dependency-parser/6.bert-biaffine-attention-cross-entropy.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# !wget https://raw.githubusercontent.com/UniversalDependencies/UD_English-EWT/master/en_ewt-ud-dev.conllu\n",
    "# !wget https://raw.githubusercontent.com/UniversalDependencies/UD_English-EWT/master/en_ewt-ud-train.conllu\n",
    "# !wget https://raw.githubusercontent.com/UniversalDependencies/UD_English-EWT/master/en_ewt-ud-test.conllu\n",
    "# !wget https://storage.googleapis.com/bert_models/2018_10_18/cased_L-12_H-768_A-12.zip\n",
    "# !unzip cased_L-12_H-768_A-12.zip"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "os.environ['CUDA_VISIBLE_DEVICES'] = '1'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "tag2idx = {'PAD': 0, 'X': 1}\n",
    "tag_idx = 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "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"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/bert/optimization.py:87: The name tf.train.Optimizer is deprecated. Please use tf.compat.v1.train.Optimizer instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/bert/tokenization.py:125: The name tf.gfile.GFile is deprecated. Please use tf.io.gfile.GFile instead.\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/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 bert\n",
    "from bert import run_classifier\n",
    "from bert import optimization\n",
    "from bert import tokenization\n",
    "from bert import modeling\n",
    "import tensorflow as tf\n",
    "import numpy as np\n",
    "\n",
    "BERT_VOCAB = 'cased_L-12_H-768_A-12/vocab.txt'\n",
    "BERT_INIT_CHKPNT = 'cased_L-12_H-768_A-12/bert_model.ckpt'\n",
    "BERT_CONFIG = 'cased_L-12_H-768_A-12/bert_config.json'\n",
    "\n",
    "tokenizer = tokenization.FullTokenizer(\n",
    "      vocab_file=BERT_VOCAB, do_lower_case=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "def process_corpus(corpus, until = None):\n",
    "    global word2idx, tag2idx, char2idx, word_idx, tag_idx, char_idx\n",
    "    sentences, words, depends, labels, pos, sequences = [], [], [], [], [], []\n",
    "    temp_sentence, temp_word, temp_depend, temp_label, temp_pos = [], [], [], [], []\n",
    "    first_time = True\n",
    "    for sentence in corpus:\n",
    "        try:\n",
    "            if len(sentence):\n",
    "                if sentence[0] == '#':\n",
    "                    continue\n",
    "                if first_time:\n",
    "                    print(sentence)\n",
    "                    first_time = False\n",
    "                sentence = sentence.split('\\t')\n",
    "                if sentence[7] not in tag2idx:\n",
    "                    tag2idx[sentence[7]] = tag_idx\n",
    "                    tag_idx += 1\n",
    "                temp_word.append(sentence[1])\n",
    "                temp_depend.append(int(sentence[6]) + 1)\n",
    "                temp_label.append(tag2idx[sentence[7]])\n",
    "                temp_sentence.append(sentence[1])\n",
    "                temp_pos.append(sentence[3])\n",
    "            else:\n",
    "                if len(temp_sentence) < 2 or len(temp_word) != len(temp_label):\n",
    "                    temp_word = []\n",
    "                    temp_depend = []\n",
    "                    temp_label = []\n",
    "                    temp_sentence = []\n",
    "                    temp_pos = []\n",
    "                    continue\n",
    "                bert_tokens = ['[CLS]']\n",
    "                labels_ = [0]\n",
    "                depends_ = [0]\n",
    "                seq_ = []\n",
    "                for no, orig_token in enumerate(temp_word):\n",
    "                    labels_.append(temp_label[no])\n",
    "                    depends_.append(temp_depend[no])\n",
    "                    t = tokenizer.tokenize(orig_token)\n",
    "                    bert_tokens.extend(t)\n",
    "                    labels_.extend([1] * (len(t) - 1))\n",
    "                    depends_.extend([0] * (len(t) - 1))\n",
    "                    seq_.append(no + 1)\n",
    "                bert_tokens.append('[SEP]')\n",
    "                labels_.append(0)\n",
    "                depends_.append(0)\n",
    "                words.append(tokenizer.convert_tokens_to_ids(bert_tokens))\n",
    "                depends.append(depends_)\n",
    "                labels.append(labels_)\n",
    "                sentences.append(temp_sentence)\n",
    "                pos.append(temp_pos)\n",
    "                sequences.append(seq_)\n",
    "                temp_word = []\n",
    "                temp_depend = []\n",
    "                temp_label = []\n",
    "                temp_sentence = []\n",
    "                temp_pos = []\n",
