outlier_detection/find_threshold.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensorflow       2.4.0\n",
      "json             2.0.9\n",
      "tensorflow.keras 2.4.0\n",
      "tensorflow_hub   0.9.0\n",
      "pandas           1.0.5\n",
      "autopep8         1.5.4\n",
      "numpy            1.19.5\n",
      "CPython 3.7.4\n",
      "IPython 7.8.0\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import os\n",
    "import pandas as pd\n",
    "import tensorflow as tf\n",
    "import tensorflow.keras as keras\n",
    "import tensorflow_hub as hub\n",
    "from IPython.core.interactiveshell import InteractiveShell\n",
    "InteractiveShell.ast_node_interactivity = \"all\"\n",
    "\n",
    "# enable gpu growth if gpu is available\n",
    "gpu_devices = tf.config.experimental.list_physical_devices('GPU')\n",
    "for device in gpu_devices:\n",
    "    tf.config.experimental.set_memory_growth(device, True)\n",
    "\n",
    "# tf.keras.mixed_precision.set_global_policy('mixed_float16')\n",
    "\n",
    "tf.config.optimizer.set_jit(True)\n",
    "\n",
    "%load_ext watermark\n",
    "%watermark -v -iv"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# The Data is taken from https://www.kaggle.com/mkechinov/ecommerce-behavior-data-from-multi-category-store and https://rees46.com/\n",
    "## Each record/line in the file has the following fields:\n",
    "1. event_time: When did the event happened (UTC)\n",
    "2. event_type: Event type: one of [view, cart, remove_from_cart, purchase] \n",
    "3. product_id\n",
    "4. category_id\n",
    "5. category_code: Category meaningful name (if present)\n",
    "6. brand: Brand name in lower case (if present)\n",
    "7. price\n",
    "8. user_id: Permanent user ID\n",
    "9. user_session: User session ID"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean: 284.77105468660056\n",
      "Std: 349.4674023158121\n",
      "Sessions: (61296,)\n",
      "Unique Products: (38515,)\n",
      "Unique category_code: (134,)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:58: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n"
     ]
    },
    {
     "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>embedding_0</th>\n",
       "      <th>embedding_1</th>\n",
       "      <th>embedding_2</th>\n",
       "      <th>embedding_3</th>\n",
       "      <th>embedding_4</th>\n",
       "      <th>embedding_5</th>\n",
       "      <th>embedding_6</th>\n",
       "      <th>embedding_7</th>\n",
       "      <th>embedding_8</th>\n",
       "      <th>embedding_9</th>\n",
       "      <th>...</th>\n",
       "      <th>embedding_46</th>\n",
       "      <th>embedding_47</th>\n",
       "      <th>embedding_48</th>\n",
       "      <th>embedding_49</th>\n",
       "      <th>price_standardized</th>\n",
       "      <th>user_session</th>\n",
       "      <th>Product</th>\n",
       "      <th>product_id</th>\n",
       "      <th>category_code</th>\n",
       "      <th>product_id_mapped</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.047610</td>\n",
       "      <td>-0.125734</td>\n",
       "      <td>-0.053261</td>\n",
       "      <td>0.196848</td>\n",
       "      <td>-0.016433</td>\n",
       "      <td>0.049773</td>\n",
       "      <td>0.012852</td>\n",
       "      <td>0.263229</td>\n",
       "      <td>-0.195415</td>\n",
       "      <td>0.327227</td>\n",
       "      <td>...</td>\n",
       "      <td>0.090539</td>\n",
       "      <td>0.104421</td>\n",
       "      <td>0.061444</td>\n",
       "      <td>-0.008996</td>\n",
       "      <td>-0.373028</td>\n",
       "      <td>0000afb3-2d30-4b52-84ec-07c6617efd37</td>\n",
       "      <td>light</td>\n",
       "      <td>1004838</td>\n",
       "      <td>construction.tools.light</td>\n",
       "      <td>311.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.047610</td>\n",
       "      <td>-0.125734</td>\n",
       "      <td>-0.053261</td>\n",
       "      <td>0.196848</td>\n",
       "      <td>-0.016433</td>\n",
       "      <td>0.049773</td>\n",
       "      <td>0.012852</td>\n",
       "      <td>0.263229</td>\n",
       "      <td>-0.195415</td>\n",
       "      <td>0.327227</td>\n",
       "      <td>...</td>\n",
       "      <td>0.090539</td>\n",
       "      <td>0.104421</td>\n",
       "      <td>0.061444</td>\n",
       "      <td>-0.008996</td>\n",
       "      <td>-0.373028</td>\n",
       "      <td>0000afb3-2d30-4b52-84ec-07c6617efd37</td>\n",
       "      <td>light</td>\n",
       "      <td>1004838</td>\n",
