neural-machine-translation/16.basic-seq2seq-bahdanau.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "os.environ['CUDA_VISIBLE_DEVICES'] = '0'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/husein/.local/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:516: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\n",
      "/home/husein/.local/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:517: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\n",
      "/home/husein/.local/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:518: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\n",
      "/home/husein/.local/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:519: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\n",
      "/home/husein/.local/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:520: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\n",
      "/home/husein/.local/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:525: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\n",
      "/usr/lib/python3/dist-packages/requests/__init__.py:80: RequestsDependencyWarning: urllib3 (1.25.3) or chardet (3.0.4) doesn't match a supported version!\n",
      "  RequestsDependencyWarning)\n",
      "/home/husein/.local/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:541: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\n",
      "/home/husein/.local/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:542: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\n",
      "/home/husein/.local/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:543: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\n",
      "/home/husein/.local/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:544: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\n",
      "/home/husein/.local/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:545: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\n",
      "/home/husein/.local/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:550: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "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": 4,
   "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": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "dict_keys(['from', 'to'])"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dictionary.keys()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "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": 7,
   "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": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'Rachel Pike : The science behind a climate headline EOS'"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "for i in range(len(train_X)):\n",
    "    train_X[i] += ' EOS'\n",
    "    \n",
    "train_X[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'How can I speak in <NUM> minutes about the bonds of women over three generations , about how the astonishing strength of those bonds took hold in the life of a four - year - old girl huddled with her young sister , her mother and her grandmother for five days and nights in a small boat in the China Sea more than <NUM> years ago , bonds that took hold in the life of that small girl and never let go - - that small girl now living in San Francisco and speaking to you today ? EOS'"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "for i in range(len(test_X)):\n",
    "    test_X[i] += ' EOS'\n",
    "    \n",
    "test_X[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "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",
    "        \n",
    "        def cells(reuse=False):\n",
    "            return tf.nn.rnn_cell.BasicRNNCell(size_layer,reuse=reuse)\n",
    "        \n",
    "        self.X = tf.placeholder(tf.int32, [None, None])\n",
    "        self.Y = tf.placeholder(tf.int32, [None, None])\n",
    "        \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",
    "        batch_size = tf.shape(self.X)[0]\n",
    "        \n",
    "        encoder_embeddings = tf.Variable(tf.random_uniform([from_dict_size, embedded_size], -1, 1))\n",
    "        decoder_embeddings = tf.Variable(tf.random_uniform([to_dict_size, embedded_size], -1, 1))\n",
    "        encoder_embedded = tf.nn.embedding_lookup(encoder_embeddings, self.X)\n",
    "        main = tf.strided_slice(self.X, [0, 0], [batch_size, -1], [1, 1])\n",
    "        decoder_input = tf.concat([tf.fill([batch_size, 1], GO), main], 1)\n",
    "        decoder_embedded = tf.nn.embedding_lookup(encoder_embeddings, decoder_input)\n",
