neural-machine-translation/20.lstm-birnn-seq2seq-bahdanau.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "os.environ['CUDA_VISIBLE_DEVICES'] = '2'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "import json"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open('dataset-bpe.json') as fopen:\n",
    "    data = json.load(fopen)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_X = data['train_X']\n",
    "train_Y = data['train_Y']\n",
    "test_X = data['test_X']\n",
    "test_Y = data['test_Y']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "EOS = 2\n",
    "GO = 1\n",
    "vocab_size = 32000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_Y = [i + [2] for i in train_Y]\n",
    "test_Y = [i + [2] for i in test_Y]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensor2tensor.utils import beam_search\n",
    "\n",
    "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, learning_rate):\n",
    "        \n",
    "        def cells(size_layer=size_layer, reuse=False):\n",
    "            return tf.nn.rnn_cell.LSTMCell(size_layer,initializer=tf.orthogonal_initializer(),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",
    "        embeddings = tf.Variable(tf.random_uniform([vocab_size, embedded_size], -1, 1))\n",
    "        \n",
    "        def forward(x, y, reuse = False):\n",
    "            batch_size = tf.shape(x)[0]\n",
    "            X_seq_len = tf.count_nonzero(x, 1, dtype = tf.int32)\n",
    "            Y_seq_len = tf.count_nonzero(y, 1, dtype = tf.int32)\n",
    "            with tf.variable_scope('model',reuse=reuse):\n",
    "                encoder_embedded = tf.nn.embedding_lookup(embeddings, x)\n",
    "                decoder_embedded = tf.nn.embedding_lookup(embeddings, y)\n",
    "                for n in range(num_layers):\n",
    "                    (out_fw, out_bw), (state_fw, state_bw) = tf.nn.bidirectional_dynamic_rnn(\n",
    "                        cell_fw = cells(size_layer // 2),\n",
    "                        cell_bw = cells(size_layer // 2),\n",
    "                        inputs = encoder_embedded,\n",
    "                        sequence_length = self.X_seq_len,\n",
    "                        dtype = tf.float32,\n",
    "                        scope = 'bidirectional_rnn_%d'%(n))\n",
    "                    encoder_embedded = tf.concat((out_fw, out_bw), 2)\n",
    "                    \n",
    "                bi_state_c = tf.concat((state_fw.c, state_bw.c), -1)\n",
    "                bi_state_h = tf.concat((state_fw.h, state_bw.h), -1)\n",
    "                bi_lstm_state = tf.nn.rnn_cell.LSTMStateTuple(c=bi_state_c, h=bi_state_h)\n",
    "                last_state = tuple([bi_lstm_state] * num_layers)\n",
    "                last_output = tf.concat((out_fw,out_bw), -1)\n",
    "\n",
    "            with tf.variable_scope(\"decoder\",reuse=reuse):\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=Y_seq_len,\n",
    "                                               initial_state = initial_state,\n",
    "                                               dtype = tf.float32)\n",
    "                \n",
    "                return tf.layers.dense(outputs,vocab_size)\n",
    "            \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",
    "        self.training_logits = forward(self.X, decoder_input, reuse = False)\n",
    "        \n",
    "        self.training_logits = self.training_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))\n",
    "        \n",
    "        initial_ids = tf.fill([batch_size], GO)\n",
    "        def symbols_to_logits(ids):\n",
    "            x = tf.contrib.seq2seq.tile_batch(self.X, 1)\n",
    "            logits = forward(x, ids, reuse = True)\n",
    "            return logits[:, tf.shape(ids)[1]-1, :]\n",
    "        \n",
    "        final_ids, final_probs, _ = beam_search.beam_search(\n",
    "            symbols_to_logits,\n",
    "            initial_ids,\n",
    "            1,\n",
    "            tf.reduce_max(self.X_seq_len),\n",
    "            vocab_size,\n",
    "            0.0,\n",
    "            eos_id = EOS)\n",
    "        \n",
