neural-machine-translation/38.dnc-seq2seq-bahdanau-greedy.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# !pip3 install dm-sonnet\n",
    "# !pip3 install tensorflow-probability==0.7.0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "os.environ['CUDA_VISIBLE_DEVICES'] = '3'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from dnc import DNC"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "\n",
    "with open('train-test.json') as fopen:\n",
    "    dataset = json.load(fopen)\n",
    "    \n",
    "with open('dictionary.json') as fopen:\n",
    "    dictionary = json.load(fopen)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_X = dataset['train_X']\n",
    "train_Y = dataset['train_Y']\n",
    "test_X = dataset['test_X']\n",
    "test_Y = dataset['test_Y']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "dictionary.keys()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "dictionary_from = dictionary['from']['dictionary']\n",
    "rev_dictionary_from = dictionary['from']['rev_dictionary']\n",
    "\n",
    "dictionary_to = dictionary['to']['dictionary']\n",
    "rev_dictionary_to = dictionary['to']['rev_dictionary']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "GO = dictionary_from['GO']\n",
    "PAD = dictionary_from['PAD']\n",
    "EOS = dictionary_from['EOS']\n",
    "UNK = dictionary_from['UNK']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "for i in range(len(train_X)):\n",
    "    train_X[i] += ' EOS'\n",
    "    \n",
    "train_X[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "for i in range(len(test_X)):\n",
    "    test_X[i] += ' EOS'\n",
    "    \n",
    "test_X[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_reads = 5\n",
    "num_writes = 1\n",
    "memory_size = 128\n",
    "word_size = 128\n",
    "clip_value = 20"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def pad_second_dim(x, desired_size):\n",
    "    padding = tf.tile([[[0.0]]], tf.stack([tf.shape(x)[0], desired_size - tf.shape(x)[1], tf.shape(x)[2]], 0))\n",
    "    return tf.concat([x, padding], 1)\n",
    "\n",
    "class Translator:\n",
    "    def __init__(self, size_layer, num_layers, embedded_size, \n",
    "                 from_dict_size, to_dict_size, learning_rate, batch_size,\n",
    "                attn_input_feeding=True):\n",
    "        \n",
    "        def attn_decoder_input_fn(inputs, attention):\n",
    "            if attn_input_feeding:\n",
    "                return inputs\n",
    "        \n",
    "        def attention(encoder_out, cell, seq_len, encoder_last_state, reuse=False):\n",
    "            attention_mechanism = tf.contrib.seq2seq.BahdanauAttention(num_units = size_layer, \n",
    "                                                                    memory = encoder_out,\n",
    "                                                                    memory_sequence_length = seq_len)\n",
    "            return tf.contrib.seq2seq.AttentionWrapper(\n",
    "                cell = cell, \n",
    "                attention_mechanism = attention_mechanism,\n",
    "                attention_layer_size = size_layer,\n",
    "                cell_input_fn=attn_decoder_input_fn,\n",
    "                initial_cell_state=encoder_last_state,\n",
    "                alignment_history=False)\n",
    "        \n",
    "        self.X = tf.placeholder(tf.int32, [None, None])\n",
    "        self.Y = tf.placeholder(tf.int32, [None, None])\n",
    "        self.X_seq_len = tf.count_nonzero(self.X, 1, dtype=tf.int32)\n",
    "        self.Y_seq_len = tf.count_nonzero(self.Y, 1, dtype=tf.int32)\n",
    "        access_config = {\n",
    "            \"memory_size\": memory_size,\n",
    "            \"word_size\": word_size,\n",
    "            \"num_reads\": num_reads,\n",
    "            \"num_writes\": num_writes,\n",
    "        }\n",
    "        controller_config = {\n",
    "            \"hidden_size\": size_layer,\n",
    "        }\n",
    "        self.dnc_cell = DNC(access_config=access_config, controller_config=controller_config,\n",
    "                            output_size=size_layer, clip_value=clip_value)\n",
    "        self.dnc_initial = self.dnc_cell.initial_state\n",