    "        rnn_cells = tf.nn.rnn_cell.MultiRNNCell([cells() for _ in range(num_layers)])\n",
    "        last_output, last_state = tf.nn.dynamic_rnn(rnn_cells, encoder_embedded,\n",
    "                                          sequence_length=self.X_seq_len,\n",
    "                                          dtype = tf.float32)\n",
    "        with tf.variable_scope(\"decoder\"):\n",
    "            \n",
    "            attention_mechanism = tf.contrib.seq2seq.BahdanauAttention(num_units = size_layer, \n",
    "                                                                    memory = last_output)\n",
    "            rnn_cells = tf.contrib.seq2seq.AttentionWrapper(\n",
    "                cell = tf.nn.rnn_cell.MultiRNNCell([cells() for _ in range(num_layers)]), \n",
    "                attention_mechanism = attention_mechanism,\n",
    "                attention_layer_size = size_layer)\n",
    "            \n",
    "            initial_state = rnn_cells.zero_state(batch_size, tf.float32).clone(cell_state=last_state)\n",
    "            outputs, _ = tf.nn.dynamic_rnn(rnn_cells, decoder_embedded, \n",
    "                                           sequence_length=self.X_seq_len,\n",
    "                                           initial_state = initial_state,\n",
    "                                           dtype = tf.float32)\n",
    "            self.logits = tf.layers.dense(outputs,to_dict_size)\n",
    "        \n",
    "        self.training_logits = self.logits[:, :tf.reduce_max(self.Y_seq_len)]\n",
    "        self.training_logits = pad_second_dim(self.training_logits, tf.reduce_max(self.Y_seq_len))\n",
    "            \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 = 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": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "size_layer = 512\n",
    "num_layers = 2\n",
    "embedded_size = 256\n",
    "learning_rate = 1e-3\n",
    "batch_size = 128\n",
    "epoch = 20"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING: Logging before flag parsing goes to stderr.\n",
      "W0903 10:27:32.271757 140383166703424 deprecation.py:506] From /home/husein/.local/lib/python3.6/site-packages/tensorflow/python/util/deprecation.py:507: calling count_nonzero (from tensorflow.python.ops.math_ops) with axis is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "reduction_indices is deprecated, use axis instead\n",
      "W0903 10:27:32.321250 140383166703424 deprecation.py:323] From <ipython-input-10-b3d71d724683>:10: BasicRNNCell.__init__ (from tensorflow.python.ops.rnn_cell_impl) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "This class is equivalent as tf.keras.layers.SimpleRNNCell, and will be replaced by that in Tensorflow 2.0.\n",
      "W0903 10:27:32.323070 140383166703424 deprecation.py:323] From <ipython-input-10-b3d71d724683>:25: MultiRNNCell.__init__ (from tensorflow.python.ops.rnn_cell_impl) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "This class is equivalent as tf.keras.layers.StackedRNNCells, and will be replaced by that in Tensorflow 2.0.\n",
      "W0903 10:27:32.324214 140383166703424 deprecation.py:323] From <ipython-input-10-b3d71d724683>:28: dynamic_rnn (from tensorflow.python.ops.rnn) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use `keras.layers.RNN(cell)`, which is equivalent to this API\n",
      "W0903 10:27:32.648029 140383166703424 deprecation.py:506] From /home/husein/.local/lib/python3.6/site-packages/tensorflow/python/ops/init_ops.py:1251: calling VarianceScaling.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Call initializer instance with the dtype argument instead of passing it to the constructor\n",
      "W0903 10:27:32.658675 140383166703424 deprecation.py:506] From /home/husein/.local/lib/python3.6/site-packages/tensorflow/python/ops/rnn_cell_impl.py:459: calling Zeros.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Call initializer instance with the dtype argument instead of passing it to the constructor\n",
      "W0903 10:27:32.781254 140383166703424 deprecation.py:323] From /home/husein/.local/lib/python3.6/site-packages/tensorflow/python/ops/rnn.py:244: add_dispatch_support.<locals>.wrapper (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use tf.where in 2.0, which has the same broadcast rule as np.where\n",
      "W0903 10:27:33.488704 140383166703424 lazy_loader.py:50] \n",
      "The TensorFlow contrib module will not be included in TensorFlow 2.0.\n",
      "For more information, please see:\n",
      "  * https://github.com/tensorflow/community/blob/master/rfcs/20180907-contrib-sunset.md\n",
      "  * https://github.com/tensorflow/addons\n",