    "        except Exception as e:\n",
    "            print(e, sentence)\n",
    "    return sentences[:-1], words[:-1], depends[:-1], labels[:-1], pos[:-1], sequences[:-1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1\tFrom\tfrom\tADP\tIN\t_\t3\tcase\t3:case\t_\n",
      "invalid literal for int() with base 10: '_' ['10.1', 'has', 'have', 'VERB', 'VBZ', '_', '_', '_', '8:parataxis', 'CopyOf=-1']\n",
      "invalid literal for int() with base 10: '_' ['21.1', 'has', 'have', 'VERB', 'VBZ', '_', '_', '_', '16:conj:and', 'CopyOf=-1']\n"
     ]
    }
   ],
   "source": [
    "with open('en_ewt-ud-dev.conllu') as fopen:\n",
    "    dev = fopen.read().split('\\n')\n",
    "\n",
    "sentences_dev, words_dev, depends_dev, labels_dev, _, seq_dev = process_corpus(dev)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1\tWhat\twhat\tPRON\tWP\tPronType=Int\t0\troot\t0:root\t_\n",
      "invalid literal for int() with base 10: '_' ['24.1', 'left', 'left', 'VERB', 'VBN', 'Tense=Past|VerbForm=Part', '_', '_', '6:parataxis', 'CopyOf=6']\n"
     ]
    }
   ],
   "source": [
    "with open('en_ewt-ud-test.conllu') as fopen:\n",
    "    test = fopen.read().split('\\n')\n",
    "\n",
    "sentences_test, words_test, depends_test, labels_test, _, seq_test = process_corpus(test)\n",
    "sentences_test.extend(sentences_dev)\n",
    "words_test.extend(words_dev)\n",
    "depends_test.extend(depends_dev)\n",
    "labels_test.extend(labels_dev)\n",
    "seq_test.extend(seq_dev)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1\tAl\tAl\tPROPN\tNNP\tNumber=Sing\t0\troot\t0:root\tSpaceAfter=No\n",
      "invalid literal for int() with base 10: '_' ['8.1', 'reported', 'report', 'VERB', 'VBN', 'Tense=Past|VerbForm=Part|Voice=Pass', '_', '_', '5:conj:and', 'CopyOf=5']\n",
      "invalid literal for int() with base 10: '_' ['22.1', 'used', 'use', 'VERB', 'VBN', 'Tense=Past|VerbForm=Part', '_', '_', '13:advcl:with|17:conj:and', 'CopyOf=17']\n",
      "invalid literal for int() with base 10: '_' ['22.1', 'used', 'use', 'VERB', 'VBN', 'Tense=Past|VerbForm=Part', '_', '_', '13:advcl:with|17:conj:and', 'CopyOf=17']\n",
      "invalid literal for int() with base 10: '_' ['11.1', 'called', 'call', 'VERB', 'VBN', 'Tense=Past|VerbForm=Part|Voice=Pass', '_', '_', '3:conj:and', 'CopyOf=3']\n",
      "invalid literal for int() with base 10: '_' ['14.1', 'is', 'be', 'VERB', 'VBZ', '_', '_', '_', '1:conj:and', 'CopyOf=1']\n",
      "invalid literal for int() with base 10: '_' ['20.1', 'reflect', 'reflect', 'VERB', 'VBP', 'Mood=Ind|Tense=Pres|VerbForm=Fin', '_', '_', '7:acl:relcl|9:conj', 'CopyOf=9']\n",
      "invalid literal for int() with base 10: '_' ['21.1', 'recruited', 'recruit', 'VERB', 'VBD', 'Mood=Ind|Tense=Past|VerbForm=Fin', '_', '_', '9:conj:and', 'CopyOf=9']\n",
      "invalid literal for int() with base 10: '_' ['9.1', 'wish', 'wish', 'VERB', 'VBP', 'Mood=Ind|Tense=Pres|VerbForm=Fin', '_', '_', '2:conj:and', 'CopyOf=2']\n",
      "invalid literal for int() with base 10: '_' ['38.1', 'supplied', 'supply', 'VERB', 'VBN', 'Tense=Past|VerbForm=Part|Voice=Pass', '_', '_', '16:conj:and', 'CopyOf=16']\n",
      "invalid literal for int() with base 10: '_' ['18.1', 'keep', 'keep', 'VERB', 'VB', 'Mood=Imp|VerbForm=Fin', '_', '_', '14:conj:and', 'CopyOf=14']\n",
      "invalid literal for int() with base 10: '_' ['21.1', 'keep', 'keep', 'VERB', 'VB', 'Mood=Imp|VerbForm=Fin', '_', '_', '14:conj:and', 'CopyOf=14']\n",
      "invalid literal for int() with base 10: '_' ['18.1', 'mean', 'mean', 'VERB', 'VB', 'VerbForm=Inf', '_', '_', '8:conj', 'CopyOf=8']\n",
      "invalid literal for int() with base 10: '_' ['30.1', 'play', 'play', 'VERB', 'VBP', 'Mood=Ind|Tense=Pres|VerbForm=Fin', '_', '_', '18:acl:relcl|27:conj:but', 'CopyOf=27']\n",
      "invalid literal for int() with base 10: '_' ['22.1', 'have', 'have', 'VERB', 'VBP', 'Mood=Ind|Tense=Pres|VerbForm=Fin', '_', '_', '17:conj', 'CopyOf=17']\n",
      "invalid literal for int() with base 10: '_' ['27.1', 'have', 'have', 'VERB', 'VBP', 'Mood=Ind|Tense=Pres|VerbForm=Fin', '_', '_', '17:conj', 'CopyOf=17']\n",
      "invalid literal for int() with base 10: '_' ['49.1', 'helped', 'help', 'VERB', 'VBD', '_', '_', '_', '38:conj:but', 'CopyOf=38']\n",
      "invalid literal for int() with base 10: '_' ['7.1', 'found', 'find', 'VERB', 'VBD', 'Mood=Ind|Tense=Past|VerbForm=Fin', '_', '_', '3:conj', 'CopyOf=3']\n",
      "invalid literal for int() with base 10: '_' ['10.1', 'excited', 'excited', 'ADJ', 'JJ', 'Degree=Pos', '_', '_', '4:advcl', 'CopyOf=4']\n",
      "invalid literal for int() with base 10: '_' ['15.1', \"'s\", 'be', 'VERB', 'VBZ', '_', '_', '_', '2:conj:and', 'CopyOf=2']\n",
      "invalid literal for int() with base 10: '_' ['25.1', 'took', 'take', 'VERB', 'VBD', 'Mood=Ind|Tense=Past|VerbForm=Fin', '_', '_', '17:conj:and', 'CopyOf=17']\n",