    "        self.fast_result = final_ids"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "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": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/tensorflow_core/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",
      "WARNING:tensorflow:From <ipython-input-7-83d9d0d26920>:11: LSTMCell.__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.LSTMCell, and will be replaced by that in Tensorflow 2.0.\n",
      "WARNING:tensorflow:From <ipython-input-7-83d9d0d26920>:36: bidirectional_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.Bidirectional(keras.layers.RNN(cell))`, which is equivalent to this API\n",
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/tensorflow_core/python/ops/rnn.py:464: 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",
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/tensorflow_core/python/ops/rnn_cell_impl.py:958: Layer.add_variable (from tensorflow.python.keras.engine.base_layer) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use `layer.add_weight` method instead.\n",
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/tensorflow_core/python/ops/rnn_cell_impl.py:962: 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",
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/tensorflow_core/python/ops/rnn.py:244: where (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",
      "WARNING:tensorflow:\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",
      "WARNING:tensorflow:From <ipython-input-7-83d9d0d26920>:50: 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",
      "WARNING:tensorflow:From <ipython-input-7-83d9d0d26920>:60: 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",
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/tensorflow_core/python/layers/core.py:187: Layer.apply (from tensorflow.python.keras.engine.base_layer) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use `layer.__call__` method instead.\n",
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/tensor2tensor/utils/beam_search.py:745: to_float (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use `tf.cast` instead.\n"
     ]
    }
   ],
   "source": [
    "tf.reset_default_graph()\n",
    "sess = tf.InteractiveSession()\n",
    "model = Translator(size_layer, num_layers, embedded_size, learning_rate)\n",
    "sess.run(tf.global_variables_initializer())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "pad_sequences = tf.keras.preprocessing.sequence.pad_sequences"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[array([[[    1, 29677, 30133, 30133, 15849, 15849, 17990, 26783,  9296,\n",
       "          17623, 17623, 30450, 30450, 30450,  2668,  2668,  2668,  2668,\n",
       "           9979,  9979, 10817, 25898, 25898,  2415, 31880, 18830, 18830,\n",
       "          31968, 31860, 31860, 31779, 31779, 31779, 31779, 31779, 30889,\n",
       "          10343]],\n",
       " \n",
       "        [[    1, 28625,  2824,  1555,  1555,   567,  4665,  4665,  4665,\n",
       "           4159, 25073, 22365, 22365, 25073, 25073, 22365, 22365, 22365,\n",
       "           1512,  1512, 22365, 22365,   526,   526, 22365, 22365,   526,\n",
       "          22365,  8708,  8708,  8708,  8708,  8708, 14406, 14406, 14406,\n",
       "          25357]],\n",
       " \n",
       "        [[    1, 30796,  8794, 15219,  9681, 26029, 26029, 26029, 26029,\n",
       "          17587, 17587, 17587, 17587, 10464, 10464, 10464, 21543, 16846,\n",
       "          16846, 12605, 12605,  4147,  8431, 19384, 14059, 14059,  1435,\n",
       "          22956, 22956, 12922, 17927, 17927, 21663, 21663, 10535, 27358,\n",
       "          27358]],\n",
       " \n",
       "        [[    1, 31714, 31714, 28571, 28571,  6850, 17524, 17524, 28381,\n",
       "          28381, 28381,  8733, 22211, 22211,  7192,  7192, 11377, 11377,\n",
       "          30897, 30897, 11390, 11390, 11390, 24835, 18748, 18748,  4568,\n",
       "          24835, 22142,   893, 22142,    68, 19518, 19518, 26179, 11528,\n",
       "          20166]],\n",
       " \n",
       "        [[    1, 13466, 26408, 18892, 22007, 27052, 23235,  4069,  4069,\n",
       "          29585, 22766, 22766, 13147, 13147, 13147, 27424, 27424, 21474,\n",
