Task 2/Eng_Spanish.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Machine Translation Project (English to Spanish)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pathlib\n",
    "import random\n",
    "import string\n",
    "import tensorflow.strings as tf_strings\n",
    "import tensorflow.data as tf_data\n",
    "import re\n",
    "from keras.layers import TextVectorization\n",
    "import keras\n",
    "import tensorflow as tf\n",
    "from keras import layers\n",
    "import json"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Verify access to the GPU"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[name: \"/device:CPU:0\"\n",
      "device_type: \"CPU\"\n",
      "memory_limit: 268435456\n",
      "locality {\n",
      "}\n",
      "incarnation: 11075625745611853481\n",
      "xla_global_id: -1\n",
      ", name: \"/device:GPU:0\"\n",
      "device_type: \"GPU\"\n",
      "memory_limit: 1733715559\n",
      "locality {\n",
      "  bus_id: 1\n",
      "  links {\n",
      "  }\n",
      "}\n",
      "incarnation: 17906485139926134931\n",
      "physical_device_desc: \"device: 0, name: NVIDIA GeForce RTX 2050, pci bus id: 0000:01:00.0, compute capability: 8.6\"\n",
      "xla_global_id: 416903419\n",
      "]\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.python.client import device_lib\n",
    "print(device_lib.list_local_devices())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Download and prepare the data\n",
    "source :\"http://storage.googleapis.com/download.tensorflow.org/data/spa-eng.zip\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "text_file =  keras.utils.get_file(\n",
    "    fname = \"spa-eng.zip\",\n",
    "    origin = \"http://storage.googleapis.com/download.tensorflow.org/data/spa-eng.zip\",\n",
    "    extract = True,\n",
    ")\n",
    "\n",
    "text_file = pathlib.Path(text_file).parent / \"spa-eng\" / \"spa.txt\"\n",
    "\n",
    "with open(text_file, \"r\") as f:\n",
    "    lines = f.read().split(\"\\n\")[:-1]\n",
    "    \n",
    "text_pairs = []\n",
    "\n",
    "for line in lines:\n",
    "    eng, spa = line.split(\"\\t\")\n",
    "    spa = \"[start] \" + spa + \" [end]\"\n",
    "    text_pairs.append((eng, spa))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "random.shuffle(text_pairs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('Tom came here on his own.', '[start] Tom vino aquí por sí solo. [end]')\n",
      "(\"I don't want Tom to see me naked.\", '[start] No quiero que Tom me vea desnuda. [end]')\n",
      "('Stand back!', '[start] ¡Retrocede! [end]')\n",
      "(\"He's quite formal when he meets a stranger.\", '[start] Es muy ceremonioso cuando le presentan una persona desconocida. [end]')\n",
      "(\"I'm stuck in a traffic jam.\", '[start] Estoy atrapado en un atasco. [end]')\n"
     ]
    }
   ],
   "source": [
    "for i in range(5):\n",
    "    print(text_pairs[i])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Structure of the Dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "118964 total pairs\n",
      "83276 training pairs\n",
      "17844 validation pairs\n",
      "17844 test pairs\n"
     ]
    }
   ],
   "source": [
    "num_val_samples = int(0.15 * len(text_pairs))\n",
    "num_train_samples = len(text_pairs) - 2 * num_val_samples\n",
    "train_pairs = text_pairs[:num_train_samples]\n",
    "val_pairs = text_pairs[num_train_samples:num_train_samples + num_val_samples]\n",
    "test_pairs = text_pairs[num_train_samples + num_val_samples:]\n",
    "\n",
    "print(f\"{len(text_pairs)} total pairs\")\n",
    "print(f\"{len(train_pairs)} training pairs\")\n",
    "print(f\"{len(val_pairs)} validation pairs\")\n",
    "print(f\"{len(test_pairs)} test pairs\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# parameters\n",
    "strip_chars = string.punctuation + \"¿\"\n",
    "strip_chars = strip_chars.replace(\"[\", \"\")\n",
    "strip_chars = strip_chars.replace(\"]\", \"\")\n",
    "\n",
    "vocab_size = 15000\n",
    "sequence_length = 20\n",