      "invalid literal for int() with base 10: '_' ['10.1', 'loss', 'lose', 'VERB', 'VBD', 'Mood=Ind|Tense=Past|VerbForm=Fin', '_', '_', '3:conj:and', 'CopyOf=3']\n",
      "invalid literal for int() with base 10: '_' ['11.1', 'leave', 'leave', 'VERB', 'VB', 'VerbForm=Inf', '_', '_', '7:parataxis', 'CopyOf=7']\n",
      "invalid literal for int() with base 10: '_' ['24.1', 'charge', 'charge', 'VERB', 'VBP', 'Mood=Ind|Tense=Pres|VerbForm=Fin', '_', '_', '16:conj:and', 'CopyOf=16']\n"
     ]
    }
   ],
   "source": [
    "with open('en_ewt-ud-train.conllu') as fopen:\n",
    "    train = fopen.read().split('\\n')\n",
    "\n",
    "sentences_train, words_train, depends_train, labels_train, _, _ = process_corpus(train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(12000, 3824)"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(sentences_train), len(sentences_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "idx2tag = {v:k for k, v in tag2idx.items()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_X = words_train\n",
    "train_Y = labels_train\n",
    "train_depends = depends_train\n",
    "\n",
    "test_X = words_test\n",
    "test_Y = labels_test\n",
    "test_depends = depends_test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "epoch = 15\n",
    "batch_size = 32\n",
    "warmup_proportion = 0.1\n",
    "num_train_steps = int(len(train_X) / batch_size * epoch)\n",
    "num_warmup_steps = int(num_train_steps * warmup_proportion)\n",
    "bert_config = modeling.BertConfig.from_json_file(BERT_CONFIG)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "class BiAAttention:\n",
    "    def __init__(self, input_size_encoder, input_size_decoder, num_labels):\n",
    "        self.input_size_encoder = input_size_encoder\n",
    "        self.input_size_decoder = input_size_decoder\n",
    "        self.num_labels = num_labels\n",
    "        \n",
    "        self.W_d = tf.get_variable(\"W_d\", shape=[self.num_labels, self.input_size_decoder],\n",
    "           initializer=tf.contrib.layers.xavier_initializer())\n",
    "        self.W_e = tf.get_variable(\"W_e\", shape=[self.num_labels, self.input_size_encoder],\n",
    "           initializer=tf.contrib.layers.xavier_initializer())\n",
    "        self.U = tf.get_variable(\"U\", shape=[self.num_labels, self.input_size_decoder, self.input_size_encoder],\n",
    "           initializer=tf.contrib.layers.xavier_initializer())\n",
    "        \n",
    "    def forward(self, input_d, input_e, mask_d=None, mask_e=None):\n",
    "        batch = tf.shape(input_d)[0]\n",
    "        length_decoder = tf.shape(input_d)[1]\n",
    "        length_encoder = tf.shape(input_e)[1]\n",
    "        out_d = tf.expand_dims(tf.matmul(self.W_d, tf.transpose(input_d, [0, 2, 1])), 3)\n",
    "        out_e = tf.expand_dims(tf.matmul(self.W_e, tf.transpose(input_e, [0, 2, 1])), 2)\n",
    "        output = tf.matmul(tf.expand_dims(input_d, 1), self.U)\n",
    "        output = tf.matmul(output, tf.transpose(tf.expand_dims(input_e, 1), [0, 1, 3, 2]))\n",
    "        \n",
    "        output = output + out_d + out_e\n",
    "        \n",
    "        if mask_d is not None:\n",
    "            d = tf.expand_dims(tf.expand_dims(mask_d, 1), 3)\n",
    "            e = tf.expand_dims(tf.expand_dims(mask_e, 1), 2)\n",
    "            output = output * d * e\n",
    "            \n",
    "        return output\n",
    "    \n",
    "class BiLinear:\n",
    "    def __init__(self, left_features, right_features, out_features):\n",
    "        self.left_features = left_features\n",
    "        self.right_features = right_features\n",
    "        self.out_features = out_features\n",
    "        \n",
    "        self.U = tf.get_variable(\"U-bi\", shape=[out_features, left_features, right_features],\n",
    "           initializer=tf.contrib.layers.xavier_initializer())\n",
    "        self.W_l = tf.get_variable(\"Wl\", shape=[out_features, left_features],\n",
    "           initializer=tf.contrib.layers.xavier_initializer())\n",
    "        self.W_r = tf.get_variable(\"Wr\", shape=[out_features, right_features],\n",
    "           initializer=tf.contrib.layers.xavier_initializer())\n",
    "    \n",
    "    def forward(self, input_left, input_right):\n",
    "        left_size = tf.shape(input_left)\n",
    "        output_shape = tf.concat([left_size[:-1], [self.out_features]], axis = 0)\n",
    "        batch = tf.cast(tf.reduce_prod(left_size[:-1]), tf.int32)\n",
    "        input_left = tf.reshape(input_left, (batch, self.left_features))\n",
    "        input_right = tf.reshape(input_right, (batch, self.right_features))\n",
    "        tiled = tf.tile(tf.expand_dims(input_left, axis = 0), (self.out_features,1,1))\n",