       "          21474,  2656,  2656,  2656, 29531,  1177,  1177, 22270,  3658,\n",
       "          29097, 29097,  7023,  8830,  7702,  7702, 14232, 14232, 10014,\n",
       "          10014]],\n",
       " \n",
       "        [[    1, 20497, 23054, 23054,  5279, 15070, 15070, 18520, 18883,\n",
       "          12342, 12342, 27659, 15949, 15949, 15949, 18654, 18654, 18654,\n",
       "          12661, 12661, 12661, 12661, 19096, 19096, 30439, 30439, 30439,\n",
       "          31804, 31804, 31804,  8403, 25440, 25440, 14680, 27067, 27067,\n",
       "          27067]],\n",
       " \n",
       "        [[    1, 19684, 22355, 22355, 26875, 26875, 26875, 26875, 26875,\n",
       "          27079, 27079, 27079, 30184, 16978, 16978, 16978, 10372, 10372,\n",
       "          10372, 10372,  9004,  9004,  9004,  9004,  9004, 15619, 15712,\n",
       "          15712, 15712, 15712, 15712, 31474, 30561, 30561, 30561, 23194,\n",
       "          10844]],\n",
       " \n",
       "        [[    1,  5385, 29135, 29135,  4818, 15122, 15122, 18116, 18116,\n",
       "          14234, 17984, 17984, 17984, 17984, 18441, 18441,  1092,  3161,\n",
       "          10841,  4278,  4278, 26077, 26077, 26077, 26878,  1920,  5682,\n",
       "           5682,  5682,  5682,  5971,  2937,  2937,  2937, 28584,  3637,\n",
       "          30212]],\n",
       " \n",
       "        [[    1, 26365, 26365, 26365,  8293,  7628,  7628, 10133,  7980,\n",
       "           7980,  7980, 26537,   627,   627,   627,   627,   627, 13805,\n",
       "            877,   877,   877,   877,   877,   877, 30117,  9406,  9406,\n",
       "           9406,  9406,  9406,  9406,  9406, 26735, 26735,  2897,  2897,\n",
       "           2897]],\n",
       " \n",
       "        [[    1, 28441,  4381,  4381,  4381,  4381, 21041, 21041, 21041,\n",
       "          21041, 21041, 21041, 11387, 11387, 11387, 14820, 14820, 14820,\n",
       "          18413, 18413, 18413, 18413, 18413,  2800, 29814, 15107, 15107,\n",
       "          15107, 26337, 17125, 26337, 26337, 17125,  1885,  1885,  2155,\n",
       "           2155]]], dtype=int32), 10.374287, 0.0]"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "batch_x = pad_sequences(train_X[:10], padding='post')\n",
    "batch_y = pad_sequences(train_Y[:10], padding='post')\n",
    "\n",
    "sess.run([model.fast_result, model.cost, model.accuracy], \n",
    "         feed_dict = {model.X: batch_x, model.Y: batch_y})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1563/1563 [12:11<00:00,  2.14it/s, accuracy=0.126, cost=6.27]\n",
      "minibatch loop: 100%|██████████| 40/40 [00:08<00:00,  4.91it/s, accuracy=0.167, cost=6.07]\n",
      "minibatch loop:   0%|          | 0/1563 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1, training avg loss 6.778524, training avg acc 0.126677\n",
      "epoch 1, testing avg loss 6.146084, testing avg acc 0.153073\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1563/1563 [12:18<00:00,  2.12it/s, accuracy=0.161, cost=5.41]\n",
      "minibatch loop: 100%|██████████| 40/40 [00:07<00:00,  5.08it/s, accuracy=0.145, cost=5.75]\n",
      "minibatch loop:   0%|          | 0/1563 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 2, training avg loss 5.781809, training avg acc 0.165808\n",
      "epoch 2, testing avg loss 5.681263, testing avg acc 0.169008\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1563/1563 [12:25<00:00,  2.10it/s, accuracy=0.19, cost=4.81] \n",
      "minibatch loop: 100%|██████████| 40/40 [00:08<00:00,  4.94it/s, accuracy=0.145, cost=5.55]\n",
      "minibatch loop:   0%|          | 0/1563 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 3, training avg loss 5.277455, training avg acc 0.183807\n",
      "epoch 3, testing avg loss 5.506363, testing avg acc 0.170784\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1563/1563 [12:26<00:00,  2.09it/s, accuracy=0.227, cost=4.31]\n",
      "minibatch loop: 100%|██████████| 40/40 [00:08<00:00,  4.91it/s, accuracy=0.151, cost=5.36]\n",
      "minibatch loop:   0%|          | 0/1563 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 4, training avg loss 4.922794, training avg acc 0.199603\n",
