movietransformer.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "fa56771e-9f59-4d7e-96d5-aba221897efd",
   "metadata": {},
   "source": [
    "# MovieTransformer\n",
    "\n",
    "A small experiment training a transformer to translate moviewarez filenames into actual movie titles. Still far from being complete, but hopefully a nice project to play with!\n",
    "\n",
    "Strongly based on the [Neural machine translation with a Transformer and Keras](https://www.tensorflow.org/text/tutorials/transformer) tutorial. \n",
    "\n",
    "Kudos to [OpenSubtitles](https://opensubtitles.org) for sharing their data!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f597fa3d-e887-4739-a2ba-f711f3880d14",
   "metadata": {},
   "outputs": [],
   "source": [
    "!pip install tensorflow_datasets tensorflow tensorflow-text pandas"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "bca05946-e361-4af0-9984-c63feef4e0aa",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2023-01-30 15:13:55.503684: I tensorflow/core/platform/cpu_feature_guard.cc:193] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 FMA\n",
      "To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.\n"
     ]
    }
   ],
   "source": [
    "import logging\n",
    "import time\n",
    "\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "import tensorflow_datasets as tfds\n",
    "import tensorflow as tf\n",
    "\n",
    "import tensorflow_text\n",
    "\n",
    "import pandas as pd"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "25b4e39b-6967-4787-b4f4-579b41bbbe17",
   "metadata": {},
   "source": [
    "# Data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8828de5e-b550-430f-b314-f8b7fde5a50e",
   "metadata": {},
   "source": [
    "## 0. Download dataset\n",
    "\n",
    "The following code downloads the full opensubtitles dataset (~300M). Smaller ones (`subtitles_day`, `subtitles_week`, and `subtitles_month`) are available too."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "f6b2eadf-52f7-4e9d-9eda-c587b0ba00fe",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "--2023-01-30 15:22:17--  http://dl.opensubtitles.org/addons/export/subtitles_all.txt.gz\n",
      "Resolving dl.opensubtitles.org (dl.opensubtitles.org)... 2606:4700:e0::ac40:6618, 2606:4700:e0::ac40:6718, 172.64.103.24, ...\n",
      "Connecting to dl.opensubtitles.org (dl.opensubtitles.org)|2606:4700:e0::ac40:6618|:80... connected.\n",
      "HTTP request sent, awaiting response... 200 OK\n",
      "Length: 318565459 (304M) [application/x-gzip]\n",
      "Saving to: ‘subtitles_all.txt.gz’\n",
      "\n",
      "subtitles_all.txt.g 100%[===================>] 303,81M  1021KB/s    in 6m 19s  \n",
      "\n",
      "2023-01-30 15:28:36 (821 KB/s) - ‘subtitles_all.txt.gz’ saved [318565459/318565459]\n",
      "\n"
     ]
    }
   ],
   "source": [
    "dataset_url = \"http://dl.opensubtitles.org/addons/export/subtitles_all.txt.gz\"\n",
    "dataset_fname = dataset_url.split(\"/\")[-1]\n",
    "\n",
    "!wget {dataset_url}\n",
    "!gunzip {dataset_fname}"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "334ea7d4-acfa-4c5b-93db-8f9a4b98f80f",
   "metadata": {},
   "source": [
    "## 1. Import TSV file\n",
    "\n",
    "Here we import the data, which is saved in the txt file as tab-separated values. Note that the dataset contains some bad rows, so we just ignore them (but you will see a bunch of errors in this cell...)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "b80c20c7-7320-4d49-8843-8d329f503b71",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/3w/1spf9bkn2gjbc8026chntkb80000gn/T/ipykernel_56483/579959582.py:1: FutureWarning: The error_bad_lines argument has been deprecated and will be removed in a future version. Use on_bad_lines in the future.\n",
      "\n",
      "\n",
      "  df = pd.read_csv('subtitles_all.txt', sep='\\t', error_bad_lines=False)\n",
      "Skipping line 2890: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 82417: expected 16 fields, saw 24\n",
      "Skipping line 88907: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 103205: expected 16 fields, saw 24\n",
      "\n",
      "Skipping line 141390: expected 16 fields, saw 17\n",
      "Skipping line 149209: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 184474: expected 16 fields, saw 17\n",
      "Skipping line 189187: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 213065: expected 16 fields, saw 17\n",
      "Skipping line 213813: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 244113: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 283930: expected 16 fields, saw 17\n",
      "Skipping line 291999: expected 16 fields, saw 17\n",
      "Skipping line 292180: expected 16 fields, saw 17\n",
      "Skipping line 292260: expected 16 fields, saw 17\n",
      "Skipping line 293000: expected 16 fields, saw 17\n",
      "Skipping line 294783: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 323131: expected 16 fields, saw 24\n",
      "\n",
      "Skipping line 334163: expected 16 fields, saw 24\n",
      "Skipping line 334819: expected 16 fields, saw 24\n",
      "Skipping line 335911: expected 16 fields, saw 24\n",
      "Skipping line 336726: expected 16 fields, saw 24\n",
      "Skipping line 338083: expected 16 fields, saw 24\n",
      "Skipping line 340544: expected 16 fields, saw 24\n",
      "Skipping line 340688: expected 16 fields, saw 24\n",
      "Skipping line 341481: expected 16 fields, saw 24\n",
      "Skipping line 341792: expected 16 fields, saw 24\n",
      "Skipping line 342574: expected 16 fields, saw 24\n",
      "Skipping line 342633: expected 16 fields, saw 24\n",
      "Skipping line 342652: expected 16 fields, saw 24\n",
      "Skipping line 342943: expected 16 fields, saw 24\n",
      "Skipping line 343254: expected 16 fields, saw 24\n",
      "Skipping line 343273: expected 16 fields, saw 24\n",
      "Skipping line 343638: expected 16 fields, saw 24\n",
      "Skipping line 344007: expected 16 fields, saw 24\n",
      "Skipping line 344264: expected 16 fields, saw 24\n",
      "Skipping line 344566: expected 16 fields, saw 24\n",
      "Skipping line 344687: expected 16 fields, saw 24\n",
      "Skipping line 344757: expected 16 fields, saw 24\n",
      "Skipping line 345434: expected 16 fields, saw 24\n",
      "Skipping line 347442: expected 16 fields, saw 24\n",
      "Skipping line 348372: expected 16 fields, saw 24\n",
      "Skipping line 349167: expected 16 fields, saw 24\n",
      "Skipping line 349221: expected 16 fields, saw 24\n",
      "Skipping line 349257: expected 16 fields, saw 24\n",
      "Skipping line 349764: expected 16 fields, saw 24\n",
      "Skipping line 349883: expected 16 fields, saw 24\n",
      "Skipping line 350190: expected 16 fields, saw 24\n",
      "Skipping line 350256: expected 16 fields, saw 24\n",
      "Skipping line 350765: expected 16 fields, saw 24\n",
      "Skipping line 350981: expected 16 fields, saw 24\n",
      "Skipping line 351194: expected 16 fields, saw 24\n",
      "Skipping line 351562: expected 16 fields, saw 24\n",
      "Skipping line 351929: expected 16 fields, saw 24\n",
      "Skipping line 352007: expected 16 fields, saw 24\n",
      "Skipping line 353818: expected 16 fields, saw 24\n",
      "Skipping line 354147: expected 16 fields, saw 24\n",
      "Skipping line 355884: expected 16 fields, saw 24\n",
      "Skipping line 356029: expected 16 fields, saw 24\n",
      "Skipping line 356104: expected 16 fields, saw 24\n",
      "Skipping line 356760: expected 16 fields, saw 24\n",
      "Skipping line 357490: expected 16 fields, saw 24\n",
      "Skipping line 357952: expected 16 fields, saw 24\n",
      "Skipping line 359019: expected 16 fields, saw 24\n",
      "\n",
      "Skipping line 387010: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 417235: expected 16 fields, saw 17\n",
      "Skipping line 420211: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 439598: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 758614: expected 16 fields, saw 17\n",
      "Skipping line 759716: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 1203571: expected 16 fields, saw 19\n",
      "\n",
      "Skipping line 1223832: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 1268668: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 1459009: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 1475552: expected 16 fields, saw 17\n",
      "Skipping line 1478698: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 1921179: expected 16 fields, saw 24\n",
      "\n",
      "Skipping line 2196325: expected 16 fields, saw 24\n",
      "Skipping line 2200514: expected 16 fields, saw 24\n",
      "Skipping line 2204091: expected 16 fields, saw 24\n",
      "Skipping line 2209322: expected 16 fields, saw 24\n",
      "Skipping line 2216478: expected 16 fields, saw 24\n",
      "Skipping line 2220419: expected 16 fields, saw 24\n",
      "Skipping line 2226964: expected 16 fields, saw 24\n",
      "\n",
      "Skipping line 2620422: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 2666201: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 2751399: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 2980831: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 2986542: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 3144024: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 3339655: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 3347216: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 3598401: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 3731769: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 3781539: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 3959118: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 4006471: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 4119142: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 4193245: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 4274817: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 4319642: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 4532695: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 4555018: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 4763572: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 5261508: expected 16 fields, saw 24\n",
