neural-machine-translation/13.basic-seq2seq-luong.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": 15,
   "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.LuongAttention(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": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "W0902 22:44:13.330178 139806083041088 deprecation.py:323] From <ipython-input-15-64e5c45dc7dd>: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": 17,
   "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": 18,
   "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": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [05:58<00:00,  2.91it/s, accuracy=0.0776, cost=6.88]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.65it/s, accuracy=0.0904, cost=6.7] \n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1, training avg loss 6.816859, training avg acc 0.072502\n",
      "epoch 1, testing avg loss 6.789898, testing avg acc 0.079024\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:08<00:00,  2.83it/s, accuracy=0.0712, cost=6.65]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.85it/s, accuracy=0.0904, cost=6.76]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 2, training avg loss 6.592953, training avg acc 0.083859\n",
      "epoch 2, testing avg loss 6.770093, testing avg acc 0.083861\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:08<00:00,  2.83it/s, accuracy=0.0776, cost=6.54]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.89it/s, accuracy=0.096, cost=6.44] \n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 3, training avg loss 6.412199, training avg acc 0.093731\n",
      "epoch 3, testing avg loss 6.512036, testing avg acc 0.094455\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:08<00:00,  2.83it/s, accuracy=0.0776, cost=6.83]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.97it/s, accuracy=0.096, cost=6.48] \n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 4, training avg loss 7.092470, training avg acc 0.084084\n",
      "epoch 4, testing avg loss 6.448802, testing avg acc 0.088475\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:09<00:00,  2.82it/s, accuracy=0.0647, cost=6.53]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.80it/s, accuracy=0.107, cost=6.33] \n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 5, training avg loss 6.285931, training avg acc 0.093607\n",
      "epoch 5, testing avg loss 6.311054, testing avg acc 0.095191\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:08<00:00,  2.83it/s, accuracy=0.0634, cost=7.58]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.77it/s, accuracy=0.0904, cost=6.89]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 6, training avg loss 10.369638, training avg acc 0.086577\n",
      "epoch 6, testing avg loss 7.008086, testing avg acc 0.068212\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:08<00:00,  2.83it/s, accuracy=0.0582, cost=6.9] \n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.90it/s, accuracy=0.0847, cost=6.59]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 7, training avg loss 6.616851, training avg acc 0.070791\n",
      "epoch 7, testing avg loss 6.507485, testing avg acc 0.075827\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:08<00:00,  2.83it/s, accuracy=0.0595, cost=6.74]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.97it/s, accuracy=0.0847, cost=6.44]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 8, training avg loss 6.336044, training avg acc 0.078293\n",
      "epoch 8, testing avg loss 6.371596, testing avg acc 0.082980\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:08<00:00,  2.83it/s, accuracy=0.0724, cost=6.58]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.94it/s, accuracy=0.0791, cost=6.4] \n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 9, training avg loss 6.224199, training avg acc 0.082430\n",
      "epoch 9, testing avg loss 6.332495, testing avg acc 0.078819\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:09<00:00,  2.82it/s, accuracy=0.0621, cost=6.99]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.92it/s, accuracy=0.0621, cost=6.61]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 10, training avg loss 9.369098, training avg acc 0.077706\n",
      "epoch 10, testing avg loss 6.741716, testing avg acc 0.053185\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:08<00:00,  2.83it/s, accuracy=0.0582, cost=6.66]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.95it/s, accuracy=0.0508, cost=6.53]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 11, training avg loss 6.381284, training avg acc 0.066436\n",
      "epoch 11, testing avg loss 6.455024, testing avg acc 0.063894\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:08<00:00,  2.83it/s, accuracy=0.0543, cost=6.59]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.83it/s, accuracy=0.0621, cost=6.47]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 12, training avg loss 6.222093, training avg acc 0.070198\n",
      "epoch 12, testing avg loss 6.344884, testing avg acc 0.071262\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:08<00:00,  2.83it/s, accuracy=0.0712, cost=6.78]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  7.01it/s, accuracy=0.0904, cost=6.46]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 13, training avg loss 6.136240, training avg acc 0.075627\n",
      "epoch 13, testing avg loss 6.371862, testing avg acc 0.078014\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:08<00:00,  2.83it/s, accuracy=0.066, cost=6.71] \n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.90it/s, accuracy=0.0791, cost=6.45]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 14, training avg loss 6.084103, training avg acc 0.079582\n",