    "        output = tf.transpose(tf.reduce_sum(tf.matmul(tiled, self.U), axis = 2))\n",
    "        output = output + tf.matmul(input_left, tf.transpose(self.W_l))\\\n",
    "        + tf.matmul(input_right, tf.transpose(self.W_r))\n",
    "        \n",
    "        return tf.reshape(output, output_shape)\n",
    "\n",
    "class Attention:\n",
    "    def __init__(self, word_dim, num_words, char_dim, num_chars, num_filters, kernel_size,\n",
    "                 hidden_size, encoder_layers, num_labels, arc_space, type_space):\n",
    "        \n",
    "        def cells(size, reuse=False):\n",
    "            return tf.nn.rnn_cell.LSTMCell(size,\n",
    "                                           initializer=tf.orthogonal_initializer(),reuse=reuse)\n",
    "        \n",
    "        self.word_embedd = tf.Variable(tf.random_uniform([num_words, word_dim], -1, 1))\n",
    "        self.char_embedd = tf.Variable(tf.random_uniform([num_chars, char_dim], -1, 1))\n",
    "        self.conv1d = tf.layers.Conv1D(num_filters, kernel_size, 1, padding='VALID')\n",
    "        self.num_labels = num_labels\n",
    "        self.encoder = tf.nn.rnn_cell.MultiRNNCell([cells(hidden_size) for _ in range(encoder_layers)])\n",
    "\n",
    "        \n",
    "        \n",
    "    def encode(self, input_word, input_char):\n",
    "        word = tf.nn.embedding_lookup(self.word_embedd, input_word)\n",
    "        char = tf.nn.embedding_lookup(self.char_embedd, input_char)\n",
    "        b = tf.shape(char)[0]\n",
    "        wl = tf.shape(char)[1]\n",
    "        cl = tf.shape(char)[2]\n",
    "        d = char.shape[3]\n",
    "        char = tf.reshape(char, [b * wl, cl, d])\n",
    "        char = tf.reduce_max(self.conv1d(char), axis = 1)\n",
    "        char = tf.nn.tanh(char)\n",
    "        d = char.shape[-1]\n",
    "        char = tf.reshape(char, [b, wl, d])\n",
    "        \n",
    "        src_encoding = tf.concat([word, char], axis=2)\n",
    "        output, hn = tf.nn.dynamic_rnn(self.encoder, src_encoding, dtype = tf.float32,\n",
    "                                      scope = 'encoder')\n",
    "        arc_h = tf.nn.elu(self.arc_h(output))\n",
    "        arc_c = tf.nn.elu(self.arc_c(output))\n",
    "        \n",
    "        type_h = tf.nn.elu(self.type_h(output))\n",
    "        type_c = tf.nn.elu(self.type_c(output))\n",
    "        \n",
    "        return (arc_h, arc_c), (type_h, type_c), hn\n",
    "    \n",
    "    def forward(self, input_word, input_char, mask):\n",
    "        arcs, types, _ = self.encode(input_word, input_char)\n",
    "        \n",
    "        out_arc = tf.squeeze(self.attention.forward(arcs[0], arcs[1], mask_d=mask, mask_e=mask), axis = 1)\n",
    "        return out_arc, types, mask\n",
    "    \n",
    "    def loss(self, input_word, input_char, mask, heads, types):\n",
    "        out_arc, out_type, _ = self.forward(input_word, input_char, mask)\n",
    "        type_h, type_c = out_type\n",
    "        batch = tf.shape(out_arc)[0]\n",
    "        max_len = tf.shape(out_arc)[1]\n",
    "        batch_index = tf.range(0, batch)\n",
    "        t = tf.transpose(heads)\n",
    "        broadcasted = tf.broadcast_to(batch_index, tf.shape(t))\n",
    "        concatenated = tf.transpose(tf.concat([tf.expand_dims(broadcasted, axis = 0), \n",
    "                                               tf.expand_dims(t, axis = 0)], axis = 0))\n",
    "        type_h = tf.gather_nd(type_h, concatenated)\n",
    "        out_type = self.bilinear.forward(type_h, type_c)\n",
    "        minus_inf = -1e8\n",
    "        minus_mask = (1 - mask) * minus_inf\n",
    "        out_arc = out_arc + tf.expand_dims(minus_mask, axis = 2) + tf.expand_dims(minus_mask, axis = 1)\n",
    "        loss_arc = tf.nn.log_softmax(out_arc, dim=1)\n",
    "        loss_type = tf.nn.log_softmax(out_type, dim=2)\n",
    "        loss_arc = loss_arc * tf.expand_dims(mask, axis = 2) * tf.expand_dims(mask, axis = 1)\n",
    "        loss_type = loss_type * tf.expand_dims(mask, axis = 2)\n",
    "        num = tf.reduce_sum(mask) - tf.cast(batch, tf.float32)\n",
    "        child_index = tf.tile(tf.expand_dims(tf.range(0, max_len), 1), [1, batch])\n",
    "        t = tf.transpose(heads)\n",
    "        broadcasted = tf.broadcast_to(batch_index, tf.shape(t))\n",
    "        concatenated = tf.transpose(tf.concat([tf.expand_dims(broadcasted, axis = 0),\n",
    "                                               tf.expand_dims(t, axis = 0),\n",
    "                                               tf.expand_dims(child_index, axis = 0)], axis = 0))\n",
    "        loss_arc = tf.gather_nd(loss_arc, concatenated)\n",
    "        loss_arc = tf.transpose(loss_arc, [1, 0])\n",
    "        \n",