       "      <td>construction.tools.light</td>\n",
       "      <td>311.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.047610</td>\n",
       "      <td>-0.125734</td>\n",
       "      <td>-0.053261</td>\n",
       "      <td>0.196848</td>\n",
       "      <td>-0.016433</td>\n",
       "      <td>0.049773</td>\n",
       "      <td>0.012852</td>\n",
       "      <td>0.263229</td>\n",
       "      <td>-0.195415</td>\n",
       "      <td>0.327227</td>\n",
       "      <td>...</td>\n",
       "      <td>0.090539</td>\n",
       "      <td>0.104421</td>\n",
       "      <td>0.061444</td>\n",
       "      <td>-0.008996</td>\n",
       "      <td>0.068930</td>\n",
       "      <td>0000b83c-9b26-4881-8bca-e20d460f4194</td>\n",
       "      <td>light</td>\n",
       "      <td>1005252</td>\n",
       "      <td>construction.tools.light</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.047610</td>\n",
       "      <td>-0.125734</td>\n",
       "      <td>-0.053261</td>\n",
       "      <td>0.196848</td>\n",
       "      <td>-0.016433</td>\n",
       "      <td>0.049773</td>\n",
       "      <td>0.012852</td>\n",
       "      <td>0.263229</td>\n",
       "      <td>-0.195415</td>\n",
       "      <td>0.327227</td>\n",
       "      <td>...</td>\n",
       "      <td>0.090539</td>\n",
       "      <td>0.104421</td>\n",
       "      <td>0.061444</td>\n",
       "      <td>-0.008996</td>\n",
       "      <td>0.289895</td>\n",
       "      <td>0000b83c-9b26-4881-8bca-e20d460f4194</td>\n",
       "      <td>light</td>\n",
       "      <td>1004503</td>\n",
       "      <td>construction.tools.light</td>\n",
       "      <td>1259.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.047610</td>\n",
       "      <td>-0.125734</td>\n",
       "      <td>-0.053261</td>\n",
       "      <td>0.196848</td>\n",
       "      <td>-0.016433</td>\n",
       "      <td>0.049773</td>\n",
       "      <td>0.012852</td>\n",
       "      <td>0.263229</td>\n",
       "      <td>-0.195415</td>\n",
       "      <td>0.327227</td>\n",
       "      <td>...</td>\n",
       "      <td>0.090539</td>\n",
       "      <td>0.104421</td>\n",
       "      <td>0.061444</td>\n",
       "      <td>-0.008996</td>\n",
       "      <td>-0.291904</td>\n",
       "      <td>0000f7c4-8836-4507-82a1-8a10de3fb1b2</td>\n",
       "      <td>light</td>\n",
       "      <td>1005191</td>\n",
       "      <td>construction.tools.light</td>\n",
       "      <td>2.0</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",
       "      <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",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>369299</th>\n",
       "      <td>0.174397</td>\n",
       "      <td>-0.204014</td>\n",
       "      <td>-0.175919</td>\n",
       "      <td>0.105906</td>\n",
       "      <td>-0.205940</td>\n",
       "      <td>-0.210022</td>\n",
       "      <td>0.224095</td>\n",
       "      <td>0.210598</td>\n",
       "      <td>-0.162613</td>\n",
       "      <td>0.150693</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.129293</td>\n",
       "      <td>-0.176801</td>\n",
       "      <td>0.100340</td>\n",
       "      <td>0.119850</td>\n",
       "      <td>-0.630791</td>\n",
       "      <td>fffdfd5e-126c-409f-9c16-8224f22cb60b</td>\n",
       "      <td>cooler</td>\n",
       "      <td>4400467</td>\n",
       "      <td>computers.components.cooler</td>\n",
       "      <td>4002.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>369300</th>\n",
       "      <td>0.047610</td>\n",
       "      <td>-0.125734</td>\n",
       "      <td>-0.053261</td>\n",
       "      <td>0.196848</td>\n",
       "      <td>-0.016433</td>\n",
       "      <td>0.049773</td>\n",
       "      <td>0.012852</td>\n",
       "      <td>0.263229</td>\n",
       "      <td>-0.195415</td>\n",
       "      <td>0.327227</td>\n",
       "      <td>...</td>\n",
       "      <td>0.090539</td>\n",
       "      <td>0.104421</td>\n",
       "      <td>0.061444</td>\n",
       "      <td>-0.008996</td>\n",
       "      <td>-0.511696</td>\n",
       "      <td>fffe34dd-9537-4991-9f12-d81f1dda91cb</td>\n",
       "      <td>light</td>\n",
       "      <td>1004903</td>\n",
       "      <td>construction.tools.light</td>\n",
       "      <td>137.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>369301</th>\n",
       "      <td>0.047610</td>\n",
       "      <td>-0.125734</td>\n",
       "      <td>-0.053261</td>\n",
       "      <td>0.196848</td>\n",
       "      <td>-0.016433</td>\n",