      "  * https://github.com/tensorflow/io (for I/O related ops)\n",
      "If you depend on functionality not listed there, please file an issue.\n",
      "\n",
      "W0903 10:27:34.589832 140383166703424 deprecation.py:323] From <ipython-input-10-b3d71d724683>:43: dense (from tensorflow.python.layers.core) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use keras.layers.dense instead.\n"
     ]
    }
   ],
   "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": 13,
   "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": 14,
   "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": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:38<00:00,  2.61it/s, accuracy=0.0983, cost=6.5] \n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.69it/s, accuracy=0.136, cost=6.3]  \n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1, training avg loss 6.621004, training avg acc 0.095715\n",
      "epoch 1, testing avg loss 6.351532, testing avg acc 0.111844\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:30<00:00,  2.67it/s, accuracy=0.097, cost=6.23] \n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.97it/s, accuracy=0.141, cost=6.14] \n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 2, training avg loss 6.112189, training avg acc 0.123624\n",
      "epoch 2, testing avg loss 6.178668, testing avg acc 0.117637\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:31<00:00,  2.66it/s, accuracy=0.106, cost=5.9]  \n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.81it/s, accuracy=0.147, cost=6.01]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 3, training avg loss 5.853072, training avg acc 0.129371\n",
      "epoch 3, testing avg loss 6.037225, testing avg acc 0.119821\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:30<00:00,  2.67it/s, accuracy=0.12, cost=5.49]  \n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.82it/s, accuracy=0.13, cost=5.76] \n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 4, training avg loss 5.583531, training avg acc 0.137750\n",
      "epoch 4, testing avg loss 5.777971, testing avg acc 0.132109\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:31<00:00,  2.66it/s, accuracy=0.136, cost=5.16] \n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.90it/s, accuracy=0.147, cost=5.54]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 5, training avg loss 5.337791, training avg acc 0.145715\n",
      "epoch 5, testing avg loss 5.649175, testing avg acc 0.133093\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:31<00:00,  2.66it/s, accuracy=0.151, cost=5.03] \n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.83it/s, accuracy=0.141, cost=5.41]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 6, training avg loss 5.168367, training avg acc 0.150114\n",
      "epoch 6, testing avg loss 5.577363, testing avg acc 0.135001\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:32<00:00,  2.66it/s, accuracy=0.16, cost=4.86] \n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.82it/s, accuracy=0.147, cost=5.47]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 7, training avg loss 5.043260, training avg acc 0.153736\n",
      "epoch 7, testing avg loss 5.546894, testing avg acc 0.132628\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:31<00:00,  2.66it/s, accuracy=0.169, cost=4.72]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.88it/s, accuracy=0.141, cost=5.27]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 8, training avg loss 4.919789, training avg acc 0.157820\n",
      "epoch 8, testing avg loss 5.417903, testing avg acc 0.135997\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:31<00:00,  2.66it/s, accuracy=0.171, cost=4.72]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.86it/s, accuracy=0.124, cost=5.33]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 9, training avg loss 4.854082, training avg acc 0.159356\n",
      "epoch 9, testing avg loss 5.430256, testing avg acc 0.134560\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:31<00:00,  2.66it/s, accuracy=0.185, cost=4.53]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.83it/s, accuracy=0.153, cost=5.27]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 10, training avg loss 4.781552, training avg acc 0.162176\n",
      "epoch 10, testing avg loss 5.389349, testing avg acc 0.136719\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:48<00:00,  2.55it/s, accuracy=0.186, cost=4.5] \n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.89it/s, accuracy=0.13, cost=5.22] \n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 11, training avg loss 4.701983, training avg acc 0.166030\n",