    "        t = tf.transpose(types)\n",
    "        broadcasted = tf.broadcast_to(batch_index, tf.shape(t))\n",
    "        concatenated = tf.transpose(tf.concat([tf.expand_dims(broadcasted, axis = 0),\n",
    "                                               tf.expand_dims(child_index, axis = 0),\n",
    "                                               tf.expand_dims(t, axis = 0)], axis = 0))\n",
    "        loss_type = tf.gather_nd(loss_type, concatenated)\n",
    "        loss_type = tf.transpose(loss_type, [1, 0])\n",
    "        return tf.reduce_sum(-loss_arc) / num, tf.reduce_sum(-loss_type) / num\n",
    "    \n",
    "    def decode(self, input_word, input_char, mask, leading_symbolic=0):\n",
    "        out_arc, out_type, _ = self.forward(input_word, input_char, mask)\n",
    "        batch = tf.shape(out_arc)[0]\n",
    "        max_len = tf.shape(out_arc)[1]\n",
    "        sec_max_len = tf.shape(out_arc)[2]\n",
    "        out_arc = out_arc + tf.linalg.diag(tf.fill([max_len], -np.inf))\n",
    "        minus_mask = tf.expand_dims(tf.cast(1 - mask, tf.bool), axis = 2)\n",
    "        minus_mask = tf.tile(minus_mask, [1, 1, sec_max_len])\n",
    "        out_arc = tf.where(minus_mask, tf.fill(tf.shape(out_arc), -np.inf), out_arc)\n",
    "        heads = tf.argmax(out_arc, axis = 1)\n",
    "        type_h, type_c = out_type\n",
    "        batch = tf.shape(type_h)[0]\n",
    "        max_len = tf.shape(type_h)[1]\n",
    "        batch_index = tf.range(0, batch)\n",
    "        t = tf.cast(tf.transpose(heads), tf.int32)\n",
    "        broadcasted = tf.broadcast_to(batch_index, tf.shape(t))\n",
    "        concatenated = tf.transpose(tf.concat([tf.expand_dims(broadcasted, axis = 0), \n",
    "                                               tf.expand_dims(t, axis = 0)], axis = 0))\n",
    "        type_h = tf.gather_nd(type_h, concatenated)\n",
    "        out_type = self.bilinear.forward(type_h, type_c)\n",
    "        out_type = out_type[:, :, leading_symbolic:]\n",
    "        types = tf.argmax(out_type, axis = 2)\n",
    "        return heads, types\n",
    "    \n",
    "class Model:\n",
    "    def __init__(\n",
    "        self,\n",
    "        learning_rate,\n",
    "        hidden_size_word,\n",
    "        cov = 0.0):\n",
    "        \n",
    "        def cells(size, reuse = False):\n",
    "            return tf.contrib.rnn.DropoutWrapper(\n",
    "                tf.nn.rnn_cell.LSTMCell(\n",
    "                    size,\n",
    "                    initializer = tf.orthogonal_initializer(),\n",
    "                    reuse = reuse,\n",
    "                ),\n",
    "                output_keep_prob = dropout,\n",
    "            )\n",
    "        \n",
    "        self.words = tf.placeholder(tf.int32, (None, None))\n",
    "        self.heads = tf.placeholder(tf.int32, (None, None))\n",
    "        self.types = tf.placeholder(tf.int32, (None, None))\n",
    "        self.mask = tf.cast(tf.math.not_equal(self.words, 0), tf.float32)\n",
    "        self.maxlen = tf.shape(self.words)[1]\n",
    "        self.lengths = tf.count_nonzero(self.words, 1)\n",
    "        mask = self.mask\n",
    "        heads = self.heads\n",
    "        types = self.types\n",
    "        \n",
    "        self.arc_h = tf.layers.Dense(hidden_size_word)\n",
    "        self.arc_c = tf.layers.Dense(hidden_size_word)\n",
    "        self.attention = BiAAttention(hidden_size_word, hidden_size_word, 1)\n",
    "\n",
    "        self.type_h = tf.layers.Dense(hidden_size_word)\n",
    "        self.type_c = tf.layers.Dense(hidden_size_word)\n",
    "        self.bilinear = BiLinear(hidden_size_word, hidden_size_word, len(tag2idx))\n",
    "        \n",
    "        model = modeling.BertModel(\n",
    "            config=bert_config,\n",
    "            is_training=True,\n",
    "            input_ids=self.words,\n",
    "            use_one_hot_embeddings=False)\n",
    "        output_layer = model.get_sequence_output()\n",
    "        \n",
    "        arc_h = tf.nn.elu(self.arc_h(output_layer))\n",
    "        arc_c = tf.nn.elu(self.arc_c(output_layer))\n",
    "        \n",
    "        type_h = tf.nn.elu(self.type_h(output_layer))\n",
    "        type_c = tf.nn.elu(self.type_c(output_layer))\n",
    "        \n",
    "        out_arc = tf.squeeze(self.attention.forward(arc_h, arc_h, mask_d=self.mask, \n",
    "                                                    mask_e=self.mask), axis = 1)\n",
    "        \n",
    "        batch = tf.shape(out_arc)[0]\n",
    "        max_len = tf.shape(out_arc)[1]\n",
    "        sec_max_len = tf.shape(out_arc)[2]\n",
    "        batch_index = tf.range(0, batch)\n",
    "        \n",
    "        decode_arc = out_arc + tf.linalg.diag(tf.fill([max_len], -np.inf))\n",
