat (Flatten)    (None, 3)            0           PromoInterval_Embed[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "CompetitionOpenSinceYear_flat ( (None, 4)            0           CompetitionOpenSinceYear_Embed[0]\n",
      "__________________________________________________________________________________________________\n",
      "Promo2SinceYear_flat (Flatten)  (None, 4)            0           Promo2SinceYear_Embed[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "State_flat (Flatten)            (None, 6)            0           State_Embed[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "Week_flat (Flatten)             (None, 2)            0           Week_Embed[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "Events_flat (Flatten)           (None, 4)            0           Events_Embed[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "Promo_fw_flat (Flatten)         (None, 1)            0           Promo_fw_Embed[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "Promo_bw_flat (Flatten)         (None, 1)            0           Promo_bw_Embed[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "StateHoliday_fw_flat (Flatten)  (None, 1)            0           StateHoliday_fw_Embed[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "StateHoliday_bw_flat (Flatten)  (None, 1)            0           StateHoliday_bw_Embed[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "SchoolHoliday_fw_flat (Flatten) (None, 1)            0           SchoolHoliday_fw_Embed[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "SchoolHoliday_bw_flat (Flatten) (None, 1)            0           SchoolHoliday_bw_Embed[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "CompetitionDistance_input (Inpu [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Max_TemperatureC_input (InputLa [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Mean_TemperatureC_input (InputL [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Min_TemperatureC_input (InputLa [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Max_Humidity_input (InputLayer) [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Mean_Humidity_input (InputLayer [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Min_Humidity_input (InputLayer) [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Max_Wind_SpeedKm_h_input (Input [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Mean_Wind_SpeedKm_h_input (Inpu [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "CloudCover_input (InputLayer)   [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "trend_input (InputLayer)        [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "trend_DE_input (InputLayer)     [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "AfterStateHoliday_input (InputL [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "BeforeStateHoliday_input (Input [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "Promo_input (InputLayer)        [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "SchoolHoliday_input (InputLayer [(None, 1)]          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "All_Concatenate (Concatenate)   (None, 126)          0           Store_flat[0][0]                 \n",
      "                                                                 DayOfWeek_flat[0][0]             \n",
      "                                                                 Year_flat[0][0]                  \n",
      "                                                                 Month_flat[0][0]                 \n",
      "                                                                 Day_flat[0][0]                   \n",
      "                                                                 StateHoliday_flat[0][0]          \n",
      "                                                                 CompetitionMonthsOpen_flat[0][0] \n",
      "                                                                 Promo2Weeks_flat[0][0]           \n",
      "                                                                 StoreType_flat[0][0]             \n",
      "                                                                 Assortment_flat[0][0]            \n",
      "                                                                 PromoInterval_flat[0][0]         \n",
      "                                                                 CompetitionOpenSinceYear_flat[0][\n",
      "                                                                 Promo2SinceYear_flat[0][0]       \n",
      "                                                                 State_flat[0][0]                 \n",
      "                                                                 Week_flat[0][0]                  \n",
      "                                                                 Events_flat[0][0]                \n",
      "                                                                 Promo_fw_flat[0][0]              \n",
      "                                                                 Promo_bw_flat[0][0]              \n",
      "                                                                 StateHoliday_fw_flat[0][0]       \n",
      "                                                                 StateHoliday_bw_flat[0][0]       \n",
      "                                                                 SchoolHoliday_fw_flat[0][0]      \n",
      "                                                                 SchoolHoliday_bw_flat[0][0]      \n",
      "                                                                 CompetitionDistance_input[0][0]  \n",
      "                                                                 Max_TemperatureC_input[0][0]     \n",
      "                                                                 Mean_TemperatureC_input[0][0]    \n",
      "                                                                 Min_TemperatureC_input[0][0]     \n",
      "                                                                 Max_Humidity_input[0][0]         \n",
      "                                                                 Mean_Humidity_input[0][0]        \n",
      "                                                                 Min_Humidity_input[0][0]         \n",
      "                                                                 Max_Wind_SpeedKm_h_input[0][0]   \n",
      "                                                                 Mean_Wind_SpeedKm_h_input[0][0]  \n",
      "                                                                 CloudCover_input[0][0]           \n",
      "                                                                 trend_input[0][0]                \n",
      "                                                                 trend_DE_input[0][0]             \n",
      "                                                                 AfterStateHoliday_input[0][0]    \n",
      "                                                                 BeforeStateHoliday_input[0][0]   \n",
      "                                                                 Promo_input[0][0]                \n",
      "                                                                 SchoolHoliday_input[0][0]        \n",
      "==================================================================================================\n",
      "Total params: 56,677\n",
      "Trainable params: 56,677\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "submodel.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "submodel.save('embeddings_model.hdf5')"
   ]
  },
  {
   "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": "3.6.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}