       "      <td>0.049773</td>\n",
       "      <td>0.012852</td>\n",
       "      <td>0.263229</td>\n",
       "      <td>-0.195415</td>\n",
       "      <td>0.327227</td>\n",
       "      <td>...</td>\n",
       "      <td>0.090539</td>\n",
       "      <td>0.104421</td>\n",
       "      <td>0.061444</td>\n",
       "      <td>-0.008996</td>\n",
       "      <td>-0.511696</td>\n",
       "      <td>fffe34dd-9537-4991-9f12-d81f1dda91cb</td>\n",
       "      <td>light</td>\n",
       "      <td>1004903</td>\n",
       "      <td>construction.tools.light</td>\n",
       "      <td>137.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>369302</th>\n",
       "      <td>0.047610</td>\n",
       "      <td>-0.125734</td>\n",
       "      <td>-0.053261</td>\n",
       "      <td>0.196848</td>\n",
       "      <td>-0.016433</td>\n",
       "      <td>0.049773</td>\n",
       "      <td>0.012852</td>\n",
       "      <td>0.263229</td>\n",
       "      <td>-0.195415</td>\n",
       "      <td>0.327227</td>\n",
       "      <td>...</td>\n",
       "      <td>0.090539</td>\n",
       "      <td>0.104421</td>\n",
       "      <td>0.061444</td>\n",
       "      <td>-0.008996</td>\n",
       "      <td>-0.459760</td>\n",
       "      <td>fffe34dd-9537-4991-9f12-d81f1dda91cb</td>\n",
       "      <td>light</td>\n",
       "      <td>1004856</td>\n",
       "      <td>construction.tools.light</td>\n",
       "      <td>66.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>369303</th>\n",
       "      <td>0.047610</td>\n",
       "      <td>-0.125734</td>\n",
       "      <td>-0.053261</td>\n",
       "      <td>0.196848</td>\n",
       "      <td>-0.016433</td>\n",
       "      <td>0.049773</td>\n",
       "      <td>0.012852</td>\n",
       "      <td>0.263229</td>\n",
       "      <td>-0.195415</td>\n",
       "      <td>0.327227</td>\n",
       "      <td>...</td>\n",
       "      <td>0.090539</td>\n",
       "      <td>0.104421</td>\n",
       "      <td>0.061444</td>\n",
       "      <td>-0.008996</td>\n",
       "      <td>-0.459760</td>\n",
       "      <td>fffe34dd-9537-4991-9f12-d81f1dda91cb</td>\n",
       "      <td>light</td>\n",
       "      <td>1004856</td>\n",
       "      <td>construction.tools.light</td>\n",
       "      <td>66.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>369304 rows × 56 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        embedding_0  embedding_1  embedding_2  embedding_3  embedding_4  \\\n",
       "0          0.047610    -0.125734    -0.053261     0.196848    -0.016433   \n",
       "1          0.047610    -0.125734    -0.053261     0.196848    -0.016433   \n",
       "2          0.047610    -0.125734    -0.053261     0.196848    -0.016433   \n",
       "3          0.047610    -0.125734    -0.053261     0.196848    -0.016433   \n",
       "4          0.047610    -0.125734    -0.053261     0.196848    -0.016433   \n",
       "...             ...          ...          ...          ...          ...   \n",
       "369299     0.174397    -0.204014    -0.175919     0.105906    -0.205940   \n",
       "369300     0.047610    -0.125734    -0.053261     0.196848    -0.016433   \n",
       "369301     0.047610    -0.125734    -0.053261     0.196848    -0.016433   \n",
       "369302     0.047610    -0.125734    -0.053261     0.196848    -0.016433   \n",
       "369303     0.047610    -0.125734    -0.053261     0.196848    -0.016433   \n",
       "\n",
       "        embedding_5  embedding_6  embedding_7  embedding_8  embedding_9  ...  \\\n",
       "0          0.049773     0.012852     0.263229    -0.195415     0.327227  ...   \n",
       "1          0.049773     0.012852     0.263229    -0.195415     0.327227  ...   \n",
       "2          0.049773     0.012852     0.263229    -0.195415     0.327227  ...   \n",
       "3          0.049773     0.012852     0.263229    -0.195415     0.327227  ...   \n",
       "4          0.049773     0.012852     0.263229    -0.195415     0.327227  ...   \n",
       "...             ...          ...          ...          ...          ...  ...   \n",
       "369299    -0.210022     0.224095     0.210598    -0.162613     0.150693  ...   \n",
       "369300     0.049773     0.012852     0.263229    -0.195415     0.327227  ...   \n",
       "369301     0.049773     0.012852     0.263229    -0.195415     0.327227  ...   \n",
       "369302     0.049773     0.012852     0.263229    -0.195415     0.327227  ...   \n",
       "369303     0.049773     0.012852     0.263229    -0.195415     0.327227  ...   \n",