    "        minus_mask = tf.expand_dims(tf.cast(1 - mask, tf.bool), axis = 2)\n",
    "        minus_mask = tf.tile(minus_mask, [1, 1, sec_max_len])\n",
    "        decode_arc = tf.where(minus_mask, tf.fill(tf.shape(decode_arc), -np.inf), decode_arc)\n",
    "        self.heads_seq = tf.argmax(decode_arc, axis = 1)\n",
    "        \n",
    "        t = tf.cast(tf.transpose(self.heads_seq), tf.int32)\n",
    "        broadcasted = tf.broadcast_to(batch_index, tf.shape(t))\n",
    "        concatenated = tf.transpose(tf.concat([tf.expand_dims(broadcasted, axis = 0), \n",
    "                                               tf.expand_dims(t, axis = 0)], axis = 0))\n",
    "        type_h = tf.gather_nd(type_h, concatenated)\n",
    "        out_type = self.bilinear.forward(type_h, type_c)\n",
    "        self.tags_seq = tf.argmax(out_type, axis = 2)\n",
    "        \n",
    "        batch = tf.shape(out_arc)[0]\n",
    "        max_len = tf.shape(out_arc)[1]\n",
    "        batch_index = tf.range(0, batch)\n",
    "        t = tf.transpose(heads)\n",
    "        broadcasted = tf.broadcast_to(batch_index, tf.shape(t))\n",
    "        concatenated = tf.transpose(tf.concat([tf.expand_dims(broadcasted, axis = 0), \n",
    "                                               tf.expand_dims(t, axis = 0)], axis = 0))\n",
    "        type_h = tf.gather_nd(type_h, concatenated)\n",
    "        out_type = self.bilinear.forward(type_h, type_c)\n",
    "        minus_inf = -1e8\n",
    "        minus_mask = (1 - mask) * minus_inf\n",
    "        out_arc = out_arc + tf.expand_dims(minus_mask, axis = 2) + tf.expand_dims(minus_mask, axis = 1)\n",
    "        loss_arc = tf.nn.log_softmax(out_arc, dim=1)\n",
    "        loss_type = tf.nn.log_softmax(out_type, dim=2)\n",
    "        loss_arc = loss_arc * tf.expand_dims(mask, axis = 2) * tf.expand_dims(mask, axis = 1)\n",
    "        loss_type = loss_type * tf.expand_dims(mask, axis = 2)\n",
    "        num = tf.reduce_sum(mask) - tf.cast(batch, tf.float32)\n",
    "        child_index = tf.tile(tf.expand_dims(tf.range(0, max_len), 1), [1, batch])\n",
    "        t = tf.transpose(heads)\n",
    "        broadcasted = tf.broadcast_to(batch_index, tf.shape(t))\n",
    "        concatenated = tf.transpose(tf.concat([tf.expand_dims(broadcasted, axis = 0),\n",
    "                                               tf.expand_dims(t, axis = 0),\n",
    "                                               tf.expand_dims(child_index, axis = 0)], axis = 0))\n",
    "        loss_arc = tf.gather_nd(loss_arc, concatenated)\n",
    "        loss_arc = tf.transpose(loss_arc, [1, 0])\n",
    "        \n",
    "        t = tf.transpose(types)\n",
    "        broadcasted = tf.broadcast_to(batch_index, tf.shape(t))\n",
    "        concatenated = tf.transpose(tf.concat([tf.expand_dims(broadcasted, axis = 0),\n",
    "                                               tf.expand_dims(child_index, axis = 0),\n",
    "                                               tf.expand_dims(t, axis = 0)], axis = 0))\n",
    "        loss_type = tf.gather_nd(loss_type, concatenated)\n",
    "        loss_type = tf.transpose(loss_type, [1, 0])\n",
    "        self.cost = (tf.reduce_sum(-loss_arc) / num) + (tf.reduce_sum(-loss_type) / num)\n",
    "        self.optimizer = optimization.create_optimizer(self.cost, learning_rate, \n",
    "                                                       num_train_steps, num_warmup_steps, False)\n",
    "        \n",
    "        mask = tf.sequence_mask(self.lengths, maxlen = self.maxlen)\n",
    "        \n",
    "        self.prediction = tf.boolean_mask(self.tags_seq, mask)\n",
    "        mask_label = tf.boolean_mask(self.types, mask)\n",
    "        correct_pred = tf.equal(tf.cast(self.prediction, tf.int32), 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",
    "        self.prediction = tf.cast(tf.boolean_mask(self.heads_seq, mask), tf.int32)\n",
    "        mask_label = tf.boolean_mask(self.heads, mask)\n",
    "        correct_pred = tf.equal(self.prediction, mask_label)\n",
    "        correct_index = tf.cast(correct_pred, tf.float32)\n",
    "        self.accuracy_depends = tf.reduce_mean(tf.cast(correct_pred, tf.float32))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/bert/modeling.py:171: The name tf.variable_scope is deprecated. Please use tf.compat.v1.variable_scope instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/bert/modeling.py:409: The name tf.get_variable is deprecated. Please use tf.compat.v1.get_variable instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/bert/modeling.py:490: The name tf.assert_less_equal is deprecated. Please use tf.compat.v1.assert_less_equal instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/bert/modeling.py:358: calling dropout (from tensorflow.python.ops.nn_ops) with keep_prob is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use `rate` instead of `keep_prob`. Rate should be set to `rate = 1 - keep_prob`.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/husein/.local/lib/python3.6/site-packages/tensorflow/python/client/session.py:1735: UserWarning: An interactive session is already active. This can cause out-of-memory errors in some cases. You must explicitly call `InteractiveSession.close()` to release resources held by the other session(s).\n",