      "epoch 11, testing avg loss 5.405530, testing avg acc 0.132007\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:32<00:00,  2.66it/s, accuracy=0.181, cost=4.55]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.80it/s, accuracy=0.153, cost=5.29]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 12, training avg loss 4.659106, training avg acc 0.167810\n",
      "epoch 12, testing avg loss 5.394949, testing avg acc 0.134590\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:31<00:00,  2.66it/s, accuracy=0.193, cost=4.42]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.73it/s, accuracy=0.147, cost=5.35]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 13, training avg loss 4.626896, training avg acc 0.168729\n",
      "epoch 13, testing avg loss 5.343545, testing avg acc 0.137271\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:31<00:00,  2.66it/s, accuracy=0.163, cost=4.5] \n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.89it/s, accuracy=0.147, cost=5.28]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 14, training avg loss 4.613827, training avg acc 0.168623\n",
      "epoch 14, testing avg loss 5.327889, testing avg acc 0.137628\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:32<00:00,  2.66it/s, accuracy=0.149, cost=4.83]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.89it/s, accuracy=0.136, cost=5.68]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 15, training avg loss 4.698052, training avg acc 0.162271\n",
      "epoch 15, testing avg loss 5.488557, testing avg acc 0.131581\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:32<00:00,  2.66it/s, accuracy=0.147, cost=5.01] \n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.99it/s, accuracy=0.102, cost=5.7] \n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 16, training avg loss 4.705058, training avg acc 0.161847\n",
      "epoch 16, testing avg loss 5.701091, testing avg acc 0.121615\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:32<00:00,  2.66it/s, accuracy=0.176, cost=4.52]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.90it/s, accuracy=0.153, cost=5.26]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 17, training avg loss 4.632894, training avg acc 0.164997\n",
      "epoch 17, testing avg loss 5.306539, testing avg acc 0.139349\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:31<00:00,  2.66it/s, accuracy=0.175, cost=4.39]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.92it/s, accuracy=0.153, cost=5.44]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 18, training avg loss 4.591221, training avg acc 0.167762\n",
      "epoch 18, testing avg loss 5.355982, testing avg acc 0.135911\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:31<00:00,  2.66it/s, accuracy=0.181, cost=4.43]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.90it/s, accuracy=0.13, cost=5.46] \n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 19, training avg loss 4.589289, training avg acc 0.167523\n",
      "epoch 19, testing avg loss 5.386686, testing avg acc 0.133285\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:31<00:00,  2.66it/s, accuracy=0.16, cost=4.62] \n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.84it/s, accuracy=0.147, cost=5.34]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 20, training avg loss 4.679741, training avg acc 0.161262\n",
      "epoch 20, testing avg loss 5.428499, testing avg acc 0.132169\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "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": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "rev_dictionary_to = {int(k): v for k, v in rev_dictionary_to.items()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(20, 99)"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "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,model.Y: batch_y,}\n",
    "logits = np.argmax(sess.run(model.logits, feed_dict = feed), axis = 2)\n",
    "logits.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 predict: Làm sao có có nói <NUM> <NUM> <NUM> <NUM> <NUM> nữ nữ nữ <NUM> <NUM> thế thế về về về về về về những những những những những những bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé bé\n",
      "0 actual: Làm sao tôi có thể trình bày trong <NUM> phút về sợi dây liên kết những người phụ nữ qua ba thế hệ , về việc làm thế nào những sợi dây mạnh mẽ đáng kinh ngạc ấy đã níu chặt lấy cuộc sống của một cô bé bốn tuổi co quắp với đứa em gái nhỏ của cô bé , với mẹ và bà trong suốt năm ngày đêm trên con thuyền nhỏ lênh đênh trên Biển Đông hơn <NUM> năm trước , những sợi dây liên kết đã níu lấy cuộc đời cô bé ấy và không bao giờ rời đi - - cô bé ấy giờ sống ở San Francisco và đang nói chuyện với các bạn hôm nay ?\n",