       "\n",
       "        embedding_46  embedding_47  embedding_48  embedding_49  \\\n",
       "0           0.090539      0.104421      0.061444     -0.008996   \n",
       "1           0.090539      0.104421      0.061444     -0.008996   \n",
       "2           0.090539      0.104421      0.061444     -0.008996   \n",
       "3           0.090539      0.104421      0.061444     -0.008996   \n",
       "4           0.090539      0.104421      0.061444     -0.008996   \n",
       "...              ...           ...           ...           ...   \n",
       "369299     -0.129293     -0.176801      0.100340      0.119850   \n",
       "369300      0.090539      0.104421      0.061444     -0.008996   \n",
       "369301      0.090539      0.104421      0.061444     -0.008996   \n",
       "369302      0.090539      0.104421      0.061444     -0.008996   \n",
       "369303      0.090539      0.104421      0.061444     -0.008996   \n",
       "\n",
       "        price_standardized                          user_session  Product  \\\n",
       "0                -0.373028  0000afb3-2d30-4b52-84ec-07c6617efd37    light   \n",
       "1                -0.373028  0000afb3-2d30-4b52-84ec-07c6617efd37    light   \n",
       "2                 0.068930  0000b83c-9b26-4881-8bca-e20d460f4194    light   \n",
       "3                 0.289895  0000b83c-9b26-4881-8bca-e20d460f4194    light   \n",
       "4                -0.291904  0000f7c4-8836-4507-82a1-8a10de3fb1b2    light   \n",
       "...                    ...                                   ...      ...   \n",
       "369299           -0.630791  fffdfd5e-126c-409f-9c16-8224f22cb60b   cooler   \n",
       "369300           -0.511696  fffe34dd-9537-4991-9f12-d81f1dda91cb    light   \n",
       "369301           -0.511696  fffe34dd-9537-4991-9f12-d81f1dda91cb    light   \n",
       "369302           -0.459760  fffe34dd-9537-4991-9f12-d81f1dda91cb    light   \n",
       "369303           -0.459760  fffe34dd-9537-4991-9f12-d81f1dda91cb    light   \n",
       "\n",
       "        product_id                category_code  product_id_mapped  \n",
       "0          1004838     construction.tools.light              311.0  \n",
       "1          1004838     construction.tools.light              311.0  \n",
       "2          1005252     construction.tools.light                1.0  \n",
       "3          1004503     construction.tools.light             1259.0  \n",
       "4          1005191     construction.tools.light                2.0  \n",
       "...            ...                          ...                ...  \n",
       "369299     4400467  computers.components.cooler             4002.0  \n",
       "369300     1004903     construction.tools.light              137.0  \n",
       "369301     1004903     construction.tools.light              137.0  \n",
       "369302     1004856     construction.tools.light               66.0  \n",
       "369303     1004856     construction.tools.light               66.0  \n",
       "\n",
       "[369304 rows x 56 columns]"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Read first 500.000 Rows\n",
    "for chunk in pd.read_table(\"2019-Dec.csv\",\n",
    "                           sep=\",\", header=0,\n",
    "                           infer_datetime_format=True, low_memory=False, chunksize=500000):\n",
    "    # Filter out other event types than 'view'\n",
    "    chunk = chunk[chunk['event_type'] == 'view']\n",
    "    # Filter out missing 'category_code' rows\n",
    "    chunk = chunk[chunk['category_code'].isna() == False]\n",
    "    chunk.reset_index(drop=True, inplace=True)\n",
    "\n",
    "    # Filter out all Sessions of length 1\n",
    "    count_sessions = chunk.groupby('user_session').count()\n",
    "    window_length = count_sessions.max()[0]\n",
    "    unique_sessions = [count_sessions.index[i] for i in range(\n",
    "        count_sessions.shape[0]) if count_sessions.iloc[i, 0] == 1]\n",
    "    chunk = chunk[~chunk['user_session'].isin(unique_sessions)]\n",
    "    chunk.reset_index(drop=True, inplace=True)\n",
    "\n",
    "    # Text embedding based on https://tfhub.dev/google/nnlm-en-dim50/2\n",
    "    last_category = []\n",
    "    for i, el in enumerate(chunk['category_code']):\n",
    "        last_category.append(el.split('.')[-1])\n",
    "    chunk['Product'] = last_category\n",
    "    embed = hub.load(\"https://tfhub.dev/google/nnlm-en-dim50/2\")\n",
    "    embeddings = embed(chunk['Product'].tolist())\n",
    "    for dim in range(embeddings.shape[1]):\n",