      "  warnings.warn('An interactive session is already active. This can '\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/bert/modeling.py:671: 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/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",
      "WARNING:tensorflow:From <ipython-input-15-5e36c0a9af4c>:225: 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",
      "WARNING:tensorflow:From <ipython-input-15-5e36c0a9af4c>:248: calling log_softmax (from tensorflow.python.ops.nn_ops) with dim is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "dim is deprecated, use axis instead\n",
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/bert/optimization.py:27: The name tf.train.get_or_create_global_step is deprecated. Please use tf.compat.v1.train.get_or_create_global_step instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/bert/optimization.py:32: The name tf.train.polynomial_decay is deprecated. Please use tf.compat.v1.train.polynomial_decay instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/tensorflow/python/keras/optimizer_v2/learning_rate_schedule.py:409: div (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Deprecated in favor of operator or tf.math.divide.\n",
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/bert/optimization.py:70: The name tf.trainable_variables is deprecated. Please use tf.compat.v1.trainable_variables instead.\n",
      "\n"
     ]
    }
   ],
   "source": [
    "tf.reset_default_graph()\n",
    "sess = tf.InteractiveSession()\n",
    "\n",
    "learning_rate = 2e-5\n",
    "hidden_size_word = 128\n",
    "\n",
    "model = Model(learning_rate, hidden_size_word)\n",
    "sess.run(tf.global_variables_initializer())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/husein/.local/lib/python3.6/site-packages/tensorflow/python/training/saver.py:1276: checkpoint_exists (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use standard file APIs to check for files with this prefix.\n",
      "INFO:tensorflow:Restoring parameters from cased_L-12_H-768_A-12/bert_model.ckpt\n"
     ]
    }
   ],
   "source": [
    "var_lists = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, scope = 'bert')\n",
    "saver = tf.train.Saver(var_list = var_lists)\n",
    "saver.restore(sess, BERT_INIT_CHKPNT)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
    "\n",
    "batch_x = train_X[:5]\n",
    "batch_x = pad_sequences(batch_x,padding='post')\n",
    "batch_y = train_Y[:5]\n",
    "batch_y = pad_sequences(batch_y,padding='post')\n",
    "batch_depends = train_depends[:5]\n",
    "batch_depends = pad_sequences(batch_depends,padding='post')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.0070422534, 0.028169014, 12.410244]"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sess.run([model.accuracy, model.accuracy_depends, model.cost],\n",
    "        feed_dict = {model.words: batch_x,\n",
    "                model.types: batch_y,\n",
    "                model.heads: batch_depends})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([ 8,  8, 48, 34, 36, 36, 27, 30, 19,  8, 34, 29, 29, 41, 28, 41, 19,\n",
       "        20, 20, 41, 47, 20, 23, 47, 28, 19, 27, 41, 18, 48, 36, 41, 27, 34,\n",
       "        36,  4, 28,  8,  8,  8,  4,  8,  8,  4]),\n",
       " array([20, 10, 16,  2,  9, 10,  0, 21,  1,  0,  2,  2,  2, 10, 10, 10, 17,\n",
       "        36, 36, 10,  2, 36, 10,  2, 10,  0,  0, 10, 36, 16, 10, 10,  0,  2,\n",
       "        10,  1, 10,  0,  0,  0,  0,  0,  0,  0]),\n",
       " array([ 0,  1,  2,  2,  0,  2,  7,  8,  2,  8,  0,  0,  9,  9,  9,  9,  0,\n",
       "         9, 16,  9, 19, 19,  8, 22, 22, 19, 24, 22,  0,  0, 22, 29, 29, 29,\n",
       "        22,  2,  0,  0,  0,  0,  0,  0,  0,  0], dtype=int32))"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tags_seq, heads = sess.run(\n",
    "    [model.tags_seq, model.heads_seq],\n",
    "    feed_dict = {\n",
    "        model.words: batch_x,\n",
    "    },\n",
    ")\n",
    "tags_seq[0], heads[0], batch_depends[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 375/375 [01:10<00:00,  5.31it/s, accuracy=0.754, accuracy_depends=0.482, cost=2.55]  \n",