      "\n",
      "1 predict: Đây là không là chuyện chuyện chuyện . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "1 actual: Câu chuyện này chưa kết thúc .\n",
      "\n",
      "2 predict: Đó là một đặt đặt đặt đặt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "2 actual: Nó là một trò chơi ghép hình vẫn đang được xếp .\n",
      "\n",
      "3 predict: Hãy tôi cho cho các về về . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "3 actual: Hãy để tôi kể cho các bạn về vài mảnh ghép nhé .\n",
      "\n",
      "4 predict: Hình tưởng tượng đầu tiên tiên tiên đàn ông ông ông của của của . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "4 actual: Hãy tưởng tượng mảnh đầu tiên : một người đàn ông đốt cháy sự nghiệp cả đời mình .\n",
      "\n",
      "5 predict: Là là thơ thơ thơ thơ nhà nhà nhà nhà ông ông một một có có có có có có có của của của của của tự của tự tự . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "5 actual: Ông là nhà thơ , nhà viết kịch , một người mà cả cuộc đời chênh vênh trên tia hi vọng duy nhất rằng đất nước ông sẽ độc lập tự do .\n",
      "\n",
      "6 predict: Tưởng tưởng tượng hình là Sandberg , , , , , thực đời cuộc một một một . . phí . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "6 actual: Hãy tưởng tượng ông , một người cộng sản tiến vào , đối diện sự thật rằng cả cuộc đời ông đã phí hoài .\n",
      "\n",
      "7 predict: Chia là , rất lâu rất rất ông giờ ông anh anh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "7 actual: Ngôn từ , qua bao năm tháng là bạn đồng hành với ông , giờ quay ra chế giễu ông .\n",
      "\n",
      "8 predict: Ông phăng Noble rắc rối . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "8 actual: Ông rút lui vào yên lặng .\n",
      "\n",
      "9 predict: Anh bị chết qua qua . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "9 actual: Ông qua đời , bị lịch sử quật ngã .\n",
      "\n",
      "10 predict: Ông là là là . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "10 actual: Ông là ông của tôi .\n",
      "\n",
      "11 predict: Tôi bao bao biết biết ấy sống sống đời . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "11 actual: Tôi chưa bao giờ gặp ông ngoài đời .\n",
      "\n",
      "12 predict: Nhưng sống sống sống hơn hơn nhiều nhiều hơn ức ức . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "12 actual: Nhưng cuộc đời ta nhiều hơn những gì ta lưu trong kí ức nhiều .\n",
      "\n",
      "13 predict: Cô bà không tôi tôi tôi quên quên đời đời . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "13 actual: Bà tôi chưa bao giờ cho phép tôi quên cuộc đời của ông .\n",
      "\n",
      "14 predict: Mẹ nhiệm của tôi phải để để để , , học học học học học học , , , , , , , , , , , chúng chúng chúng . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "14 actual: Nhiệm vụ của tôi là không để cuộc đời ấy qua trong vô vọng , và bài học của tôi là nhận ra rằng , vâng , lịch sử đã cố quật ngã chúng tôi , nhưng chúng tôi đã chịu đựng được .\n",
      "\n",
      "15 predict: Mẫu tiếp Mẫu hàng tàu tàu tàu tàu tàu tàu hoả thuyền hòm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "15 actual: Mảnh ghép tiếp theo của tấm hình là một con thuyền trong sớm hoàng hôn lặng lẽ trườn ra biển .\n",
      "\n",
      "16 predict: Mẹ tôi , , , , , <NUM> khi khi khi - - - - một một , hôn hôn hôn hôn hôn hôn <NUM> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "16 actual: Mẹ tôi , Mai , mới <NUM> tuổi khi ba của mẹ mất - - đã lập gia đình , một cuộc hôn nhân sắp đặt trước , đã có hai cô con gái nhỏ .\n",
      "\n",
      "17 predict: Với với sống đời đời đời đã đã đã đã một một một một một một một sống sống sống sống sống sống sống . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "17 actual: Với mẹ , cuộc đời cô đọng vào nhiệm vụ duy nhất : để gia đình mẹ trốn thoát và bắt đầu cuộc sống mới ở Úc .\n",
      "\n",
      "18 predict: Nó không không không ấy cô cô cô sẽ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "18 actual: Mẹ không bao giờ chấp nhận được là mẹ sẽ không thành công .\n",
      "\n",
      "19 predict: Sau là <NUM> <NUM> <NUM> <NUM> <NUM> , , , , , thuyền thuyền ly ly ly cá cá . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "19 actual: Thế là sau bốn năm , một trường thiên đằng đẵng hơn cả trong truyện , một chiếc thuyền trườn ra biển nguỵ trang là thuyền đánh cá .\n",
      "\n"
     ]
    }
   ],
   "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
}