    "batch_size = 64"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Vectorize the data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "def custom_standardization(input_string):\n",
    "    lowercase = tf_strings.lower(input_string)\n",
    "    return tf_strings.regex_replace(lowercase, f\"[{re.escape(strip_chars)}]\", \"\")\n",
    "\n",
    "# vectorization\n",
    "eng_vectorization = TextVectorization(\n",
    "    max_tokens = vocab_size,\n",
    "    output_mode = \"int\",\n",
    "    output_sequence_length = sequence_length,\n",
    ")\n",
    "\n",
    "spa_vectorization = TextVectorization(\n",
    "    max_tokens = vocab_size,\n",
    "    output_mode = \"int\",\n",
    "    output_sequence_length = sequence_length + 1,\n",
    "    standardize = custom_standardization,\n",
    ")\n",
    "\n",
    "train_eng_texts = [pair[0] for pair in train_pairs]\n",
    "train_spa_texts = [pair[1] for pair in train_pairs]\n",
    "\n",
    "eng_vectorization.adapt(train_eng_texts)\n",
    "spa_vectorization.adapt(train_spa_texts)\n",
    "\n",
    "#save the vectorization layers\n",
    "eng_vectorization_config = eng_vectorization.get_config()\n",
    "eng_vectorization_config.pop('standardize', None)\n",
    "eng_vocab = eng_vectorization.get_vocabulary()\n",
    "with open('eng_vectorization_config.json', 'w', encoding='utf-8') as f:\n",
    "    json.dump(eng_vectorization_config, f)\n",
    "    \n",
    "with open('eng_vocab.json', 'w', encoding='utf-8') as f:\n",
    "    json.dump(eng_vocab, f)\n",
    "    \n",
    "spa_vectorization_config = spa_vectorization.get_config()\n",
    "spa_vectorization_config.pop('standardize', None)\n",
    "spa_vocab = spa_vectorization.get_vocabulary()\n",
    "with open('spa_vectorization_config.json', 'w', encoding='utf-8') as f:\n",
    "    json.dump(spa_vectorization_config, f)\n",
    "    \n",
    "with open('spa_vocab.json', 'w', encoding='utf-8') as f:\n",
    "    json.dump(spa_vocab, f)\n",
    "    \n",
    "\n",
    "def format_dataset(eng, spa):\n",
    "    eng = eng_vectorization(eng)\n",
    "    spa = spa_vectorization(spa)\n",
    "    return (\n",
    "        {\n",
    "            \"encoder_inputs\": eng,\n",
    "            \"decoder_inputs\": spa[:, :-1],\n",
    "        },\n",
    "        spa[:, 1:],\n",
    "    )\n",
    "    \n",
    "def make_dataset(pairs):\n",
    "    eng_texts, spa_texts = zip(*pairs)\n",
    "    eng_texts = list(eng_texts)\n",
    "    spa_texts = list(spa_texts)\n",
    "    dataset = tf_data.Dataset.from_tensor_slices((eng_texts, spa_texts))\n",
    "    dataset = dataset.batch(batch_size)\n",
    "    dataset = dataset.map(format_dataset)\n",
    "    return dataset.cache().shuffle(2048).prefetch(16)\n",
    "\n",
    "train_ds = make_dataset(train_pairs)\n",
    "val_ds = make_dataset(val_pairs)\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(64, 20)\n",
      "(64, 20)\n"
     ]
    }
   ],
   "source": [
    "for inputs,targets in train_ds.take(1):\n",
    "    print(inputs[\"encoder_inputs\"].shape)\n",
    "    print(targets.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Model Architecture\n",
    "![Encoder-Decoder](images/encoder-decoder-context.png)\n",
    "![Encoder-Decoder](images/encoder-decoder-translation.png)\n",
    "![Attention Mechanism](images/attention.png)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Creating an Encoder\n",
    "class TransformerEncoder(layers.Layer):\n",
    "    def __init__(self, embed_dim, dense_dim, num_heads, **kwargs):\n",
    "        super().__init__(**kwargs)\n",
    "        self.embed_dim = embed_dim\n",
    "        self.dense_dim = dense_dim\n",
    "        self.num_heads = num_heads\n",
    "        self.attention = layers.MultiHeadAttention(\n",
    "            num_heads = num_heads, key_dim = embed_dim\n",
    "        )\n",
    "        self.dense_proj = keras.Sequential(\n",
    "            [\n",
    "                layers.Dense(dense_dim, activation = \"relu\"),\n",
    "                layers.Dense(embed_dim),\n",
    "            ]\n",
    "        )\n",
    "        self.layernorm_1 = layers.LayerNormalization()\n",