    "        chunk['embedding_'+str(dim)] = embeddings[:, dim]\n",
    "\n",
    "    # Standardization\n",
    "    mean = chunk['price'].mean(axis=0)\n",
    "    print('Mean:', mean)\n",
    "    std = chunk['price'].std(axis=0)\n",
    "    print('Std:', std)\n",
    "    chunk['price_standardized'] = (chunk['price'] - mean) / std\n",
    "\n",
    "    chunk.sort_values(by=['user_session', 'event_time'], inplace=True)\n",
    "    chunk['price_standardized'] = chunk['price_standardized'].astype('float32')\n",
    "    chunk['product_id'] = chunk['product_id'].astype('int32')\n",
    "    chunk.reset_index(drop=True, inplace=True)\n",
    "\n",
    "    print('Sessions:', pd.unique(chunk['user_session']).shape)\n",
    "    print('Unique Products:', pd.unique(chunk['product_id']).shape)\n",
    "    print('Unique category_code:', pd.unique(chunk['category_code']).shape)\n",
    "\n",
    "    columns = ['embedding_'+str(i) for i in range(embeddings.shape[1])]\n",
    "    columns.append('price_standardized')\n",
    "    columns.append('user_session')\n",
    "    columns.append('Product')\n",
    "    columns.append('product_id')\n",
    "    columns.append('category_code')\n",
    "\n",
    "    df = chunk[columns]\n",
    "    break\n",
    "\n",
    "mapping = pd.read_csv('../ID_Mapping.csv')[['Item_ID', 'Mapped_ID']]\n",
    "dict_items = mapping.set_index('Item_ID').to_dict()['Mapped_ID']\n",
    "\n",
    "df['product_id_mapped'] = df['product_id'].map(dict_items)   \n",
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(30941, 31, 1)\n"
     ]
    }
   ],
   "source": [
    "sessions_padded = np.load('list_sessions_padded_autoencoder.npy')\n",
    "print(sessions_padded.shape)\n",
    "n_output_features = int(sessions_padded.max())\n",
    "n_unique_input_ids = int(sessions_padded.max())\n",
    "window_length = sessions_padded.shape[1]\n",
    "n_input_features = sessions_padded.shape[2]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### LSTM Autoencoder in functional API\n",
    "- Input: x rows (time steps) of Item IDs in a Session\n",
    "- Output: reconstructed Session"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def build_autoencoder(window_length=50,\n",
    "                      units_lstm_layer=100,\n",
    "                      n_unique_input_ids=0,\n",
    "                      embedding_dim=200,\n",
    "                      n_input_features=1,\n",
    "                      n_output_features=3,\n",
    "                      dropout_rate=0.1):\n",
    "\n",
    "    inputs = keras.layers.Input(\n",
    "        shape=[window_length, n_input_features], dtype=np.float32)\n",
    "\n",
    "    # Encoder\n",
    "    # Embedding Layer\n",
    "    embedding_layer = tf.keras.layers.Embedding(\n",
    "        n_unique_input_ids+1, embedding_dim, input_length=window_length)  # , mask_zero=True)\n",
    "    embeddings = embedding_layer(inputs[:, :, 0])\n",
    "\n",
    "    mask = inputs[:, :, 0] != 0\n",
    "\n",
    "    # LSTM Layer 1\n",
    "    lstm1_output, lstm1_state_h, lstm1_state_c = keras.layers.LSTM(units=units_lstm_layer, return_state=True,\n",
    "                                                                   return_sequences=True)(embeddings, mask=mask)\n",
    "    lstm1_state = [lstm1_state_h, lstm1_state_c]\n",
    "\n",
    "    # Decoder\n",
    "    # input: lstm1_state_c, lstm1_state_h\n",
    "    decoder_state_c = lstm1_state_c\n",
    "    decoder_state_h = lstm1_state_h\n",
    "    decoder_outputs = tf.expand_dims(lstm1_state_h, 1)\n",
    "\n",
    "    list_states = []\n",
    "    decoder_layer = keras.layers.LSTM(\n",
    "        units=units_lstm_layer, return_state=True, return_sequences=True, unroll=False)\n",
    "    for i in range(window_length):\n",
    "        decoder_outputs, decoder_state_h, decoder_state_c = decoder_layer(decoder_outputs,\n",
    "                                                                          initial_state=[decoder_state_h,\n",
    "                                                                                         decoder_state_c])\n",
    "        list_states.append(decoder_state_h)\n",
    "    stacked = tf.stack(list_states, axis=1)\n",
    "\n",
    "    fc_layer = tf.keras.layers.Dense(\n",
    "        n_output_features+1, kernel_initializer='he_normal')\n",
    "\n",
    "    fc_layer_output = tf.keras.layers.TimeDistributed(fc_layer)(\n",
    "        stacked, mask=mask)\n",
    "\n",
    "    mask_softmax = tf.tile(tf.expand_dims(mask, axis=2),\n",