      "\n",
      "Skipping line 5308521: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 5531161: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 5574140: expected 16 fields, saw 17\n",
      "\n",
      "Skipping line 5629727: expected 16 fields, saw 24\n",
      "\n",
      "Skipping line 5681286: expected 16 fields, saw 17\n",
      "Skipping line 5682024: expected 16 fields, saw 17\n",
      "\n",
      "/var/folders/3w/1spf9bkn2gjbc8026chntkb80000gn/T/ipykernel_56483/579959582.py:1: DtypeWarning: Columns (0,6) have mixed types. Specify dtype option on import or set low_memory=False.\n",
      "  df = pd.read_csv('subtitles_all.txt', sep='\\t', error_bad_lines=False)\n"
     ]
    }
   ],
   "source": [
    "df = pd.read_csv('subtitles_all.txt', sep='\\t', error_bad_lines=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "637b7f94-0afe-4eb3-9215-e4d71f44de3c",
   "metadata": {},
   "source": [
    "Let us see what the dataset looks like..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "00fa09e8-d51d-4e20-8ccd-342223ef538f",
   "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>IDSubtitle</th>\n",
       "      <th>MovieName</th>\n",
       "      <th>MovieYear</th>\n",
       "      <th>LanguageName</th>\n",
       "      <th>ISO639</th>\n",
       "      <th>SubAddDate</th>\n",
       "      <th>ImdbID</th>\n",
       "      <th>SubFormat</th>\n",
       "      <th>SubSumCD</th>\n",
       "      <th>MovieReleaseName</th>\n",
       "      <th>MovieFPS</th>\n",
       "      <th>SeriesSeason</th>\n",
       "      <th>SeriesEpisode</th>\n",
       "      <th>SeriesIMDBParent</th>\n",
       "      <th>MovieKind</th>\n",
       "      <th>URL</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>Alien3</td>\n",
       "      <td>1992.0</td>\n",
       "      <td>English</td>\n",
       "      <td>en</td>\n",
       "      <td>2004-10-31 23:54:23</td>\n",
       "      <td>103644</td>\n",
       "      <td>sub</td>\n",
       "      <td>2.0</td>\n",
       "      <td>Alien.3</td>\n",
       "      <td>11.000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>movie</td>\n",
       "      <td>http://www.opensubtitles.org/subtitles/1/alien...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>Identity</td>\n",
       "      <td>2003.0</td>\n",
       "      <td>Slovenian</td>\n",
       "      <td>sl</td>\n",
       "      <td>2004-10-31 23:54:23</td>\n",
       "      <td>309698</td>\n",
       "      <td>sub</td>\n",
       "      <td>1.0</td>\n",
       "      <td>Identity.DVDRiP.XViD</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>movie</td>\n",
       "      <td>http://www.opensubtitles.org/subtitles/2/ident...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3</td>\n",
       "      <td>Ghost in the Shell 2: Innocence</td>\n",
       "      <td>2004.0</td>\n",
       "      <td>English</td>\n",
       "      <td>en</td>\n",
       "      <td>2004-10-31 23:54:23</td>\n",
       "      <td>347246</td>\n",
       "      <td>srt</td>\n",
       "      <td>1.0</td>\n",
       "      <td>Innocence.2004.DVDRip.XviD</td>\n",
       "      <td>23.980</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>movie</td>\n",
       "      <td>http://www.opensubtitles.org/subtitles/3/ghost...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>Planet of the Apes</td>\n",
       "      <td>2001.0</td>\n",
       "      <td>Slovenian</td>\n",
       "      <td>sl</td>\n",
       "      <td>2004-10-31 23:54:23</td>\n",
       "      <td>133152</td>\n",
       "      <td>sub</td>\n",
       "      <td>1.0</td>\n",
       "      <td>Planet.Of.The. Apes.XviD</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>movie</td>\n",
       "      <td>http://www.opensubtitles.org/subtitles/4/plane...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5</td>\n",
       "      <td>The City of Lost Children</td>\n",
       "      <td>1995.0</td>\n",
       "      <td>English</td>\n",
       "      <td>en</td>\n",
       "      <td>2004-11-01 20:11:53</td>\n",
       "      <td>112682</td>\n",
       "      <td>srt</td>\n",
       "      <td>2.0</td>\n",
       "      <td>The.City.Of.Lost.Children.1995.iNTERNAL.DVDRi...</td>\n",
       "      <td>23.976</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>movie</td>\n",
       "      <td>http://www.opensubtitles.org/subtitles/5/the-c...</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",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5941101</th>\n",
       "      <td>9412705</td>\n",
       "      <td>The Last of Us Long Long Time</td>\n",
       "      <td>2023.0</td>\n",
       "      <td>French</td>\n",
       "      <td>fr</td>\n",
       "      <td>2023-01-30 09:04:22</td>\n",
       "      <td>14500888</td>\n",
       "      <td>srt</td>\n",
       "      <td>1.0</td>\n",
       "      <td>The.Last.of.Us.S01E03.Long.Long.Time.1080p.AMZ...</td>\n",
       "      <td>0.000</td>\n",
       "      <td>1.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>3581920.0</td>\n",
       "      <td>tv</td>\n",
       "      <td>http://www.opensubtitles.org/subtitles/9412705...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5941102</th>\n",
       "      <td>9412706</td>\n",
       "      <td>Trolley Episode #1.7</td>\n",
       "      <td>2023.0</td>\n",
       "      <td>German</td>\n",
       "      <td>de</td>\n",
       "      <td>2023-01-30 09:10:28</td>\n",
       "      <td>24787062</td>\n",
       "      <td>srt</td>\n",
       "      <td>1.0</td>\n",
       "      <td>Trolley.S01E07.The.Promise.1080p.NF.WEB-DL.DD...</td>\n",
       "      <td>23.976</td>\n",
       "      <td>1.0</td>\n",
       "      <td>7.0</td>\n",
       "      <td>24224842.0</td>\n",
       "      <td>tv</td>\n",
       "      <td>http://www.opensubtitles.org/subtitles/9412706...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5941103</th>\n",
       "      <td>9412707</td>\n",
       "      <td>Trolley Episode #1.8</td>\n",
       "      <td>2023.0</td>\n",
       "      <td>German</td>\n",
       "      <td>de</td>\n",
       "      <td>2023-01-30 09:10:45</td>\n",
       "      <td>24787064</td>\n",
       "      <td>srt</td>\n",
       "      <td>1.0</td>\n",
       "      <td>Trolley.S01E08.The.Right.Thing.To.Do.1080p.NF...</td>\n",
       "      <td>23.976</td>\n",
       "      <td>1.0</td>\n",
       "      <td>8.0</td>\n",
       "      <td>24224842.0</td>\n",
       "      <td>tv</td>\n",
       "      <td>http://www.opensubtitles.org/subtitles/9412707...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5941104</th>\n",
       "      <td>9412708</td>\n",
       "      <td>Chyby</td>\n",
       "      <td>2021.0</td>\n",
       "      <td>Hungarian</td>\n",
       "      <td>hu</td>\n",
       "      <td>2023-01-30 09:17:04</td>\n",
       "      <td>10757922</td>\n",
       "      <td>srt</td>\n",
       "      <td>1.0</td>\n",
       "      <td>Chyby.2021.1080p.WEBRip.CZ.5.1</td>\n",
       "      <td>24.000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>movie</td>\n",
       "      <td>http://www.opensubtitles.org/subtitles/9412708...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5941105</th>\n",
       "      <td>9412709</td>\n",
       "      <td>Father Brown The Winds of Change</td>\n",
       "      <td>2023.0</td>\n",
       "      <td>Dutch</td>\n",
       "      <td>nl</td>\n",
       "      <td>2023-01-30 09:22:07</td>\n",
       "      <td>21616578</td>\n",
       "      <td>srt</td>\n",
       "      <td>1.0</td>\n",
       "      <td>Father Brown - S10E01 - The Winds of Change</td>\n",
       "      <td>25.000</td>\n",
       "      <td>10.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>2215842.0</td>\n",
       "      <td>tv</td>\n",
       "      <td>http://www.opensubtitles.org/subtitles/9412709...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5941106 rows × 16 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        IDSubtitle                         MovieName  MovieYear LanguageName  \\\n",
       "0                1                            Alien3     1992.0      English   \n",
       "1                2                          Identity     2003.0    Slovenian   \n",
       "2                3   Ghost in the Shell 2: Innocence     2004.0      English   \n",
       "3                4                Planet of the Apes     2001.0    Slovenian   \n",
       "4                5         The City of Lost Children     1995.0      English   \n",
       "...            ...                               ...        ...          ...   \n",
       "5941101    9412705     The Last of Us Long Long Time     2023.0       French   \n",
       "5941102    9412706              Trolley Episode #1.7     2023.0       German   \n",
       "5941103    9412707              Trolley Episode #1.8     2023.0       German   \n",
       "5941104    9412708                             Chyby     2021.0    Hungarian   \n",
       "5941105    9412709  Father Brown The Winds of Change     2023.0        Dutch   \n",
       "\n",
       "        ISO639           SubAddDate    ImdbID SubFormat  SubSumCD  \\\n",