      "epoch 14, testing avg loss 6.324430, testing avg acc 0.077470\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:09<00:00,  2.82it/s, accuracy=0.0815, cost=6.37]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.83it/s, accuracy=0.0904, cost=6.52]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 15, training avg loss 6.053008, training avg acc 0.082156\n",
      "epoch 15, testing avg loss 6.350252, testing avg acc 0.079757\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:07<00:00,  2.84it/s, accuracy=0.0815, cost=6.4] \n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.87it/s, accuracy=0.096, cost=6.51] \n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 16, training avg loss 6.021106, training avg acc 0.084047\n",
      "epoch 16, testing avg loss 6.375186, testing avg acc 0.080646\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:08<00:00,  2.83it/s, accuracy=0.0724, cost=6.33]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  7.01it/s, accuracy=0.0791, cost=6.39]\n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 17, training avg loss 5.997729, training avg acc 0.085079\n",
      "epoch 17, testing avg loss 6.308892, testing avg acc 0.082106\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:08<00:00,  2.83it/s, accuracy=0.0724, cost=6.35]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.87it/s, accuracy=0.096, cost=6.36] \n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 18, training avg loss 6.011559, training avg acc 0.083714\n",
      "epoch 18, testing avg loss 6.312873, testing avg acc 0.082769\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:08<00:00,  2.83it/s, accuracy=0.0776, cost=6.34]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.90it/s, accuracy=0.096, cost=6.48] \n",
      "minibatch loop:   0%|          | 0/1042 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 19, training avg loss 5.982512, training avg acc 0.086026\n",
      "epoch 19, testing avg loss 6.350138, testing avg acc 0.084315\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1042/1042 [06:08<00:00,  2.83it/s, accuracy=0.0349, cost=33.2]\n",
      "minibatch loop: 100%|██████████| 23/23 [00:03<00:00,  6.82it/s, accuracy=0.0113, cost=18.8]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 20, training avg loss 41.891171, training avg acc 0.057395\n",
      "epoch 20, testing avg loss 23.723181, testing avg acc 0.023959\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": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "rev_dictionary_to = {int(k): v for k, v in rev_dictionary_to.items()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(20, 99)"
      ]
     },
     "execution_count": 21,
     "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": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 predict: Nhận lõi lõi lõi Paul lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi\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: Nhận ta lõi ! Paul lõi Paul hứng . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "1 actual: Câu chuyện này chưa kết thúc .\n",
      "\n",
      "2 predict: Nhận xứ lõi lõi lõi lõi lõi lõi lõi lõi lõi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "2 actual: Nó là một trò chơi ghép hình vẫn đang được xếp .\n",
      "\n",
      "3 predict: Nhận lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "3 actual: Hãy để tôi kể cho các bạn về vài mảnh ghép nhé .\n",
      "\n",
      "4 predict: Nhận lõi tế lõi , lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "4 actual: Hãy tưởng tượng mảnh đầu tiên : một người đàn ông đốt cháy sự nghiệp cả đời mình .\n",
      "\n",
      "5 predict: Nhận : lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\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: Nhận xứ lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\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: Nhận : xứ lõi Paul lõi Paul lõi Paul lõi Paul lõi Paul . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\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: Nhận lõi lõi lõi lõi lõi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "8 actual: Ông rút lui vào yên lặng .\n",
      "\n",
      "9 predict: Nhận : sử lõi lõi lõi , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "9 actual: Ông qua đời , bị lịch sử quật ngã .\n",
      "\n",
      "10 predict: Nhận lõi lõi lõi lõi lõi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "10 actual: Ông là ông của tôi .\n",
      "\n",
      "11 predict: Nhận xứ lõi lõi lõi lõi lõi lõi lõi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "11 actual: Tôi chưa bao giờ gặp ông ngoài đời .\n",
      "\n",
      "12 predict: Nhận <NUM> lõi lõi Paul lõi lõi lõi lõi lõi Paul . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "12 actual: Nhưng cuộc đời ta nhiều hơn những gì ta lưu trong kí ức nhiều .\n",
      "\n",
      "13 predict: Nhận lõi không lõi lõi lõi lõi lõi lõi lõi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "13 actual: Bà tôi chưa bao giờ cho phép tôi quên cuộc đời của ông .\n",
      "\n",
      "14 predict: Nhận : không lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\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: Nhận : lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\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: Nhận : thu lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\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: Nhận xứ lõi lõi Paul lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\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: Paul ấy tế lõi lõi lõi lõi lõi lõi lõi lõi lõi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\n",
      "18 actual: Mẹ không bao giờ chấp nhận được là mẹ sẽ không thành công .\n",
      "\n",
      "19 predict: Nhận xứ lõi lõi Paul lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi lõi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .\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
}