    "                           [1, 1, n_output_features+1])\n",
    "\n",
    "    softmax = tf.keras.layers.Softmax(axis=2, dtype=tf.float32)(\n",
    "        fc_layer_output, mask=mask_softmax)\n",
    "\n",
    "    model = keras.models.Model(inputs=[inputs],\n",
    "                               outputs=[softmax])\n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.training.tracking.util.CheckpointLoadStatus at 0x19b0f5de0c8>"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = build_autoencoder(window_length=window_length,\n",
    "                                             n_output_features=n_output_features,\n",
    "                                             n_unique_input_ids=n_unique_input_ids,\n",
    "                                             n_input_features=n_input_features,\n",
    "                                             embedding_dim=200,\n",
    "                                             units_lstm_layer=300,\n",
    "                                             dropout_rate=0.0)\n",
    "model.load_weights(\"weights\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "id_mapping = pd.read_csv('../ID_Mapping.csv')\n",
    "distribution_items = df.groupby('product_id_mapped').count()\n",
    "\n",
    "if os.path.isfile('errors.npy'):\n",
    "    errors = np.load('errors.npy')\n",
    "else:\n",
    "    list_errors = []\n",
    "    bce = tf.keras.losses.SparseCategoricalCrossentropy(\n",
    "        from_logits=False, reduction='sum')\n",
    "    batch_size = 128 # lower batch size, if your gpu memory isnt high enough\n",
    "\n",
    "    try:\n",
    "        for i in range(0, sessions_padded.shape[0], batch_size):\n",
    "            y_pred = model(sessions_padded[i:i+batch_size])\n",
    "            y_true = sessions_padded[i:i+batch_size]\n",
    "\n",
    "            for batch in range(0, batch_size, 1):\n",
    "                total = 0.0\n",
    "                for j in range(sessions_padded.shape[1]):\n",
    "                    if y_true[batch, j, 0] > 0:\n",
    "                        loss = bce(y_true[batch, j, 0], y_pred[batch, j, :])\n",
    "                        total = total + loss\n",
    "                    else:\n",
    "                        break\n",
    "                list_errors.append(total.numpy())\n",
    "    except Exception as e:\n",
    "        print(e)\n",
    "    errors = np.array(list_errors)\n",
    "    np.save('errors.npy', errors)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x19b90adcb88>"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x1080 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "mask = sessions_padded[:, :, 0]!=0\n",
    "lengths_session = mask.sum(axis=1)\n",
    "mean_errors = errors/lengths_session\n",
    "pd.DataFrame(mean_errors).plot(figsize=(20,15))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([  169,   246,   394,   498,   630,  1039,  1578,  2008,  2040,\n",
       "         2447,  2557,  2609,  3179,  3276,  3481,  3615,  3813,  4179,\n",
       "         4361,  4794,  5077,  6184,  6369,  7347,  7596,  8415,  8761,\n",
       "         8773,  9011,  9404,  9504,  9613,  9880,  9907,  9978, 10050,\n",
       "        10229, 10573, 10654, 11196, 11429, 11477, 11493, 11654, 11975,\n",
       "        12135, 13526, 13659, 13729, 14139, 14469, 14910, 15203, 15429,\n",
       "        15934, 15982, 16310, 16352, 16504, 16647, 16743, 17046, 17085,\n",
       "        17302, 17342, 17449, 18584, 18702, 18711, 18770, 19204, 19642,\n",
       "        19758, 19863, 19891, 20135, 20244, 20652, 20865, 20899, 21077,\n",
       "        21680, 23338, 23407, 23892, 24101, 24257, 24259, 24396, 25078,\n",
       "        25127, 25380, 25576, 26071, 26082, 26123, 26323, 26373, 27007,\n",
       "        27629, 27664, 27833, 28388, 28739, 29576, 29588, 30381, 30529,\n",
       "        30873, 30930], dtype=int64),)"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tmp = np.where(mean_errors>1.9)\n",
    "tmp"
   ]
  },
  {
   "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>category_code</th>\n",
       "      <th>Item_ID</th>\n",
       "      <th>Item_ID_Mapped</th>\n",
       "      <th>Count</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>electronics.audio.acoustic</td>\n",
       "      <td>14100052.0</td>\n",
       "      <td>349</td>\n",
       "      <td>27</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>electronics.audio.acoustic</td>\n",
       "      <td>14100285.0</td>\n",
       "      <td>350</td>\n",