       "0           en  2004-10-31 23:54:23    103644       sub       2.0   \n",
       "1           sl  2004-10-31 23:54:23    309698       sub       1.0   \n",
       "2           en  2004-10-31 23:54:23    347246       srt       1.0   \n",
       "3           sl  2004-10-31 23:54:23    133152       sub       1.0   \n",
       "4           en  2004-11-01 20:11:53    112682       srt       2.0   \n",
       "...        ...                  ...       ...       ...       ...   \n",
       "5941101     fr  2023-01-30 09:04:22  14500888       srt       1.0   \n",
       "5941102     de  2023-01-30 09:10:28  24787062       srt       1.0   \n",
       "5941103     de  2023-01-30 09:10:45  24787064       srt       1.0   \n",
       "5941104     hu  2023-01-30 09:17:04  10757922       srt       1.0   \n",
       "5941105     nl  2023-01-30 09:22:07  21616578       srt       1.0   \n",
       "\n",
       "                                          MovieReleaseName  MovieFPS  \\\n",
       "0                                                  Alien.3    11.000   \n",
       "1                                     Identity.DVDRiP.XViD     0.000   \n",
       "2                               Innocence.2004.DVDRip.XviD    23.980   \n",
       "3                                 Planet.Of.The. Apes.XviD     0.000   \n",
       "4         The.City.Of.Lost.Children.1995.iNTERNAL.DVDRi...    23.976   \n",
       "...                                                    ...       ...   \n",
       "5941101  The.Last.of.Us.S01E03.Long.Long.Time.1080p.AMZ...     0.000   \n",
       "5941102   Trolley.S01E07.The.Promise.1080p.NF.WEB-DL.DD...    23.976   \n",
       "5941103   Trolley.S01E08.The.Right.Thing.To.Do.1080p.NF...    23.976   \n",
       "5941104                     Chyby.2021.1080p.WEBRip.CZ.5.1    24.000   \n",
       "5941105        Father Brown - S10E01 - The Winds of Change    25.000   \n",
       "\n",
       "         SeriesSeason  SeriesEpisode  SeriesIMDBParent MovieKind  \\\n",
       "0                 0.0            0.0               0.0     movie   \n",
       "1                 0.0            0.0               0.0     movie   \n",
       "2                 0.0            0.0               0.0     movie   \n",
       "3                 0.0            0.0               0.0     movie   \n",
       "4                 0.0            0.0               0.0     movie   \n",
       "...               ...            ...               ...       ...   \n",
       "5941101           1.0            3.0         3581920.0        tv   \n",
       "5941102           1.0            7.0        24224842.0        tv   \n",
       "5941103           1.0            8.0        24224842.0        tv   \n",
       "5941104           0.0            0.0               0.0     movie   \n",
       "5941105          10.0            1.0         2215842.0        tv   \n",
       "\n",
       "                                                       URL  \n",
       "0        http://www.opensubtitles.org/subtitles/1/alien...  \n",
       "1        http://www.opensubtitles.org/subtitles/2/ident...  \n",
       "2        http://www.opensubtitles.org/subtitles/3/ghost...  \n",
       "3        http://www.opensubtitles.org/subtitles/4/plane...  \n",
       "4        http://www.opensubtitles.org/subtitles/5/the-c...  \n",
       "...                                                    ...  \n",
       "5941101  http://www.opensubtitles.org/subtitles/9412705...  \n",
       "5941102  http://www.opensubtitles.org/subtitles/9412706...  \n",
       "5941103  http://www.opensubtitles.org/subtitles/9412707...  \n",
       "5941104  http://www.opensubtitles.org/subtitles/9412708...  \n",
       "5941105  http://www.opensubtitles.org/subtitles/9412709...  \n",
       "\n",
       "[5941106 rows x 16 columns]"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3323b356-3485-48ee-9f18-e9b953ab79a5",
   "metadata": {},
   "source": [
    "## 2. Clean data\n",
    "\n",
    "This section is devoted to cleaning the dataset so it is ready to be used for training.\n",
    "\n",
    "Note that the following is the bare minimum set of transformations which provide decent results, but there are plenty more that can be done to improve our model's performances (e.g. removing individual episode titles which do not appear in the filenames, add explicit text for year and season/episode info, etc)."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0ed25201-0e70-4ca9-a49d-bf72dff6409a",
   "metadata": {},
   "source": [
    "### 2.1 Filter columns, drop NaN cells, cast as string\n",
    "\n",
    "Here we only select the two columns we are interested in, then drop those rows where one of the field is Nan and make sure everything is a string"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 165,
   "id": "25137870-bfda-4de6-b3d7-65db22b3d3b9",
   "metadata": {},
   "outputs": [],
   "source": [
    "df = df[['MovieName', 'MovieReleaseName']].dropna().astype('string')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "afcda147-a967-4cc2-9164-85dfc3a9225e",
   "metadata": {},
   "source": [
    "### 2.2 Remove junk rows\n",
    "Some rows contain junky text such as `NO MOVIE TITLE YET (CLICK ON REPORT AND INSERT IMDB LINK)`, so we remove them (feel free to add whatever other checks seem reasonable to you here)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 166,
   "id": "1ada5e7e-409e-4ec9-bbea-2152979b084f",
   "metadata": {},
   "outputs": [],
   "source": [
    "df = df.drop(df[df['MovieName'].str.contains(\"NO MOVIE\") | df['MovieReleaseName'].str.contains(\"NO MOVIE\")].index)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2089d73c-6be2-4359-83de-6bb7f8387795",
   "metadata": {},
   "source": [
    "---\n",
    "We also drop the few lines which contain `\\t` or `\\n` in the movie or file name, so that we can use them as start- and end-of-sequence markers (yeah it is not ideal, but works for our simple use case and tokenizer) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 167,
   "id": "7e34f40f-3ed0-4257-a1e3-f3d6a35109c7",
   "metadata": {},
   "outputs": [],
   "source": [
    "df = df.drop(df[df['MovieName'].str.contains(\"\\t\") | df['MovieReleaseName'].str.contains(\"\\t\")].index)\n",
    "df = df.drop(df[df['MovieName'].str.contains(\"\\n\") | df['MovieReleaseName'].str.contains(\"\\n\")].index)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e5946516-1e3f-49ae-a92b-63ef7180e926",
   "metadata": {},
   "source": [
    "---\n",
    "Strip whitespace:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 168,
   "id": "f9ef7c0f-a735-40ea-b430-5df867d342d0",
   "metadata": {},
   "outputs": [],
   "source": [
    "df['MovieName'] = df['MovieName'].str.strip()\n",
    "df['MovieReleaseName'] = df['MovieReleaseName'].str.strip()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b76f4d4a-3f61-49aa-ab5f-9cc74ec3db6e",
   "metadata": {},
   "source": [
    "### 2.3 Add start and end characters\n",
    "\n",
    "The `UnicodeCharTokenizer` we will soon use does not have start/end tokens, so we are going to use tab and newline for that and manually add them to our strings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "bc0f2d5c-e647-4cdb-a137-c009511cf5d1",
   "metadata": {},
   "outputs": [],
   "source": [
    "df['MovieName'] = df['MovieName'].map(\"\\t{}\\n\".format)\n",
    "df['MovieReleaseName'] = df['MovieReleaseName'].map(\"\\t{}\\n\".format)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7fae740e-d02e-4d5d-a508-5ed5e481ead6",
   "metadata": {},
   "source": [
    "## 3. Build TF Dataset\n",
    "\n",
    "Ok, this is probably not the best way to do it (and we'll waste some time to shuffle the dataset more than once) but it does the job for now and is flexible enough to work (using a smaller `shuffle_size`) with less RAM."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "f5b94ea5-3ad8-41c8-b989-24031c475582",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(<tf.Tensor: shape=(), dtype=string, numpy=b'\\t Alien.3\\n'>, <tf.Tensor: shape=(), dtype=string, numpy=b'\\tAlien3\\n'>)\n",
      "(<tf.Tensor: shape=(), dtype=string, numpy=b'\\t Identity.DVDRiP.XViD\\n'>, <tf.Tensor: shape=(), dtype=string, numpy=b'\\tIdentity\\n'>)\n",
      "(<tf.Tensor: shape=(), dtype=string, numpy=b'\\t Innocence.2004.DVDRip.XviD\\n'>, <tf.Tensor: shape=(), dtype=string, numpy=b'\\tGhost in the Shell 2: Innocence\\n'>)\n"
     ]
    }
   ],
   "source": [
    "feats = tf.data.Dataset.from_tensor_slices(df['MovieReleaseName'])\n",
    "labels = tf.data.Dataset.from_tensor_slices(df['MovieName'])\n",
    "ds = tf.data.Dataset.zip((feats, labels))\n",
    "\n",
    "for row in ds.take(3):\n",
    "    print(row)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "e441f6e7-57ae-4eb1-a9fe-ecb834d2c8e8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[i] Dataset sizes for train, val, test:\n",
      "    5346995\n",
      "    588169\n",
      "    5942\n"
     ]
    }
   ],
   "source": [
    "# define a function to build partitions\n",
    "# NOTE that test split is tiny, for 2 reasons: \n",
    "#      1) We will just use it for qualitative evaluation\n",
    "#      2) The dataset is huge, so we will still end up with ~6K samples to play with\n",
    "\n",
    "def get_dataset_partitions_tf(ds, ds_size, train_split=0.9, val_split=0.099, test_split=0.001, \n",
    "                              shuffle=True, shuffle_size=10000, shuffle_seed=42):\n",
    "    assert (train_split + test_split + val_split) == 1\n",
    "    \n",
    "    if shuffle:\n",
    "        # Specify seed to always have the same split distribution between runs\n",