    "        self.layernorm_2 = layers.LayerNormalization()\n",
    "        self.supports_masking = True\n",
    "        \n",
    "    def call(self, inputs, mask=None):\n",
    "        if mask is not None:\n",
    "            padding_mask = tf.cast(mask[:, None, :], dtype = tf.int32)\n",
    "        else:\n",
    "            padding_mask = None\n",
    "            \n",
    "        attention_output = self.attention(\n",
    "            query = inputs,\n",
    "            value = inputs,\n",
    "            key = inputs,\n",
    "            attention_mask = padding_mask,\n",
    "        )\n",
    "        proj_input = self.layernorm_1(inputs + attention_output)\n",
    "        proj_output = self.dense_proj(proj_input)\n",
    "        return self.layernorm_2(proj_input + proj_output)\n",
    "    \n",
    "    def get_config(self):\n",
    "        config = super().get_config()\n",
    "        config.update({\n",
    "            \"embed_dim\": self.embed_dim,\n",
    "            \"dense_dim\": self.dense_dim,\n",
    "            \"num_heads\": self.num_heads,\n",
    "        })\n",
    "        return config\n",
    "    \n",
    "# Creating a Positional Embedding\n",
    "class PositionalEmbedding(layers.Layer):\n",
    "    def __init__(self, sequence_length, vocab_size, embed_dim, **kwargs):\n",
    "        super().__init__(**kwargs)\n",
    "        self.token_embeddings = layers.Embedding(\n",
    "            input_dim = vocab_size, output_dim = embed_dim\n",
    "        )\n",
    "        self.position_embeddings = layers.Embedding(\n",
    "            input_dim = sequence_length, output_dim = embed_dim\n",
    "        )\n",
    "        self.sequence_length = sequence_length\n",
    "        self.vocab_size = vocab_size\n",
    "        self.embed_dim = embed_dim\n",
    "        \n",
    "    def call(self, inputs):\n",
    "        length = tf.shape(inputs)[-1]\n",
    "        positions = tf.range(start = 0, limit = length, delta = 1)\n",
    "        embedded_tokens = self.token_embeddings(inputs)\n",
    "        embedded_positions = self.position_embeddings(positions)\n",
    "        return embedded_tokens + embedded_positions\n",
    "    \n",
    "    def compute_mask(self, inputs, mask=None):\n",
    "        if mask is not None:\n",
    "            return tf.not_equal(inputs, 0)\n",
    "        else:\n",
    "            return None\n",
    "        \n",
    "    def get_config(self):\n",
    "        config = super().get_config()\n",
    "        config.update({\n",
    "            \"vocab_size\": self.vocab_size,\n",
    "            \"sequence_length\": self.sequence_length,\n",
    "            \"embed_dim\": self.embed_dim,\n",
    "        })\n",
    "        return config\n",
    "    \n",
    "# Creating a Decoder\n",
    "class TransformerDecoder(layers.Layer):\n",
    "    def __init__(self, embed_dim, latent_dim, num_heads, **kwargs):\n",
    "        super().__init__(**kwargs)\n",
    "        self.embed_dim = embed_dim\n",
    "        self.latent_dim = latent_dim\n",
    "        self.num_heads = num_heads\n",
    "        self.attention_1 = layers.MultiHeadAttention(\n",
    "            num_heads = num_heads, key_dim = embed_dim\n",
    "        )\n",
    "        self.attention_2 = layers.MultiHeadAttention(\n",
    "            num_heads = num_heads, key_dim = embed_dim\n",
    "        )\n",
    "        self.dense_proj = keras.Sequential(\n",
    "            [\n",
    "                layers.Dense(latent_dim, activation = \"relu\"),\n",
    "                layers.Dense(embed_dim),\n",
    "            ]\n",
    "        )\n",
    "        self.layernorm_1 = layers.LayerNormalization()\n",
    "        self.layernorm_2 = layers.LayerNormalization()\n",
    "        self.layernorm_3 = layers.LayerNormalization()\n",
    "        self.supports_masking = True\n",
    "        \n",
    "    def call(self, inputs, encoder_outputs, mask=None):\n",
    "        casual_mask = self.get_causal_attention_mask(inputs)\n",
    "        if mask is not None:\n",