       "      <td>16</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>electronics.audio.acoustic</td>\n",
       "      <td>14100052.0</td>\n",
       "      <td>349</td>\n",
       "      <td>27</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>apparel.shoes</td>\n",
       "      <td>4100126.0</td>\n",
       "      <td>351</td>\n",
       "      <td>188</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>computers.components.power_supply</td>\n",
       "      <td>11500273.0</td>\n",
       "      <td>352</td>\n",
       "      <td>84</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>apparel.shoes</td>\n",
       "      <td>9300000.0</td>\n",
       "      <td>291</td>\n",
       "      <td>82</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>construction.tools.light</td>\n",
       "      <td>1004249.0</td>\n",
       "      <td>246</td>\n",
       "      <td>1145</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>construction.tools.light</td>\n",
       "      <td>1005105.0</td>\n",
       "      <td>152</td>\n",
       "      <td>1607</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>construction.tools.light</td>\n",
       "      <td>1005159.0</td>\n",
       "      <td>353</td>\n",
       "      <td>981</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>construction.tools.light</td>\n",
       "      <td>1005235.0</td>\n",
       "      <td>354</td>\n",
       "      <td>413</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>construction.tools.light</td>\n",
       "      <td>1002540.0</td>\n",
       "      <td>355</td>\n",
       "      <td>137</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>construction.tools.light</td>\n",
       "      <td>1005143.0</td>\n",
       "      <td>356</td>\n",
       "      <td>125</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>apparel.shoes</td>\n",
       "      <td>9300000.0</td>\n",
       "      <td>291</td>\n",
       "      <td>82</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>computers.components.power_supply</td>\n",
       "      <td>11500273.0</td>\n",
       "      <td>352</td>\n",
       "      <td>84</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>electronics.audio.acoustic</td>\n",
       "      <td>14100285.0</td>\n",
       "      <td>350</td>\n",
       "      <td>16</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>apparel.shoes</td>\n",
       "      <td>4100126.0</td>\n",
       "      <td>351</td>\n",
       "      <td>188</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>computers.components.power_supply</td>\n",
       "      <td>11500273.0</td>\n",
       "      <td>352</td>\n",
       "      <td>84</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>electronics.audio.acoustic</td>\n",
       "      <td>14100052.0</td>\n",
       "      <td>349</td>\n",
       "      <td>27</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                        category_code     Item_ID  Item_ID_Mapped  Count\n",
       "0          electronics.audio.acoustic  14100052.0             349     27\n",
       "1          electronics.audio.acoustic  14100285.0             350     16\n",
       "2          electronics.audio.acoustic  14100052.0             349     27\n",
       "3                       apparel.shoes   4100126.0             351    188\n",
       "4   computers.components.power_supply  11500273.0             352     84\n",
       "5                       apparel.shoes   9300000.0             291     82\n",
       "6            construction.tools.light   1004249.0             246   1145\n",
       "7            construction.tools.light   1005105.0             152   1607\n",
       "8            construction.tools.light   1005159.0             353    981\n",
       "9            construction.tools.light   1005235.0             354    413\n",
       "10           construction.tools.light   1002540.0             355    137\n",
       "11           construction.tools.light   1005143.0             356    125\n",
       "12                      apparel.shoes   9300000.0             291     82\n",
       "13  computers.components.power_supply  11500273.0             352     84\n",
       "14         electronics.audio.acoustic  14100285.0             350     16\n",
       "15                      apparel.shoes   4100126.0             351    188\n",
       "16  computers.components.power_supply  11500273.0             352     84\n",
       "17         electronics.audio.acoustic  14100052.0             349     27"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "(1.910614013671875, 34.391052)"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "example = 169 # 23279\n",