    "        ds = ds.shuffle(shuffle_size, seed=shuffle_seed)\n",
    "    \n",
    "    train_size = int(train_split * ds_size)\n",
    "    val_size = int(val_split * ds_size)\n",
    "    \n",
    "    train_ds = ds.take(train_size)    \n",
    "    val_ds = ds.skip(train_size).take(val_size)\n",
    "    test_ds = ds.skip(train_size).skip(val_size)\n",
    "    \n",
    "    return train_ds, val_ds, test_ds\n",
    "\n",
    "\n",
    "ds_size = int(ds.cardinality())\n",
    "train_ds, val_ds, test_ds = get_dataset_partitions_tf(ds, ds_size, shuffle_size=ds_size)\n",
    "\n",
    "print(\"[i] Dataset sizes for train, val, test:\")\n",
    "for dataset in (train_ds, val_ds, test_ds):\n",
    "    print(\"   \", int(dataset.cardinality()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "34dd691d-1a80-44bf-8693-51041053ab1d",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2023-01-30 15:50:03.954274: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 3830314 of 5941106\n",
      "2023-01-30 15:50:09.417413: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:417] Shuffle buffer filled.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "> Examples of filenames:\n",
      "\tYing hung boon sik\n",
      "\n",
      "\t Blue.Miracle.2021.1080p.NF.WEB-DL.DDP5.1.Atmos.x264-EVO\n",
      "\n",
      "\t House - [1x07] - Fidelity\n",
      "\n",
      "\n",
      "> Examples of movie titles:\n",
      "\tA Better Tomorrow\n",
      "\n",
      "\tBlue Miracle\n",
      "\n",
      "\tHouse Fidelity\n",
      "\n"
     ]
    }
   ],
   "source": [
    "for filenames, movietitles in train_ds.batch(3).take(1):\n",
    "    print('> Examples of filenames:')\n",
    "    for fn in filenames.numpy():\n",
    "        print(fn.decode('utf-8'))\n",
    "\n",
    "    print('\\n> Examples of movie titles:')\n",
    "    for mt in movietitles.numpy():\n",
    "        print(mt.decode('utf-8'))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "984e7d13-9cd2-4622-8ae7-a483f9665ba5",
   "metadata": {},
   "source": [
    "# 4. Tokenizing and batching data\n",
    "\n",
    "To avoid messing up with different ways to split tokens (especially in filenames), we use the `UnicodeCharTokenizer` so every character is going to be a token for us. The vocabulary size for this tokenizer is 2^16, ie. 65536."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "7f02ef44-8f0d-4c9b-923f-bc32365f45d3",
   "metadata": {},
   "outputs": [],
   "source": [
    "tokenizer = tensorflow_text.UnicodeCharTokenizer()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "028c37c9-722d-421b-82af-39242b912fe2",
   "metadata": {},
   "source": [
    "---\n",
    "Here we take a few movie titles and tokenize them. Note that all the lists of tokens start and end with the same tokens (9 and 10, which correspond to `\\t` and `\\n`)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "66d96d57-f9da-49f0-b8d6-2447d7ede131",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "> This is a batch of strings:\n",
      "\tA Better Tomorrow\n",
      "\n",
      "\tBlue Miracle\n",
      "\n",
      "\tHouse Fidelity\n",
      "\n",
      "> This is a padded-batch of token IDs:\n",
      "[9, 65, 32, 66, 101, 116, 116, 101, 114, 32, 84, 111, 109, 111, 114, 114, 111, 119, 10]\n",
      "[9, 66, 108, 117, 101, 32, 77, 105, 114, 97, 99, 108, 101, 10]\n",
      "[9, 72, 111, 117, 115, 101, 32, 70, 105, 100, 101, 108, 105, 116, 121, 10]\n"
     ]
    }
   ],
   "source": [
    "print('> This is a batch of strings:')\n",
    "for mt in movietitles.numpy():\n",
    "    print(mt.decode('utf-8'))\n",
    "    \n",
    "encoded = tokenizer.tokenize(movietitles)\n",
    "\n",
    "print('> This is a padded-batch of token IDs:')\n",
    "for row in encoded.to_list():\n",
    "    print(row)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ad30b5c8-8322-4ac3-8329-84516541ab9a",
   "metadata": {},
   "source": [
    "---\n",
    "\n",
    "If we detokenize the token lists, we get the original strings:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "380ab51c-2c96-4d52-9dfa-e03a75e4e817",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "> This is human-readable text:\n",
      "\tA Better Tomorrow\n",
      "\n",
      "\tBlue Miracle\n",
      "\n",
      "\tHouse Fidelity\n",
      "\n"
     ]
    }
   ],
   "source": [
    "round_trip = tokenizer.detokenize(encoded)\n",
    "\n",
    "print('> This is human-readable text:')\n",
    "for line in round_trip.numpy():\n",
    "    print(line.decode('utf-8'))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1196852b-df57-48ff-9175-1628ecb6b167",
   "metadata": {},
   "source": [
    "---\n",
    "The following code is used to estimate the amount of tokens a sample typically contains.\n",
    "This helps us choose a threshold above which we can cut our input strings.\n",
    "\n",
    "Note that there are a few commented lines in the code below. If you uncomment them, the maximum possible token value will be estimated from the data itself. This estimation was useful to understand all values were encoded with two bytes (16 bits, max value 65535), but it is quite slow so I left it commented to save some time... Do not worry if all you see is a 0 at the end :-)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "70f52ce1-d02e-4065-88f2-24f570df6c3e",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2023-01-30 16:02:29.546910: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 3884946 of 5941106\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "......."
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2023-01-30 16:02:34.762072: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:417] Shuffle buffer filled.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "...........................................................................................0\n"
     ]
    }
   ],
   "source": [
    "lengths = []\n",
    "\n",
    "npmax = 0\n",
    "\n",
    "for pt_examples, en_examples in train_ds.take(100000).batch(1024):\n",
    "    pt_tokens = tokenizer.tokenize(pt_examples)\n",
    "    lengths.append(pt_tokens.row_lengths())\n",
    "    # npmax_ = np.max([np.max(i) for i in pt_tokens])\n",
    "    # npmax = npmax_ if npmax_ > npmax else npmax\n",
    "\n",
    "    en_tokens = tokenizer.tokenize(en_examples)\n",
    "    lengths.append(en_tokens.row_lengths())\n",
    "    # npmax_ = np.max([np.max(i) for i in en_tokens])\n",
    "    # npmax = npmax_ if npmax_ > npmax else npmax\n",
    "    print('.', end='', flush=True)\n",
    "    \n",
    "print(npmax)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "c16451f1-4f52-4de0-96cc-71dfae8853a7",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "all_lengths = np.concatenate(lengths)\n",
    "\n",
    "plt.hist(all_lengths, np.linspace(0, 500, 101))\n",
    "plt.ylim(plt.ylim())\n",
    "max_length = max(all_lengths)\n",
    "plt.plot([max_length, max_length], plt.ylim())\n",
    "plt.title(f'Maximum tokens per example: {max_length}');"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b7836298-2e53-410b-a25b-63687fe679e6",
   "metadata": {},
   "source": [
    "As you can see in the plot above, while the max value of tokens per example is quite large (in my version of the dataset I get 693) almost all of the samples are below just a bit more than 100 tokens. For this reason, we can safely cut our strings to 128 characters and pad shorter strings with zeroes so that all the input token lists have the same length."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "c7a7122a-172f-4427-ab57-b1e764dea476",
   "metadata": {},
   "outputs": [],
   "source": [
    "MAX_TOKENS=128\n",
    "\n",
    "def prepare_batch(fn, mt):\n",
    "    fn = tokenizer.tokenize(fn)      # Output is ragged.\n",
    "    fn = fn[:, :MAX_TOKENS]    # Trim to MAX_TOKENS.\n",
    "    fn = fn.to_tensor()  # Convert to 0-padded dense Tensor\n",
    "\n",
    "    mt = tokenizer.tokenize(mt)\n",
    "    mt = mt[:, :(MAX_TOKENS+1)]\n",
    "    mt_inputs = mt[:, :-1].to_tensor()  # Drop the [END] tokens\n",
    "    mt_labels = mt[:, 1:].to_tensor()   # Drop the [START] tokens\n",
    "\n",
    "    return (fn, mt_inputs), mt_labels"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4868305e-ab4d-4631-a06c-eeddc051a1bb",
   "metadata": {},
   "source": [
    "---\n",
    "`make_batches` uses the `prepare_batch` function defined above to provide ready to use batches from a dataset (note the data is shuffled again, which is probably not needed):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "2cdb4df1-1ede-4101-a3fd-c8bc71541a90",
   "metadata": {},
   "outputs": [],
   "source": [
    "BUFFER_SIZE = 20000\n",
    "BATCH_SIZE = 64\n",
    "\n",
    "def make_batches(ds):\n",
    "    return (\n",
    "      ds\n",
    "      .shuffle(BUFFER_SIZE)\n",
    "      .batch(BATCH_SIZE)\n",
    "      .map(prepare_batch, tf.data.AUTOTUNE)\n",
    "      .prefetch(buffer_size=tf.data.AUTOTUNE))\n",
    "\n",
    "train_batches = make_batches(train_ds)\n",
    "val_batches = make_batches(val_ds)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1ec3d9d5-daba-4840-b701-c3d5c8ce92a8",
   "metadata": {},
   "source": [
    "# 5. Model definition\n",
    "\n",
    "The following code has been taken as is from Tensorflow's tutorial on transformers. I removed pictures and tests to save space, but I'd suggest you to [check it out](https://www.tensorflow.org/text/tutorials/transformer) if you want to understand how this model works!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "id": "a83796fe-ab6d-42d3-b4f4-af19552bae81",
   "metadata": {},
   "outputs": [],
   "source": [
    "def positional_encoding(length, depth):\n",
    "    depth = depth/2\n",