    "            padding_mask = tf.cast(mask[:, None, :], dtype = tf.int32)\n",
    "            padding_mask = tf.minimum(padding_mask, casual_mask)\n",
    "        else:\n",
    "            padding_mask = None\n",
    "            \n",
    "        attention_output_1 = self.attention_1(\n",
    "            query = inputs,\n",
    "            value = inputs,\n",
    "            key = inputs,\n",
    "            attention_mask = casual_mask,\n",
    "        )\n",
    "        out_1 = self.layernorm_1(inputs + attention_output_1)\n",
    "        \n",
    "        attention_output_2 = self.attention_2(\n",
    "            query = out_1,\n",
    "            value = encoder_outputs,\n",
    "            key = encoder_outputs,\n",
    "            attention_mask = padding_mask,\n",
    "        )\n",
    "        \n",
    "        out_2 = self.layernorm_2(out_1 + attention_output_2)\n",
    "        proj_output = self.dense_proj(out_2)\n",
    "        \n",
    "        return self.layernorm_3(out_2 + proj_output)\n",
    "    \n",
    "    def get_causal_attention_mask(self, inputs):\n",
    "        input_shape = tf.shape(inputs)\n",
    "        batch_size, sequence_length = input_shape[0], input_shape[1]\n",
    "        i = tf.range(sequence_length)[:, None]\n",
    "        j = tf.range(sequence_length)\n",
    "        mask = tf.cast(i >= j, tf.int32)\n",
    "        mask = tf.reshape(mask,(1, input_shape[1], input_shape[1]))\n",
    "        mult = tf.concat(\n",
    "            [\n",
    "                tf.expand_dims(batch_size, -1),\n",
    "                tf.convert_to_tensor([1, 1]),\n",
    "            ],\n",
    "            axis = 0,\n",
    "        )\n",
    "        return tf.tile(mask, mult)\n",
    "    \n",
    "    def get_config(self):\n",
    "        config = super().get_config()\n",
    "        config.update({\n",
    "            \"embed_dim\": self.embed_dim,\n",
    "            \"latent_dim\": self.latent_dim,\n",
    "            \"num_heads\": self.num_heads,\n",
    "        })\n",
    "        return config\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# define emmbedding dimensions, latent dimensions, and number of heads\n",
    "embed_dim = 256\n",
    "latent_dim = 2048\n",
    "num_heads = 8\n",
    "\n",
    "#Encoder\n",
    "encoder_inputs = keras.Input(shape = (None,), dtype = \"int64\", name = \"encoder_inputs\")\n",
    "\n",
    "x = PositionalEmbedding(sequence_length, vocab_size, embed_dim)(encoder_inputs)\n",
    "\n",
    "encoder_outputs = TransformerEncoder(embed_dim, latent_dim, num_heads)(x)\n",
    "\n",
    "encoder = keras.Model(encoder_inputs, encoder_outputs, name = \"encoder\")\n",
    "\n",
    "#Decoder\n",
    "decoder_inputs = keras.Input(shape = (None,), dtype = \"int64\", name = \"decoder_inputs\")\n",
    "encoder_seq_inputs = keras.Input(shape = (None, embed_dim), name = \"encoder_seq_inputs\")\n",
    "\n",
    "x = PositionalEmbedding(sequence_length, vocab_size, embed_dim)(decoder_inputs)\n",
    "\n",
    "x = TransformerDecoder(embed_dim, latent_dim, num_heads)(x, encoder_seq_inputs)\n",
    "\n",
    "x = layers.Dropout(0.5)(x)\n",
    "\n",
    "decoder_outputs = layers.Dense(vocab_size, activation = \"softmax\")(x)\n",
    "\n",
    "decoder = keras.Model([decoder_inputs, encoder_seq_inputs], decoder_outputs, name = \"decoder\")\n",
    "\n",
    "# Define the final model\n",
    "decoder_outputs = decoder([decoder_inputs, encoder_outputs])\n",
    "\n",
    "transformer = keras.Model(\n",
    "    [encoder_inputs, decoder_inputs], decoder_outputs, name = \"transformer\"\n",
    ")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"transformer\"\n",
      "__________________________________________________________________________________________________\n",
      " Layer (type)                   Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      " encoder_inputs (InputLayer)    [(None, None)]       0           []                               \n",
      "                                                                                                  \n",