      "epoch 4, testing avg loss 5.446328, testing avg acc 0.172173\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1563/1563 [12:26<00:00,  2.09it/s, accuracy=0.271, cost=3.91]\n",
      "minibatch loop: 100%|██████████| 40/40 [00:08<00:00,  4.88it/s, accuracy=0.161, cost=5.35]\n",
      "minibatch loop:   0%|          | 0/1563 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 5, training avg loss 4.643044, training avg acc 0.215329\n",
      "epoch 5, testing avg loss 5.441227, testing avg acc 0.173015\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1563/1563 [12:27<00:00,  2.09it/s, accuracy=0.269, cost=3.76]\n",
      "minibatch loop: 100%|██████████| 40/40 [00:08<00:00,  4.95it/s, accuracy=0.199, cost=5.27]\n",
      "minibatch loop:   0%|          | 0/1563 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 6, training avg loss 4.417888, training avg acc 0.230713\n",
      "epoch 6, testing avg loss 5.441321, testing avg acc 0.172314\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1563/1563 [12:27<00:00,  2.09it/s, accuracy=0.301, cost=3.53]\n",
      "minibatch loop: 100%|██████████| 40/40 [00:08<00:00,  4.91it/s, accuracy=0.21, cost=5.27] \n",
      "minibatch loop:   0%|          | 0/1563 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 7, training avg loss 4.234431, training avg acc 0.246316\n",
      "epoch 7, testing avg loss 5.470088, testing avg acc 0.168257\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop: 100%|██████████| 1563/1563 [13:17<00:00,  1.96it/s, accuracy=0.337, cost=3.25]\n",
      "minibatch loop: 100%|██████████| 40/40 [00:08<00:00,  4.93it/s, accuracy=0.172, cost=5.43]\n",
      "minibatch loop:   0%|          | 0/1563 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 8, training avg loss 4.075385, training avg acc 0.262108\n",
      "epoch 8, testing avg loss 5.628513, testing avg acc 0.166524\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop:  12%|█▏        | 191/1563 [01:29<11:02,  2.07it/s, accuracy=0.26, cost=4.16] IOPub message rate exceeded.\n",
      "The notebook server will temporarily stop sending output\n",
      "to the client in order to avoid crashing it.\n",
      "To change this limit, set the config variable\n",
      "`--NotebookApp.iopub_msg_rate_limit`.\n",
      "\n",
      "Current values:\n",
      "NotebookApp.iopub_msg_rate_limit=1000.0 (msgs/sec)\n",
      "NotebookApp.rate_limit_window=3.0 (secs)\n",
      "\n",
      "minibatch loop: 100%|██████████| 1563/1563 [12:28<00:00,  2.09it/s, accuracy=0.362, cost=3.07]\n",
      "minibatch loop: 100%|██████████| 40/40 [00:07<00:00,  5.03it/s, accuracy=0.188, cost=5.44]\n",
      "minibatch loop:   0%|          | 0/1563 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 10, training avg loss 3.831156, training avg acc 0.287826\n",
      "epoch 10, testing avg loss 5.751677, testing avg acc 0.166575\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop:  64%|██████▎   | 994/1563 [08:40<04:02,  2.34it/s, accuracy=0.331, cost=3.6] IOPub message rate exceeded.\n",
      "The notebook server will temporarily stop sending output\n",
      "to the client in order to avoid crashing it.\n",
      "To change this limit, set the config variable\n",
      "`--NotebookApp.iopub_msg_rate_limit`.\n",
      "\n",
      "Current values:\n",
      "NotebookApp.iopub_msg_rate_limit=1000.0 (msgs/sec)\n",
      "NotebookApp.rate_limit_window=3.0 (secs)\n",
      "\n",
      "minibatch loop: 100%|██████████| 1563/1563 [12:30<00:00,  2.08it/s, accuracy=0.426, cost=2.68]\n",
      "minibatch loop: 100%|██████████| 40/40 [00:08<00:00,  4.88it/s, accuracy=0.151, cost=5.63]\n",
      "minibatch loop:   0%|          | 0/1563 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 13, training avg loss 3.531115, training avg acc 0.326408\n",
      "epoch 13, testing avg loss 6.021354, testing avg acc 0.159851\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop:  18%|█▊        | 288/1563 [02:16<09:27,  2.25it/s, accuracy=0.323, cost=3.58]IOPub message rate exceeded.\n",
      "The notebook server will temporarily stop sending output\n",
      "to the client in order to avoid crashing it.\n",