    "\n",
    "    positions = np.arange(length)[:, np.newaxis]     # (seq, 1)\n",
    "    depths = np.arange(depth)[np.newaxis, :]/depth   # (1, depth)\n",
    "\n",
    "    angle_rates = 1 / (10000**depths)         # (1, depth)\n",
    "    angle_rads = positions * angle_rates      # (pos, depth)\n",
    "\n",
    "    pos_encoding = np.concatenate(\n",
    "      [np.sin(angle_rads), np.cos(angle_rads)],\n",
    "      axis=-1) \n",
    "\n",
    "    return tf.cast(pos_encoding, dtype=tf.float32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "id": "8e02f248-6fbd-4b66-ba55-891bdde7259f",
   "metadata": {},
   "outputs": [],
   "source": [
    "class PositionalEmbedding(tf.keras.layers.Layer):\n",
    "    def __init__(self, vocab_size, d_model):\n",
    "        super().__init__()\n",
    "        self.d_model = d_model\n",
    "        self.embedding = tf.keras.layers.Embedding(vocab_size, d_model, mask_zero=True) \n",
    "        self.pos_encoding = positional_encoding(length=2048, depth=d_model)\n",
    "\n",
    "    def compute_mask(self, *args, **kwargs):\n",
    "        return self.embedding.compute_mask(*args, **kwargs)\n",
    "\n",
    "    def call(self, x):\n",
    "        length = tf.shape(x)[1]\n",
    "        x = self.embedding(x)\n",
    "        # This factor sets the relative scale of the embedding and positonal_encoding.\n",
    "        x *= tf.math.sqrt(tf.cast(self.d_model, tf.float32))\n",
    "        x = x + self.pos_encoding[tf.newaxis, :length, :]\n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "id": "a30fe748-b9d5-491e-a39d-0a992b8997a9",
   "metadata": {},
   "outputs": [],
   "source": [
    "class BaseAttention(tf.keras.layers.Layer):\n",
    "    def __init__(self, **kwargs):\n",
    "        super().__init__()\n",
    "        self.mha = tf.keras.layers.MultiHeadAttention(**kwargs)\n",
    "        self.layernorm = tf.keras.layers.LayerNormalization()\n",
    "        self.add = tf.keras.layers.Add()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "id": "cd491bcf-b58f-4204-855c-0122b848dfd2",
   "metadata": {},
   "outputs": [],
   "source": [
    "class CrossAttention(BaseAttention):\n",
    "    def call(self, x, context):\n",
    "        attn_output, attn_scores = self.mha(\n",
    "            query=x,\n",
    "            key=context,\n",
    "            value=context,\n",
    "            return_attention_scores=True)\n",
    "\n",
    "        # Cache the attention scores for plotting later.\n",
    "        self.last_attn_scores = attn_scores\n",
    "\n",
    "        x = self.add([x, attn_output])\n",
    "        x = self.layernorm(x)\n",
    "\n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "b0744b68-b939-4ad0-add4-6da6d9b4bed7",
   "metadata": {},
   "outputs": [],
   "source": [
    "class GlobalSelfAttention(BaseAttention):\n",
    "    def call(self, x):\n",
    "        attn_output = self.mha(\n",
    "            query=x,\n",
    "            value=x,\n",
    "            key=x)\n",
    "        x = self.add([x, attn_output])\n",
    "        x = self.layernorm(x)\n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "c4498f7e-900b-482e-a598-3acf293471ba",
   "metadata": {},
   "outputs": [],
   "source": [
    "class CausalSelfAttention(BaseAttention):\n",
    "    def call(self, x):\n",
    "        attn_output = self.mha(\n",
    "            query=x,\n",
    "            value=x,\n",
    "            key=x,\n",
    "            use_causal_mask = True)\n",
    "        x = self.add([x, attn_output])\n",
    "        x = self.layernorm(x)\n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "id": "79faaafe-ca11-41fc-b806-3eee60d188f5",
   "metadata": {},
   "outputs": [],
   "source": [
    "class FeedForward(tf.keras.layers.Layer):\n",
    "    def __init__(self, d_model, dff, dropout_rate=0.1):\n",
    "        super().__init__()\n",
    "        self.seq = tf.keras.Sequential([\n",
    "          tf.keras.layers.Dense(dff, activation='relu'),\n",
    "          tf.keras.layers.Dense(d_model),\n",
    "          tf.keras.layers.Dropout(dropout_rate)\n",
    "        ])\n",
    "        self.add = tf.keras.layers.Add()\n",
    "        self.layer_norm = tf.keras.layers.LayerNormalization()\n",
    "\n",
    "    def call(self, x):\n",
    "        x = self.add([x, self.seq(x)])\n",
    "        x = self.layer_norm(x) \n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "id": "722e6744-986d-4324-a852-95df1eb541c4",
   "metadata": {},
   "outputs": [],
   "source": [
    "class EncoderLayer(tf.keras.layers.Layer):\n",
    "    def __init__(self,*, d_model, num_heads, dff, dropout_rate=0.1):\n",
    "        super().__init__()\n",
    "\n",
    "        self.self_attention = GlobalSelfAttention(\n",
    "            num_heads=num_heads,\n",
    "            key_dim=d_model,\n",
    "            dropout=dropout_rate)\n",
    "\n",
    "        self.ffn = FeedForward(d_model, dff)\n",
    "\n",
    "    def call(self, x):\n",
    "        x = self.self_attention(x)\n",
    "        x = self.ffn(x)\n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "id": "4664fe5b-fe64-4b55-b3a9-d14bb407d49b",
   "metadata": {},
   "outputs": [],
   "source": [
    "class Encoder(tf.keras.layers.Layer):\n",
    "    def __init__(self, *, num_layers, d_model, num_heads,\n",
    "               dff, vocab_size, dropout_rate=0.1):\n",
    "        super().__init__()\n",
    "\n",
    "        self.d_model = d_model\n",
    "        self.num_layers = num_layers\n",
    "\n",
    "        self.pos_embedding = PositionalEmbedding(\n",
    "            vocab_size=vocab_size, d_model=d_model)\n",
    "\n",
    "        self.enc_layers = [\n",
    "            EncoderLayer(d_model=d_model,\n",
    "                         num_heads=num_heads,\n",
    "                         dff=dff,\n",
    "                         dropout_rate=dropout_rate)\n",
    "            for _ in range(num_layers)]\n",
    "        self.dropout = tf.keras.layers.Dropout(dropout_rate)\n",
    "\n",
    "    def call(self, x):\n",
    "        # `x` is token-IDs shape: (batch, seq_len)\n",
    "        x = self.pos_embedding(x)  # Shape `(batch_size, seq_len, d_model)`.\n",
    "\n",
    "        # Add dropout.\n",
    "        x = self.dropout(x)\n",
    "\n",
    "        for i in range(self.num_layers):\n",
    "            x = self.enc_layers[i](x)\n",
    "\n",
    "        return x  # Shape `(batch_size, seq_len, d_model)`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "id": "94957da1-0d63-44cc-9f67-18c69dc5d02d",
   "metadata": {},
   "outputs": [],
   "source": [
    "class DecoderLayer(tf.keras.layers.Layer):\n",
    "    def __init__(self,\n",
    "                   *,\n",
    "                   d_model,\n",
    "                   num_heads,\n",
    "                   dff,\n",
    "                   dropout_rate=0.1):\n",
    "        super(DecoderLayer, self).__init__()\n",
    "\n",
    "        self.causal_self_attention = CausalSelfAttention(\n",
    "            num_heads=num_heads,\n",
    "            key_dim=d_model,\n",
    "            dropout=dropout_rate)\n",
    "\n",
    "        self.cross_attention = CrossAttention(\n",
    "            num_heads=num_heads,\n",
    "            key_dim=d_model,\n",
    "            dropout=dropout_rate)\n",
    "\n",
    "        self.ffn = FeedForward(d_model, dff)\n",
    "\n",
    "    def call(self, x, context):\n",
    "        x = self.causal_self_attention(x=x)\n",
    "        x = self.cross_attention(x=x, context=context)\n",
    "\n",
    "        # Cache the last attention scores for plotting later\n",
    "        self.last_attn_scores = self.cross_attention.last_attn_scores\n",
    "\n",
    "        x = self.ffn(x)  # Shape `(batch_size, seq_len, d_model)`.\n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "id": "5922d961-b35b-4451-b036-48e710bf87c5",
   "metadata": {},
   "outputs": [],
   "source": [
    "class Decoder(tf.keras.layers.Layer):\n",
    "    def __init__(self, *, num_layers, d_model, num_heads, dff, vocab_size,\n",
    "                   dropout_rate=0.1):\n",
    "        super(Decoder, self).__init__()\n",
    "\n",
    "        self.d_model = d_model\n",
    "        self.num_layers = num_layers\n",
    "\n",
    "        self.pos_embedding = PositionalEmbedding(vocab_size=vocab_size,\n",
    "                                                 d_model=d_model)\n",
    "        self.dropout = tf.keras.layers.Dropout(dropout_rate)\n",
    "        self.dec_layers = [\n",
    "            DecoderLayer(d_model=d_model, num_heads=num_heads,\n",
    "                         dff=dff, dropout_rate=dropout_rate)\n",
    "            for _ in range(num_layers)]\n",
    "\n",
    "        self.last_attn_scores = None\n",
    "\n",
    "    def call(self, x, context):\n",
    "        # `x` is token-IDs shape (batch, target_seq_len)\n",
    "        x = self.pos_embedding(x)  # (batch_size, target_seq_len, d_model)\n",
    "\n",
    "        x = self.dropout(x)\n",
    "\n",
    "        for i in range(self.num_layers):\n",
    "            x  = self.dec_layers[i](x, context)\n",
    "\n",
    "        self.last_attn_scores = self.dec_layers[-1].last_attn_scores\n",
    "\n",
    "        # The shape of x is (batch_size, target_seq_len, d_model).\n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "id": "8d9cb377-bfed-465b-bb93-419927826e29",
   "metadata": {},
   "outputs": [],
   "source": [
    "class Transformer(tf.keras.Model):\n",
    "    def __init__(self, *, num_layers, d_model, num_heads, dff,\n",
    "                   input_vocab_size, target_vocab_size, dropout_rate=0.1):\n",
    "        super().__init__()\n",