    "        \n",
    "        encoder_embeddings = tf.Variable(tf.random_uniform([from_dict_size, embedded_size], -1, 1))\n",
    "        encoder_embedded = tf.nn.embedding_lookup(encoder_embeddings, self.X)\n",
    "        \n",
    "        initial_state = self.dnc_initial(batch_size)\n",
    "        self.encoder_out, self.encoder_state = tf.nn.dynamic_rnn(\n",
    "            cell=self.dnc_cell, inputs=encoder_embedded,\n",
    "            sequence_length=self.X_seq_len, dtype=tf.float32,\n",
    "            initial_state=initial_state)\n",
    "        main = tf.strided_slice(self.Y, [0, 0], [batch_size, -1], [1, 1])\n",
    "        decoder_input = tf.concat([tf.fill([batch_size, 1], GO), main], 1)\n",
    "        # decoder\n",
    "        decoder_embeddings = tf.Variable(tf.random_uniform([to_dict_size, embedded_size], -1, 1))\n",
    "        decoder_cell = attention(self.encoder_out, self.dnc_cell, self.X_seq_len,self.encoder_state)\n",
    "        dense_layer = tf.layers.Dense(to_dict_size)\n",
    "        training_helper = tf.contrib.seq2seq.TrainingHelper(\n",
    "            inputs = tf.nn.embedding_lookup(decoder_embeddings, decoder_input),\n",
    "            sequence_length = self.Y_seq_len,\n",
    "            time_major = False)\n",
    "        training_decoder = tf.contrib.seq2seq.BasicDecoder(\n",
    "            cell = decoder_cell,\n",
    "            helper = training_helper,\n",
    "            initial_state = decoder_cell.zero_state(batch_size=batch_size, dtype=tf.float32),\n",
    "            output_layer = dense_layer)\n",
    "        training_decoder_output, _, _ = tf.contrib.seq2seq.dynamic_decode(\n",
    "            decoder = training_decoder,\n",
    "            impute_finished = True,\n",
    "            output_time_major=False,\n",
    "            maximum_iterations = tf.reduce_max(self.Y_seq_len))\n",
    "        \n",
    "        predicting_helper = tf.contrib.seq2seq.GreedyEmbeddingHelper(\n",
    "                embedding = decoder_embeddings,\n",
    "                start_tokens = tf.tile(tf.constant([GO], dtype=tf.int32), [batch_size]),\n",
    "                end_token = EOS)\n",
    "        predicting_decoder = tf.contrib.seq2seq.BasicDecoder(\n",
    "                cell = decoder_cell,\n",
    "                helper = predicting_helper,\n",
    "                initial_state = decoder_cell.zero_state(batch_size=batch_size, dtype=tf.float32),\n",
    "                output_layer = dense_layer)\n",
    "        predicting_decoder_output, _, _ = tf.contrib.seq2seq.dynamic_decode(\n",
    "                decoder = predicting_decoder,\n",
    "                impute_finished = True,\n",
    "                maximum_iterations = 2 * tf.reduce_max(self.X_seq_len))\n",
    "        self.training_logits = training_decoder_output.rnn_output\n",
    "        self.predicting_ids = predicting_decoder_output.sample_id\n",
    "        masks = tf.sequence_mask(self.Y_seq_len, tf.reduce_max(self.Y_seq_len), dtype=tf.float32)\n",
    "        self.cost = tf.contrib.seq2seq.sequence_loss(logits = self.training_logits,\n",
    "                                                     targets = self.Y,\n",
    "                                                     weights = masks)\n",
    "        self.optimizer = tf.train.AdamOptimizer(learning_rate).minimize(self.cost)\n",
    "        y_t = tf.argmax(self.training_logits,axis=2)\n",
    "        y_t = tf.cast(y_t, tf.int32)\n",
    "        self.prediction = tf.boolean_mask(y_t, masks)\n",
    "        mask_label = tf.boolean_mask(self.Y, masks)\n",
    "        correct_pred = tf.equal(self.prediction, mask_label)\n",
    "        correct_index = tf.cast(correct_pred, tf.float32)\n",
    "        self.accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "size_layer = 256\n",
    "num_layers = 2\n",
    "embedded_size = 256\n",
    "learning_rate = 1e-3\n",
    "batch_size = 96\n",
    "epoch = 20"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "tf.reset_default_graph()\n",
    "sess = tf.InteractiveSession()\n",
    "model = Translator(size_layer, num_layers, embedded_size, len(dictionary_from), \n",