      "To change this limit, set the config variable\n",
      "`--NotebookApp.iopub_msg_rate_limit`.\n",
      "\n",
      "Current values:\n",
      "NotebookApp.iopub_msg_rate_limit=1000.0 (msgs/sec)\n",
      "NotebookApp.rate_limit_window=3.0 (secs)\n",
      "\n",
      "minibatch loop: 100%|██████████| 1563/1563 [12:23<00:00,  2.10it/s, accuracy=0.464, cost=2.53]\n",
      "minibatch loop: 100%|██████████| 40/40 [00:08<00:00,  4.98it/s, accuracy=0.172, cost=5.85]\n",
      "minibatch loop:   0%|          | 0/1563 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 15, training avg loss 3.374760, training avg acc 0.348911\n",
      "epoch 15, testing avg loss 6.183997, testing avg acc 0.156699\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop:  68%|██████▊   | 1069/1563 [08:18<03:57,  2.08it/s, accuracy=0.324, cost=3.55]IOPub message rate exceeded.\n",
      "The notebook server will temporarily stop sending output\n",
      "to the client in order to avoid crashing it.\n",
      "To change this limit, set the config variable\n",
      "`--NotebookApp.iopub_msg_rate_limit`.\n",
      "\n",
      "Current values:\n",
      "NotebookApp.iopub_msg_rate_limit=1000.0 (msgs/sec)\n",
      "NotebookApp.rate_limit_window=3.0 (secs)\n",
      "\n",
      "minibatch loop: 100%|██████████| 1563/1563 [11:57<00:00,  2.18it/s, accuracy=0.471, cost=2.4] \n",
      "minibatch loop: 100%|██████████| 40/40 [00:07<00:00,  5.14it/s, accuracy=0.172, cost=6.17]\n",
      "minibatch loop:   0%|          | 0/1563 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 18, training avg loss 3.179215, training avg acc 0.378525\n",
      "epoch 18, testing avg loss 6.412740, testing avg acc 0.155674\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "minibatch loop:  25%|██▌       | 391/1563 [02:56<07:37,  2.56it/s, accuracy=0.407, cost=3.04]IOPub message rate exceeded.\n",
      "The notebook server will temporarily stop sending output\n",
      "to the client in order to avoid crashing it.\n",
      "To change this limit, set the config variable\n",
      "`--NotebookApp.iopub_msg_rate_limit`.\n",
      "\n",
      "Current values:\n",
      "NotebookApp.iopub_msg_rate_limit=1000.0 (msgs/sec)\n",
      "NotebookApp.rate_limit_window=3.0 (secs)\n",
      "\n",
      "minibatch loop: 100%|██████████| 1563/1563 [11:58<00:00,  2.18it/s, accuracy=0.507, cost=2.26]\n",
      "minibatch loop: 100%|██████████| 40/40 [00:07<00:00,  5.13it/s, accuracy=0.167, cost=6.18]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 20, training avg loss 3.081857, training avg acc 0.393816\n",
      "epoch 20, testing avg loss 6.668028, testing avg acc 0.157161\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",
    "        batch_x = pad_sequences(train_X[i : index], padding='post')\n",
    "        batch_y = pad_sequences(train_Y[i : index], padding='post')\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 = pad_sequences(test_X[i : index], padding='post')\n",
    "        batch_y = pad_sequences(test_Y[i : index], padding='post')\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": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensor2tensor.utils import bleu_hook"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 40/40 [04:33<00:00,  6.84s/it]\n"
     ]
    }
   ],
   "source": [
    "results = []\n",
    "for i in tqdm.tqdm(range(0, len(test_X), batch_size)):\n",
    "    index = min(i + batch_size, len(test_X))\n",
    "    batch_x = pad_sequences(test_X[i : index], padding='post')\n",
    "    feed = {model.X: batch_x}\n",
    "    p = sess.run(model.fast_result,feed_dict = feed)[:,0,:]\n",
    "    result = []\n",
    "    for row in p:\n",
    "        result.append([i for i in row if i > 3])\n",
    "    results.extend(result)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "rights = []\n",
    "for r in test_Y:\n",
    "    rights.append([i for i in r if i > 3])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.054097746"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "bleu_hook.compute_bleu(reference_corpus = rights,\n",
    "                       translation_corpus = results)"
   ]
  },
  {
   "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
}