    "        self.encoder = Encoder(num_layers=num_layers, d_model=d_model,\n",
    "                               num_heads=num_heads, dff=dff,\n",
    "                               vocab_size=input_vocab_size,\n",
    "                               dropout_rate=dropout_rate)\n",
    "\n",
    "        self.decoder = Decoder(num_layers=num_layers, d_model=d_model,\n",
    "                               num_heads=num_heads, dff=dff,\n",
    "                               vocab_size=target_vocab_size,\n",
    "                               dropout_rate=dropout_rate)\n",
    "\n",
    "        self.final_layer = tf.keras.layers.Dense(target_vocab_size)\n",
    "\n",
    "    def call(self, inputs):\n",
    "        # To use a Keras model with `.fit` you must pass all your inputs in the\n",
    "        # first argument.\n",
    "        context, x  = inputs\n",
    "\n",
    "        context = self.encoder(context)  # (batch_size, context_len, d_model)\n",
    "        x = self.decoder(x, context)  # (batch_size, target_len, d_model)\n",
    "\n",
    "        # Final linear layer output.\n",
    "        logits = self.final_layer(x)  # (batch_size, target_len, target_vocab_size)\n",
    "\n",
    "        try:\n",
    "          # Drop the keras mask, so it doesn't scale the losses/metrics.\n",
    "          # b/250038731\n",
    "            del logits._keras_mask\n",
    "        except AttributeError:\n",
    "            pass\n",
    "\n",
    "        # Return the final output and the attention weights.\n",
    "        return logits"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "id": "ae5893c3-a97c-4663-b4d5-5045a5c4a9cf",
   "metadata": {},
   "outputs": [],
   "source": [
    "num_layers = 4\n",
    "d_model = 128\n",
    "dff = 512\n",
    "num_heads = 8\n",
    "dropout_rate = 0.1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "id": "7f335ca7-24f5-4170-b8cd-1059377f237a",
   "metadata": {},
   "outputs": [],
   "source": [
    "transformer = Transformer(\n",
    "    num_layers=num_layers,\n",
    "    d_model=d_model,\n",
    "    num_heads=num_heads,\n",
    "    dff=dff,\n",
    "    input_vocab_size=65535,\n",
    "    target_vocab_size=65535,\n",
    "    dropout_rate=dropout_rate)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "id": "6ff4df29-7c9e-4208-a833-5c78bda83823",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"transformer\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " encoder (Encoder)           multiple                  11027328  \n",
      "                                                                 \n",
      " decoder (Decoder)           multiple                  13138304  \n",
      "                                                                 \n",
      " dense_16 (Dense)            multiple                  8454015   \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 32,619,647\n",
      "Trainable params: 32,619,647\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "transformer.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2400cd1f-e281-4c6e-8fe4-0318a04824a0",
   "metadata": {},
   "source": [
    "# 6. Define LR schedule, optimizer, and metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "id": "398b1470-38fa-4ae3-97c4-c83922019d4e",
   "metadata": {},
   "outputs": [],
   "source": [
    "class CustomSchedule(tf.keras.optimizers.schedules.LearningRateSchedule):\n",
    "    def __init__(self, d_model, warmup_steps=4000):\n",
    "        super().__init__()\n",
    "\n",
    "        self.d_model = d_model\n",
    "        self.d_model = tf.cast(self.d_model, tf.float32)\n",
    "\n",
    "        self.warmup_steps = warmup_steps\n",
    "\n",
    "    def __call__(self, step):\n",
    "        step = tf.cast(step, dtype=tf.float32)\n",
    "        arg1 = tf.math.rsqrt(step)\n",
    "        arg2 = step * (self.warmup_steps ** -1.5)\n",
    "\n",
    "        return tf.math.rsqrt(self.d_model) * tf.math.minimum(arg1, arg2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "id": "eb3f42cc-595f-47ea-a34c-45518dbb1c17",
   "metadata": {},
   "outputs": [],
   "source": [
    "learning_rate = CustomSchedule(d_model)\n",
    "\n",
    "optimizer = tf.keras.optimizers.Adam(learning_rate, beta_1=0.9, beta_2=0.98,\n",
    "                                     epsilon=1e-9)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "id": "a91c586b-f5dd-48f2-a3a8-080e183b7009",
   "metadata": {},
   "outputs": [],
   "source": [
    "# plt.plot(learning_rate(tf.range(80000, dtype=tf.float32)))\n",
    "# plt.ylabel('Learning Rate')\n",
    "# plt.xlabel('Train Step')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "id": "2a196905-9ce9-41df-a2cc-b978559d1a36",
   "metadata": {},
   "outputs": [],
   "source": [
    "def masked_loss(label, pred):\n",
    "    mask = label != 0\n",
    "    loss_object = tf.keras.losses.SparseCategoricalCrossentropy(\n",
    "    from_logits=True, reduction='none')\n",
    "    loss = loss_object(label, pred)\n",
    "\n",
    "    mask = tf.cast(mask, dtype=loss.dtype)\n",
    "    loss *= mask\n",
    "\n",
    "    loss = tf.reduce_sum(loss)/tf.reduce_sum(mask)\n",
    "    return loss\n",
    "\n",
    "\n",
    "def masked_accuracy(label, pred):\n",
    "    pred = tf.argmax(pred, axis=2)\n",
    "    label = tf.cast(label, pred.dtype)\n",
    "    match = label == pred\n",
    "\n",
    "    mask = label != 0\n",
    "\n",
    "    match = match & mask\n",
    "\n",
    "    match = tf.cast(match, dtype=tf.float32)\n",
    "    mask = tf.cast(mask, dtype=tf.float32)\n",
    "    return tf.reduce_sum(match)/tf.reduce_sum(mask)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c0897ea3-4462-42cc-990e-526c8a6be7ba",
   "metadata": {},
   "source": [
    "# 7. Train model\n",
    "\n",
    "Uncomment the `transformer.fit` cell below to train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "id": "53c2ee15-d52b-4156-a165-4c97217d2b5c",
   "metadata": {},
   "outputs": [],
   "source": [
    "transformer.compile(\n",
    "    loss=masked_loss,\n",
    "    optimizer=optimizer,\n",
    "    metrics=[masked_accuracy])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 239,
   "id": "09cb0c96-0783-486d-9d28-8aadea3b9369",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/3\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2023-01-28 00:01:41.661644: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 1497349 of 5879992\n",
      "2023-01-28 00:01:51.661641: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 2993466 of 5879992\n",
      "2023-01-28 00:02:01.661642: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 4489460 of 5879992\n",
      "2023-01-28 00:02:10.943448: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:417] Shuffle buffer filled.\n",
      "2023-01-28 00:02:10.943517: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 1 of 20000\n",
      "2023-01-28 00:02:10.943552: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 2 of 20000\n",
      "2023-01-28 00:02:10.943566: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 3 of 20000\n",
      "2023-01-28 00:02:11.057148: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:417] Shuffle buffer filled.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "82688/82688 [==============================] - ETA: 0s - loss: 0.6999 - masked_accuracy: 0.8064"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2023-01-28 03:04:23.700289: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 1520983 of 5879992\n",
      "2023-01-28 03:04:33.700287: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 3044188 of 5879992\n",
      "2023-01-28 03:04:43.700287: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 4568413 of 5879992\n",
      "2023-01-28 03:04:52.321352: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:417] Shuffle buffer filled.\n",
      "2023-01-28 03:05:02.839160: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 1 of 20000\n",
      "2023-01-28 03:05:02.839229: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 2 of 20000\n",
      "2023-01-28 03:05:02.839244: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 3 of 20000\n",
      "2023-01-28 03:05:02.839258: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 4 of 20000\n",
      "2023-01-28 03:05:02.981219: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:417] Shuffle buffer filled.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "82688/82688 [==============================] - 11697s 141ms/step - loss: 0.6999 - masked_accuracy: 0.8064 - val_loss: 0.5099 - val_masked_accuracy: 0.8514\n",
      "Epoch 2/3\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2023-01-28 03:16:20.184706: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 1518701 of 5879992\n",
      "2023-01-28 03:16:30.184699: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 3037721 of 5879992\n",
      "2023-01-28 03:16:40.184696: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 4560204 of 5879992\n",
      "2023-01-28 03:16:48.855674: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:417] Shuffle buffer filled.\n",
      "2023-01-28 03:16:48.855738: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 1 of 20000\n",
      "2023-01-28 03:16:48.855771: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 2 of 20000\n",
      "2023-01-28 03:16:48.855784: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 3 of 20000\n",
      "2023-01-28 03:16:48.966671: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:417] Shuffle buffer filled.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "82688/82688 [==============================] - ETA: 0s - loss: 0.5279 - masked_accuracy: 0.8457"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2023-01-28 06:18:32.447848: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 1562013 of 5879992\n",