    "                len(dictionary_to), learning_rate,batch_size)\n",
    "sess.run(tf.global_variables_initializer())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def str_idx(corpus, dic):\n",
    "    X = []\n",
    "    for i in corpus:\n",
    "        ints = []\n",
    "        for k in i.split():\n",
    "            ints.append(dic.get(k,UNK))\n",
    "        X.append(ints)\n",
    "    return X\n",
    "\n",
    "def pad_sentence_batch(sentence_batch, pad_int):\n",
    "    padded_seqs = []\n",
    "    seq_lens = []\n",
    "    max_sentence_len = max([len(sentence) for sentence in sentence_batch])\n",
    "    for sentence in sentence_batch:\n",
    "        padded_seqs.append(sentence + [pad_int] * (max_sentence_len - len(sentence)))\n",
    "        seq_lens.append(len(sentence))\n",
    "    return padded_seqs, seq_lens"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_X = str_idx(train_X, dictionary_from)\n",
    "test_X = str_idx(test_X, dictionary_from)\n",
    "train_Y = str_idx(train_Y, dictionary_to)\n",
    "test_Y = str_idx(test_Y, dictionary_to)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tqdm\n",
    "\n",
    "for e in range(epoch):\n",
    "    pbar = tqdm.tqdm(\n",
    "        range(0, len(train_X), batch_size), desc = 'minibatch loop')\n",
    "    train_loss, train_acc, test_loss, test_acc = [], [], [], []\n",
    "    for i in pbar:\n",
    "        index = min(i + batch_size, len(train_X))\n",
    "        maxlen = max([len(s) for s in train_X[i : index] + train_Y[i : index]])\n",
    "        batch_x, seq_x = pad_sentence_batch(train_X[i : index], PAD)\n",
    "        batch_y, seq_y = pad_sentence_batch(train_Y[i : index], PAD)\n",
    "        feed = {model.X: batch_x,\n",
    "                model.Y: batch_y}\n",
    "        accuracy, loss, _ = sess.run([model.accuracy,model.cost,model.optimizer],\n",
    "                                    feed_dict = feed)\n",
    "        train_loss.append(loss)\n",
    "        train_acc.append(accuracy)\n",
    "        pbar.set_postfix(cost = loss, accuracy = accuracy)\n",
    "    \n",
    "    \n",
    "    pbar = tqdm.tqdm(\n",
    "        range(0, len(test_X), batch_size), desc = 'minibatch loop')\n",
    "    for i in pbar:\n",
    "        index = min(i + batch_size, len(test_X))\n",
    "        batch_x, seq_x = pad_sentence_batch(test_X[i : index], PAD)\n",
    "        batch_y, seq_y = pad_sentence_batch(test_Y[i : index], PAD)\n",
    "        feed = {model.X: batch_x,\n",
    "                model.Y: batch_y,}\n",
    "        accuracy, loss = sess.run([model.accuracy,model.cost],\n",
    "                                    feed_dict = feed)\n",
    "\n",
    "        test_loss.append(loss)\n",
    "        test_acc.append(accuracy)\n",
    "        pbar.set_postfix(cost = loss, accuracy = accuracy)\n",
    "    \n",
    "    print('epoch %d, training avg loss %f, training avg acc %f'%(e+1,\n",
    "                                                                 np.mean(train_loss),np.mean(train_acc)))\n",
    "    print('epoch %d, testing avg loss %f, testing avg acc %f'%(e+1,\n",
    "                                                              np.mean(test_loss),np.mean(test_acc)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "rev_dictionary_to = {int(k): v for k, v in rev_dictionary_to.items()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_size = 20\n",
    "\n",
    "batch_x, seq_x = pad_sentence_batch(test_X[: test_size], PAD)\n",
    "batch_y, seq_y = pad_sentence_batch(test_Y[: test_size], PAD)\n",
    "feed = {model.X: batch_x}\n",
    "logits = sess.run(model.predicting_ids, feed_dict = feed)\n",
    "logits.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "rejected = ['PAD', 'EOS', 'UNK', 'GO']\n",
    "\n",
    "for i in range(test_size):\n",
    "    predict = [rev_dictionary_to[i] for i in logits[i] if rev_dictionary_to[i] not in rejected]\n",
    "    actual = [rev_dictionary_to[i] for i in batch_y[i] if rev_dictionary_to[i] not in rejected]\n",
    "    print(i, 'predict:', ' '.join(predict))\n",
    "    print(i, 'actual:', ' '.join(actual))\n",
    "    print()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}