    "session = pd.DataFrame()\n",
    "session['category_code'] = [id_mapping['category_code'][int(i)-1] for i in sessions_padded[example,:,0] if i>0]\n",
    "session['Item_ID'] = [id_mapping['Item_ID'][int(i)-1] for i in sessions_padded[example,:,0] if i>0]\n",
    "session['Item_ID_Mapped'] = [int(i) for i in sessions_padded[example,:,0] if i>0]\n",
    "session['Count'] = [distribution_items.loc[i, 'embedding_0'] for i in session['Item_ID_Mapped']]\n",
    "#session.drop(['Item_ID_Mapped'], axis=1, inplace=True)\n",
    "session\n",
    "mean_errors[example], errors[example]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Custom Session"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_true = np.zeros((1, 31, 1), dtype=np.int32)\n",
    "y_true[0,0,0] = 2499\n",
    "y_true[0,1,0] = 29\n",
    "y_true[0,2,0] = 100\n",
    "y_true[0,3,0] = 1001 #1000\n",
    "y_true[0,4,0] = 1507\n",
    "y_true[0,5,0] = 501 #3000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([8.75680351])"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "list_errors = []\n",
    "bce = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False, reduction='sum')\n",
    "batch_size = 1\n",
    "\n",
    "for i in range(0, y_true.shape[0], batch_size):\n",
    "    y_pred = model(y_true)\n",
    "    \n",
    "    for batch in range(0, batch_size, 1):\n",
    "        total = 0.0\n",
    "        for j in range(sessions_padded.shape[1]):\n",
    "            if y_true[batch,j,0] > 0:\n",
    "                loss = bce(y_true[batch,j,0], y_pred[batch,j,:])\n",
    "                total = total + loss\n",
    "            else:\n",
    "                break\n",
    "        list_errors.append(total.numpy())\n",
    "\n",
    "errors_2 = np.array(list_errors)\n",
    "mask = y_true[:, :, 0]!=0\n",
    "lengths_session = mask.sum(axis=1)\n",
    "mean_errors_2 = errors_2/lengths_session\n",
    "mean_errors_2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "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>category_code</th>\n",
       "      <th>Item_ID</th>\n",
       "      <th>Item_ID_Mapped</th>\n",
       "      <th>Count</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>appliances.kitchen.blender</td>\n",
       "      <td>3100204.0</td>\n",
       "      <td>2499</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>appliances.personal.massager</td>\n",
       "      <td>1801226.0</td>\n",
       "      <td>29</td>\n",
       "      <td>175</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>computers.ebooks</td>\n",
       "      <td>5100605.0</td>\n",
       "      <td>100</td>\n",
       "      <td>15</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>furniture.kitchen.chair</td>\n",
       "      <td>10100202.0</td>\n",
       "      <td>1001</td>\n",
       "      <td>18</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>furniture.kitchen.chair</td>\n",
       "      <td>11400736.0</td>\n",
       "      <td>1507</td>\n",
       "      <td>15</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>apparel.shoes.keds</td>\n",
       "      <td>2300307.0</td>\n",
       "      <td>501</td>\n",
       "      <td>19</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                  category_code     Item_ID  Item_ID_Mapped  Count\n",
       "0    appliances.kitchen.blender   3100204.0            2499      9\n",
       "1  appliances.personal.massager   1801226.0              29    175\n",
       "2              computers.ebooks   5100605.0             100     15\n",
       "3       furniture.kitchen.chair  10100202.0            1001     18\n",
       "4       furniture.kitchen.chair  11400736.0            1507     15\n",
       "5            apparel.shoes.keds   2300307.0             501     19"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "session = pd.DataFrame()\n",
    "session['category_code'] = [id_mapping['category_code'][int(i)-1] for i in y_true[0,:,0] if i>0]\n",
    "session['Item_ID'] = [id_mapping['Item_ID'][int(i)-1] for i in y_true[0,:,0] if i>0]\n",
    "session['Item_ID_Mapped'] = [int(i) for i in y_true[0,:,0] if i>0]\n",
    "session['Count'] = [distribution_items.loc[i, 'embedding_0'] for i in session['Item_ID_Mapped']]\n",
    "#session.drop(['Item_ID_Mapped'], axis=1, inplace=True)\n",
    "session"
   ]
  },
  {
   "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": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}