      "2023-01-28 06:18:42.447848: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 3019876 of 5879992\n",
      "2023-01-28 06:18:52.447855: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 4451715 of 5879992\n",
      "2023-01-28 06:19:02.447846: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 5795953 of 5879992\n",
      "2023-01-28 06:19:03.222693: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:417] Shuffle buffer filled.\n",
      "2023-01-28 06:19:14.552979: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 1 of 20000\n",
      "2023-01-28 06:19:14.553046: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 2 of 20000\n",
      "2023-01-28 06:19:14.553063: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 3 of 20000\n",
      "2023-01-28 06:19:14.553077: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 4 of 20000\n",
      "2023-01-28 06:19:14.553091: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 5 of 20000\n",
      "2023-01-28 06:19:14.696920: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:417] Shuffle buffer filled.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "82688/82688 [==============================] - 11653s 140ms/step - loss: 0.5279 - masked_accuracy: 0.8457 - val_loss: 0.4587 - val_masked_accuracy: 0.8658\n",
      "Epoch 3/3\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2023-01-28 06:30:34.930874: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 1496686 of 5879992\n",
      "2023-01-28 06:30:44.930873: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 2990353 of 5879992\n",
      "2023-01-28 06:30:54.930875: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 4488804 of 5879992\n",
      "2023-01-28 06:31:04.211045: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:417] Shuffle buffer filled.\n",
      "2023-01-28 06:31:04.211115: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 1 of 20000\n",
      "2023-01-28 06:31:04.211151: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 2 of 20000\n",
      "2023-01-28 06:31:04.211165: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 3 of 20000\n",
      "2023-01-28 06:31:04.324165: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:417] Shuffle buffer filled.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "82688/82688 [==============================] - ETA: 0s - loss: 0.4972 - masked_accuracy: 0.8543"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2023-01-28 09:33:04.211649: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 1515884 of 5879992\n",
      "2023-01-28 09:33:14.211644: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 3028246 of 5879992\n",
      "2023-01-28 09:33:24.211648: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 4545743 of 5879992\n",
      "2023-01-28 09:33:33.002748: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:417] Shuffle buffer filled.\n",
      "2023-01-28 09:33:43.644815: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 1 of 20000\n",
      "2023-01-28 09:33:43.644881: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 2 of 20000\n",
      "2023-01-28 09:33:43.644897: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 3 of 20000\n",
      "2023-01-28 09:33:43.644911: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 4 of 20000\n",
      "2023-01-28 09:33:43.785794: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:417] Shuffle buffer filled.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "82688/82688 [==============================] - 11666s 141ms/step - loss: 0.4972 - masked_accuracy: 0.8543 - val_loss: 0.4321 - val_masked_accuracy: 0.8734\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7f3ff014cd60>"
      ]
     },
     "execution_count": 239,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# transformer.fit(train_batches,\n",
    "#                 epochs=3,\n",
    "#                 validation_data=val_batches)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d81a2e37-db95-43ed-89b3-2ed2ab062977",
   "metadata": {},
   "source": [
    "# 8. Save/load model\n",
    "\n",
    "The cells below are used to save (and optionally load a saved) model and then load it again (uncomment those you want to use)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "id": "6ed0fbb5-5f4d-4a11-8a57-2334151f3fe3",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.checkpoint.checkpoint.CheckpointLoadStatus at 0x7feb9dbe3fd0>"
      ]
     },
     "execution_count": 67,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Use saved_model.save to save the whole model\n",
    "# tf.saved_model.save(transformer, export_dir='transformer')\n",
    "\n",
    "# Save the model weights only\n",
    "# transformer.save_weights(\"trasformer_weights/epoch03.ckpt\")\n",
    "\n",
    "transformer.load_weights(\"./transformer_weights/epoch03.ckpt\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "id": "766f195c-5c7b-4297-8c7b-4b1910f0de85",
   "metadata": {},
   "outputs": [],
   "source": [
    "class Translator(tf.Module):\n",
    "    def __init__(self, tokenizers, transformer):\n",
    "        self.tokenizer = tokenizer\n",
    "        self.transformer = transformer\n",
    "\n",
    "    def __call__(self, sentence, max_length=MAX_TOKENS):\n",
    "\n",
    "        assert isinstance(sentence, tf.Tensor)\n",
    "        if len(sentence.shape) == 0:\n",
    "            sentence = sentence[tf.newaxis]\n",
    "\n",
    "        sentence = self.tokenizer.tokenize(sentence).to_tensor()\n",
    "\n",
    "        encoder_input = sentence\n",
    "\n",
    "        start_end = self.tokenizer.tokenize(['\\t\\n'])[0]\n",
    "        start = start_end[0][tf.newaxis]\n",
    "        end = start_end[1][tf.newaxis]\n",
    "\n",
    "        # `tf.TensorArray` is required here (instead of a Python list), so that the\n",
    "        # dynamic-loop can be traced by `tf.function`.\n",
    "        output_array = tf.TensorArray(dtype=tf.int32, size=0, dynamic_size=True)\n",
    "        output_array = output_array.write(0, start)\n",
    "\n",
    "        for i in tf.range(max_length):\n",
    "            output = tf.transpose(output_array.stack())\n",
    "            predictions = self.transformer([encoder_input, output], training=False)\n",
    "\n",
    "            # Select the last token from the `seq_len` dimension.\n",
    "            predictions = predictions[:, -1:, :]  # Shape `(batch_size, 1, vocab_size)`.\n",
    "\n",
    "            predicted_id = tf.cast(tf.argmax(predictions, axis=-1), tf.int32)\n",
    "\n",
    "            # Concatenate the `predicted_id` to the output which is given to the\n",
    "            # decoder as its input.\n",
    "            output_array = output_array.write(i+1, predicted_id[0], tf.int32)\n",
    "\n",
    "            if predicted_id == end:\n",
    "                break\n",
    "\n",
    "        output = tf.transpose(output_array.stack())\n",
    "        # The output shape is `(1, tokens)`.\n",
    "        text = tokenizer.detokenize(output)[0]  # Shape: `()`.\n",
    "\n",
    "        tokens = tf.convert_to_tensor(list(text.numpy().decode(\"utf-8\"))) #tokenizers.en.lookup(output)[0]\n",
    "\n",
    "        # `tf.function` prevents us from using the attention_weights that were\n",
    "        # calculated on the last iteration of the loop.\n",
    "        # So, recalculate them outside the loop.\n",
    "        self.transformer([encoder_input, output[:,:-1]], training=False)\n",
    "        attention_weights = self.transformer.decoder.last_attn_scores\n",
    "\n",
    "        return text, tokens, attention_weights\n",
    "    \n",
    "\n",
    "def print_translation(filename, predicted_movietitle, ground_truth):\n",
    "    print(f'{\"Input:\":15s}: {filename}')\n",
    "    print(f'{\"Prediction\":15s}: {predicted_movietitle.numpy().decode(\"utf-8\")}')\n",
    "    print(f'{\"Ground truth\":15s}: {ground_truth}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "id": "dd1abe50-3534-4f6d-b01b-5f2775984a99",
   "metadata": {},
   "outputs": [],
   "source": [
    "translator = Translator(tokenizer, transformer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "id": "e0bc9943-7783-417f-96d4-ebbcdbdd7813",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Input:         : Hellhole.2022.720p.WEBDL.English.Dubbed\n",
      "Prediction     : \tHellhole\n",
      "\n",
      "Ground truth   : Hellhole\n"
     ]
    }
   ],
   "source": [
    "filename = 'Hellhole.2022.720p.WEBDL.English.Dubbed'\n",
    "movietitle = 'Hellhole'\n",
    "\n",
    "predicted_movietitle, predicted_tokens, attention_weights = translator(tf.constant(filename))\n",
    "print_translation(filename, predicted_movietitle, movietitle)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 127,
   "id": "9f230a71-077c-48f6-90ba-c0258e11a451",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2023-01-30 22:19:12.033181: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:392] Filling up shuffle buffer (this may take a while): 3055120 of 5941106\n",
      "2023-01-30 22:19:21.593444: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:417] Shuffle buffer filled.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Input:         : b'\\tMerlin - 03x12 - The coming of arthur.FOV\\n'\n",
      "Prediction     : \tMerlin The Coming of Arthur\n",
      "\n",
      "Ground truth   : b'\\tMerlin The Coming of Arthur: Part One\\n'\n",
      "---\n",
      "Input:         : b'\\t 30 Rock - Rockefeller Plaza 30 - Napisy PL Sezon 6 HDTV XviD S06E12\\n'\n",