      " positional_embedding (Position  (None, None, 256)   3845120     ['encoder_inputs[0][0]']         \n",
      " alEmbedding)                                                                                     \n",
      "                                                                                                  \n",
      " decoder_inputs (InputLayer)    [(None, None)]       0           []                               \n",
      "                                                                                                  \n",
      " transformer_encoder (Transform  (None, None, 256)   3155456     ['positional_embedding[0][0]']   \n",
      " erEncoder)                                                                                       \n",
      "                                                                                                  \n",
      " decoder (Functional)           (None, None, 15000)  12959640    ['decoder_inputs[0][0]',         \n",
      "                                                                  'transformer_encoder[0][0]']    \n",
      "                                                                                                  \n",
      "==================================================================================================\n",
      "Total params: 19,960,216\n",
      "Trainable params: 19,960,216\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "Epoch 1/20\n",
      "1302/1302 [==============================] - 186s 138ms/step - loss: 1.9811 - accuracy: 0.7206 - val_loss: 1.6848 - val_accuracy: 0.7416\n",
      "Epoch 2/20\n",
      "1302/1302 [==============================] - 179s 137ms/step - loss: 1.7133 - accuracy: 0.7508 - val_loss: 1.5721 - val_accuracy: 0.7638\n",
      "Epoch 3/20\n",
      "1302/1302 [==============================] - 178s 136ms/step - loss: 1.5370 - accuracy: 0.7759 - val_loss: 1.3673 - val_accuracy: 0.7908\n",
      "Epoch 4/20\n",
      "1302/1302 [==============================] - 178s 137ms/step - loss: 1.3938 - accuracy: 0.7957 - val_loss: 1.2725 - val_accuracy: 0.8077\n",
      "Epoch 5/20\n",
      "1302/1302 [==============================] - 177s 136ms/step - loss: 1.2863 - accuracy: 0.8128 - val_loss: 1.2105 - val_accuracy: 0.8216\n",
      "Epoch 6/20\n",
      "1302/1302 [==============================] - 178s 137ms/step - loss: 1.2217 - accuracy: 0.8268 - val_loss: 1.1603 - val_accuracy: 0.8325\n",
      "Epoch 7/20\n",
      "1302/1302 [==============================] - 178s 137ms/step - loss: 1.1745 - accuracy: 0.8369 - val_loss: 1.1168 - val_accuracy: 0.8407\n",
      "Epoch 8/20\n",
      "1302/1302 [==============================] - 178s 136ms/step - loss: 1.1379 - accuracy: 0.8448 - val_loss: 1.1005 - val_accuracy: 0.8429\n",
      "Epoch 9/20\n",
      "1302/1302 [==============================] - 177s 136ms/step - loss: 1.1092 - accuracy: 0.8506 - val_loss: 1.0816 - val_accuracy: 0.8485\n",
      "Epoch 10/20\n",
      "1302/1302 [==============================] - 178s 136ms/step - loss: 1.0849 - accuracy: 0.8552 - val_loss: 1.0628 - val_accuracy: 0.8515\n",
      "Epoch 11/20\n",
      "1302/1302 [==============================] - 177s 136ms/step - loss: 1.0652 - accuracy: 0.8590 - val_loss: 1.0541 - val_accuracy: 0.8510\n",
      "Epoch 12/20\n",
      "1302/1302 [==============================] - 178s 136ms/step - loss: 1.0469 - accuracy: 0.8622 - val_loss: 1.0370 - val_accuracy: 0.8556\n",
      "Epoch 13/20\n",
      "1302/1302 [==============================] - 178s 136ms/step - loss: 1.0317 - accuracy: 0.8649 - val_loss: 1.0275 - val_accuracy: 0.8577\n",
      "Epoch 14/20\n",
      "1302/1302 [==============================] - 178s 136ms/step - loss: 1.0163 - accuracy: 0.8674 - val_loss: 1.0242 - val_accuracy: 0.8584\n",
      "Epoch 15/20\n",
      "1302/1302 [==============================] - 178s 136ms/step - loss: 1.0016 - accuracy: 0.8696 - val_loss: 1.0176 - val_accuracy: 0.8602\n",
      "Epoch 16/20\n",
      "1302/1302 [==============================] - 178s 136ms/step - loss: 0.9884 - accuracy: 0.8717 - val_loss: 1.0102 - val_accuracy: 0.8606\n",
      "Epoch 17/20\n",
      "1302/1302 [==============================] - 178s 136ms/step - loss: 0.9767 - accuracy: 0.8736 - val_loss: 1.0234 - val_accuracy: 0.8591\n",
      "Epoch 18/20\n",
      "1302/1302 [==============================] - 178s 136ms/step - loss: 0.9665 - accuracy: 0.8752 - val_loss: 0.9950 - val_accuracy: 0.8634\n",
      "Epoch 19/20\n",
      "1302/1302 [==============================] - 178s 136ms/step - loss: 0.9549 - accuracy: 0.8771 - val_loss: 1.0000 - val_accuracy: 0.8621\n",
      "Epoch 20/20\n",
      "1302/1302 [==============================] - 178s 136ms/step - loss: 0.9456 - accuracy: 0.8783 - val_loss: 1.0106 - val_accuracy: 0.8617\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x214f97284c0>"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "epochs = 20\n",
    "\n",
    "transformer.summary()\n",
    "\n",
    "transformer.compile(\n",
    "    \"rmsprop\", loss = \"sparse_categorical_crossentropy\", metrics = [\"accuracy\"]\n",
    ")\n",
    "\n",
    "transformer.fit(train_ds, epochs = epochs, validation_data = val_ds)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING:absl:Found untraced functions such as embedding_layer_call_fn, embedding_layer_call_and_return_conditional_losses, embedding_1_layer_call_fn, embedding_1_layer_call_and_return_conditional_losses, multi_head_attention_layer_call_fn while saving (showing 5 of 60). These functions will not be directly callable after loading.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Assets written to: transformer_model\\assets\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Assets written to: transformer_model\\assets\n"
     ]
    }
   ],
   "source": [
    "transformer.save(\"transformer_model\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "input: her days are numbered\n",
      "translated: [start] sus días están en el [UNK] [end]\n",
      "\n",
      "input: i heard the door close\n",
      "translated: [start] he oído la puerta [end]\n",
      "\n",
      "input: you go first\n",
      "translated: [start] vas al primero [end]\n",
      "\n",
      "input: you used to look up to your father\n",
      "translated: [start] te [UNK] a tu padre [end]\n",
      "\n",
      "input: i am acquainted with the chairman of the committee\n",
      "translated: [start] estoy en el [UNK] de la [UNK] [end]\n",
      "\n"
     ]
    }
   ],
   "source": [
    "spa_vocab = spa_vectorization.get_vocabulary()\n",
    "spa_index_lookup = dict(zip(range(len(spa_vocab)), spa_vocab))\n",
    "max_decoded_sentence_length = sequence_length\n",
    "\n",
    "def decode_sentence(input_sentence):\n",
    "    tokenized_input_sentence = eng_vectorization([input_sentence])\n",
    "    decoded_sentence = \"[start]\"\n",
    "    for i in range(max_decoded_sentence_length):\n",
    "        tokenized_target_sentence = spa_vectorization([decoded_sentence])[:, :-1]\n",
    "        predictions = transformer([tokenized_input_sentence, tokenized_target_sentence])\n",
    "        sampled_token_index = tf.argmax(predictions[0, i, :]).numpy().item(0)\n",
    "        sampled_token = spa_index_lookup[sampled_token_index]\n",
    "        decoded_sentence += \" \" + sampled_token\n",
    "        if sampled_token == \"[end]\":\n",
    "            break\n",
    "    return decoded_sentence\n",
    "\n",
    "test_eng_texts = [pair[0] for pair in test_pairs]\n",
    "for _ in range(5):\n",
    "    input_sentence = random.choice(test_eng_texts)\n",
    "    input_sentence = input_sentence.lower()\n",
    "    input_sentence = input_sentence.translate(str.maketrans('', '', strip_chars))\n",
    "    translated = decode_sentence(input_sentence)\n",
    "    print(f\"input: {input_sentence}\")\n",
    "    print(f\"translated: {translated}\")\n",
    "    print()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "base",
   "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": "3.9.19"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}