      "Prediction     : \t30 Rock Napisy Planet of the Sun\n",
      "\n",
      "Ground truth   : b\"\\t30 Rock St. Patrick's Day\\n\"\n",
      "---\n",
      "Input:         : b'\\t Game.of.Silence.S01E09.WEB-DL.XviD-FUM[ettv]\\n'\n",
      "Prediction     : \tGame of Silence The Secret of Silence\n",
      "\n",
      "Ground truth   : b'\\tGame of Silence The Truth\\n'\n",
      "---\n",
      "Input:         : b'\\t Six.Feet.Under.S04E01.DVDRiP.x264-DBO\\n'\n",
      "Prediction     : \tSix Feet Under The Secret of the Sin\n",
      "\n",
      "Ground truth   : b'\\tSix Feet Under Falling into Place\\n'\n",
      "---\n",
      "Input:         : b'\\t720p.BluRay.x264-SPARKS (COLOR Hearing Impaired)\\n'\n",
      "Prediction     : \tThe Strangers\n",
      "\n",
      "Ground truth   : b'\\tA Million Ways to Die in the West\\n'\n",
      "---\n",
      "Input:         : b'\\tBetter.Call.Saul.S04E01.Smoke.720p.NF.WEBRip.DDP5.1.x264-NTb\\n'\n",
      "Prediction     : \tBetter Call Saul Smoke\n",
      "\n",
      "Ground truth   : b'\\tBetter Call Saul Smoke\\n'\n",
      "---\n",
      "Input:         : b'\\tWonder Woman 2x16 - Death in Disguise.mkv\\n'\n",
      "Prediction     : \tWonder Woman Death in Disguise\n",
      "\n",
      "Ground truth   : b'\\tWonder Woman Death in Disguise\\n'\n",
      "---\n",
      "Input:         : b'\\tHist\\xc3\\xb2ria de la meva mort (Story of my death)\\n'\n",
      "Prediction     : \tStory of My Death\n",
      "\n",
      "Ground truth   : b'\\tHistoria de la meva mort\\n'\n",
      "---\n",
      "Input:         : b'\\t 24.S06E04.Day.6_.9_00.A.M..-.10_00.A.M..720p.WEB-DL.DD5.1.H.264\\n'\n",
      "Prediction     : \t24 Day 6: 9:00 a.m.-10:00 a.m.\n",
      "\n",
      "Ground truth   : b'\\t24 Day 6: 9:00 a.m.-10:00 a.m.\\n'\n",
      "---\n",
      "Input:         : b'\\tHDTV.x264-RiVER\\n'\n",
      "Prediction     : \tThe Big Bang Theory The Convergence Interview\n",
      "\n",
      "Ground truth   : b'\\tBluestone 42 Episode #1.8\\n'\n",
      "---\n"
     ]
    }
   ],
   "source": [
    "for filenames, movietitles in test_ds.batch(10).take(1):\n",
    "    for i in range(len(filenames)):\n",
    "        filename = filenames[i]\n",
    "        movietitle = movietitles[i]\n",
    "\n",
    "        predicted_movietitle, predicted_tokens, attention_weights = translator(tf.constant(filename))\n",
    "        print_translation(filename, predicted_movietitle, movietitle)        \n",
    "        print(\"---\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ae22840c-5de5-49bd-a97c-4fb68871bfa1",
   "metadata": {},
   "source": [
    "# Show attention weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "id": "1b965c9e-ada9-4c2e-b616-504d3ad9b4b4",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Input:         : \t Elite.S02E03.36.hours.missing.720p.NF.WEB DL.DDP5.1.x264 MZABI\n",
      "\n",
      "Prediction     : \tElite 36 horas desaparecido\n",
      "\n",
      "Ground truth   : \tElite 36 horas desaparecido\n",
      "\n"
     ]
    }
   ],
   "source": [
    "sentence = '\\t Elite.S02E03.36.hours.missing.720p.NF.WEB DL.DDP5.1.x264 MZABI\\n'\n",
    "ground_truth = \"\\tElite 36 horas desaparecido\\n\"\n",
    "\n",
    "translated_text, translated_tokens, attention_weights = translator(\n",
    "    tf.constant(sentence))\n",
    "print_translation(sentence, translated_text, ground_truth)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "id": "0839611c-e113-48eb-9787-e1b0b16350f2",
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_attention_head(in_tokens, translated_tokens, attention):\n",
    "    # The model didn't generate `<START>` in the output. Skip it.\n",
    "    translated_tokens = translated_tokens[1:]\n",
    "\n",
    "    ax = plt.gca()\n",
    "    # ax.figure.set_size_inches(10, 10)\n",
    "    ax.matshow(attention)\n",
    "    ax.set_xticks(range(len(in_tokens)))\n",
    "    ax.set_yticks(range(len(translated_tokens)))\n",
    "\n",
    "    labels = [label.decode('utf-8') for label in in_tokens.numpy()]\n",
    "    ax.set_xticklabels(\n",
    "      labels) #, rotation=90)\n",
    "\n",
    "    labels = [label.decode('utf-8') for label in translated_tokens.numpy()]\n",
    "    ax.set_yticklabels(labels)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "id": "1b05d189-ddb8-4f6b-80b9-0c7563711afc",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "TensorShape([28, 65])"
      ]
     },
     "execution_count": 96,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "head = 0\n",
    "# Shape: `(batch=1, num_heads, seq_len_q, seq_len_k)`.\n",
    "attention_heads = tf.squeeze(attention_weights, 0)\n",
    "attention = attention_heads[head]\n",
    "attention.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "id": "e0a59183-27d9-41d1-b054-2bc486209e0d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(65,), dtype=string, numpy=\n",
       "array([b'\\t', b' ', b'E', b'l', b'i', b't', b'e', b'.', b'S', b'0', b'2',\n",
       "       b'E', b'0', b'3', b'.', b'3', b'6', b'.', b'h', b'o', b'u', b'r',\n",
       "       b's', b'.', b'm', b'i', b's', b's', b'i', b'n', b'g', b'.', b'7',\n",
       "       b'2', b'0', b'p', b'.', b'N', b'F', b'.', b'W', b'E', b'B', b' ',\n",
       "       b'D', b'L', b'.', b'D', b'D', b'P', b'5', b'.', b'1', b'.', b'x',\n",
       "       b'2', b'6', b'4', b' ', b'M', b'Z', b'A', b'B', b'I', b'\\n'],\n",
       "      dtype=object)>"
      ]
     },
     "execution_count": 97,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "in_tokens = tf.convert_to_tensor(list(sentence))\n",
    "in_tokens"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "id": "0ea3d18c-e7d4-42de-a461-0e3ae72c7ebc",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(29,), dtype=string, numpy=\n",
       "array([b'\\t', b'E', b'l', b'i', b't', b'e', b' ', b'3', b'6', b' ', b'h',\n",
       "       b'o', b'r', b'a', b's', b' ', b'd', b'e', b's', b'a', b'p', b'a',\n",
       "       b'r', b'e', b'c', b'i', b'd', b'o', b'\\n'], dtype=object)>"
      ]
     },
     "execution_count": 98,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "translated_tokens"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "id": "b0a2fa61-55ef-415e-902a-e737a097613e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_attention_head(in_tokens, translated_tokens, attention)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 120,
   "id": "a7a77125-c2b0-486e-b10b-757d472389b2",
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_attention_weights(sentence, translated_tokens, attention_heads):\n",
    "    in_tokens = tf.convert_to_tensor(list(sentence))\n",
    "    \n",
    "    fig = plt.figure(figsize=(10, 10))\n",
    "\n",
    "    for h, head in enumerate(attention_heads):\n",
    "        ax = fig.add_subplot(4, 2, h+1)\n",
    "        plot_attention_head(in_tokens, translated_tokens, head)\n",
    "        ax.set_xlabel(f'Head {h+1}')\n",
    "\n",
    "    plt.tight_layout()\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 121,
   "id": "04038e3f-f3e4-4cf6-8891-77d80d4d40e3",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/3w/1spf9bkn2gjbc8026chntkb80000gn/T/ipykernel_56483/1771425617.py:11: UserWarning: Glyph 9 (\t) missing from current font.\n",
      "  plt.tight_layout()\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1000x1000 with 8 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_attention_weights(sentence,\n",
    "                       translated_tokens,\n",
    "                       attention_weights[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 122,
   "id": "e4ce72cf-46bb-4a8e-b845-7ccb02436967",
   "metadata": {},
   "outputs": [],
   "source": [
    "class ExportTranslator(tf.Module):\n",
    "    def __init__(self, translator):\n",
    "        self.translator = translator\n",
    "\n",
    "    # @tf.function(input_signature=[tf.TensorSpec(shape=[], dtype=tf.string)])\n",
    "    def __call__(self, sentence):\n",
    "        (result,\n",
    "         tokens,\n",
    "         attention_weights) = self.translator(tf.constant(sentence), max_length=MAX_TOKENS)\n",
    "\n",
    "        return result"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 123,
   "id": "f0734152-9902-4839-8c5f-99a08dcde958",
   "metadata": {},
   "outputs": [],
   "source": [
    "my_translator = ExportTranslator(translator)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "id": "16c00da2-8977-44b6-b404-c6be83c5c61d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "b'\\tBig Mouth\\n'"
      ]
     },
     "execution_count": 126,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "my_translator(b'\\tBig.Mouth.2022.1080p.DSNP.WEB-DL.AAC2.0.H.264-ECLiPSE\\n').numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 393,
   "id": "9d30c210-6300-474f-88dd-ed125adabea6",
   "metadata": {},
   "outputs": [],
   "source": [
    "### A couple of ways to save the translator model\n",
    "\n",
    "# 1) use tf.saved_model\n",
    "# tf.saved_model.save(my_translator, export_dir='my_translator')\n",
    "\n",
    "# 2) use model.save\n",
    "##   NOTE: This one breaks due to custom LR schedule\n",
    "##   TODO: fix by defining config as\n",
    "##   def get_config(self):\n",
    "##      config = {\n",
    "##      'd_model': self.d_model,\n",
    "##      'warmup_steps': self.warmup_steps,\n",
    "##\n",
    "##       }\n",
    "##      return config\n",
    "\n",
    "# !mkdir -p \"saved_model\"\n",
    "# transformer.save(\"saved_model/epoch03\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "fd6db193-dc4d-447a-ad5c-6ae21aaa5444",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.9.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}