      "test minibatch loop: 100%|██████████| 120/120 [00:05<00:00, 20.51it/s, accuracy=0.808, accuracy_depends=0.549, cost=2]   \n",
      "train minibatch loop:   0%|          | 1/375 [00:00<01:06,  5.64it/s, accuracy=0.746, accuracy_depends=0.383, cost=2.83]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 0, training loss: 4.682894, training acc: 0.433306, training depends: 0.308738, valid loss: 2.175135, valid acc: 0.757226, valid depends: 0.515791\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 375/375 [01:06<00:00,  5.67it/s, accuracy=0.884, accuracy_depends=0.641, cost=1.37] \n",
      "test minibatch loop: 100%|██████████| 120/120 [00:05<00:00, 20.69it/s, accuracy=0.886, accuracy_depends=0.724, cost=0.95] \n",
      "train minibatch loop:   0%|          | 1/375 [00:00<01:05,  5.72it/s, accuracy=0.848, accuracy_depends=0.53, cost=1.85]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 1, training loss: 1.797183, training acc: 0.815364, training depends: 0.561427, valid loss: 1.349600, valid acc: 0.857193, valid depends: 0.636366\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 375/375 [01:06<00:00,  5.66it/s, accuracy=0.889, accuracy_depends=0.695, cost=1.04] \n",
      "test minibatch loop: 100%|██████████| 120/120 [00:05<00:00, 20.36it/s, accuracy=0.919, accuracy_depends=0.76, cost=0.708] \n",
      "train minibatch loop:   0%|          | 1/375 [00:00<01:05,  5.68it/s, accuracy=0.877, accuracy_depends=0.61, cost=1.42]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 2, training loss: 1.193647, training acc: 0.869602, training depends: 0.653151, valid loss: 1.071987, valid acc: 0.879075, valid depends: 0.677740\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 375/375 [01:06<00:00,  5.67it/s, accuracy=0.912, accuracy_depends=0.691, cost=0.926]\n",
      "test minibatch loop: 100%|██████████| 120/120 [00:05<00:00, 20.55it/s, accuracy=0.919, accuracy_depends=0.779, cost=0.63] \n",
      "train minibatch loop:   0%|          | 1/375 [00:00<01:05,  5.68it/s, accuracy=0.893, accuracy_depends=0.627, cost=1.16]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 3, training loss: 0.931854, training acc: 0.892288, training depends: 0.696346, valid loss: 1.005326, valid acc: 0.883707, valid depends: 0.692016\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 375/375 [01:06<00:00,  5.67it/s, accuracy=0.914, accuracy_depends=0.739, cost=0.762]\n",
      "test minibatch loop: 100%|██████████| 120/120 [00:05<00:00, 20.45it/s, accuracy=0.912, accuracy_depends=0.799, cost=0.51] \n",
      "train minibatch loop:   0%|          | 1/375 [00:00<01:06,  5.66it/s, accuracy=0.889, accuracy_depends=0.654, cost=1.01]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 4, training loss: 0.777697, training acc: 0.901257, training depends: 0.721131, valid loss: 0.964560, valid acc: 0.877505, valid depends: 0.701398\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 375/375 [01:06<00:00,  5.67it/s, accuracy=0.913, accuracy_depends=0.755, cost=0.638]\n",
      "test minibatch loop: 100%|██████████| 120/120 [00:05<00:00, 20.68it/s, accuracy=0.912, accuracy_depends=0.812, cost=0.492]\n",
      "train minibatch loop:   0%|          | 1/375 [00:00<01:04,  5.79it/s, accuracy=0.893, accuracy_depends=0.659, cost=0.932]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 5, training loss: 0.668966, training acc: 0.901568, training depends: 0.741328, valid loss: 0.928792, valid acc: 0.891125, valid depends: 0.710297\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 375/375 [01:06<00:00,  5.67it/s, accuracy=0.913, accuracy_depends=0.751, cost=0.616]\n",
      "test minibatch loop: 100%|██████████| 120/120 [00:05<00:00, 20.45it/s, accuracy=0.919, accuracy_depends=0.825, cost=0.394]\n",
      "train minibatch loop:   0%|          | 1/375 [00:00<01:06,  5.67it/s, accuracy=0.896, accuracy_depends=0.707, cost=0.809]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 6, training loss: 0.594000, training acc: 0.913134, training depends: 0.754423, valid loss: 0.943845, valid acc: 0.888382, valid depends: 0.713479\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 375/375 [01:06<00:00,  5.67it/s, accuracy=0.927, accuracy_depends=0.776, cost=0.537] \n",
      "test minibatch loop: 100%|██████████| 120/120 [00:05<00:00, 20.37it/s, accuracy=0.935, accuracy_depends=0.808, cost=0.534]\n",
      "train minibatch loop:   0%|          | 1/375 [00:00<01:06,  5.61it/s, accuracy=0.909, accuracy_depends=0.709, cost=0.77]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 7, training loss: 0.538314, training acc: 0.920553, training depends: 0.764744, valid loss: 0.930650, valid acc: 0.903622, valid depends: 0.718959\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 375/375 [01:06<00:00,  5.66it/s, accuracy=0.935, accuracy_depends=0.781, cost=0.505]\n",
      "test minibatch loop: 100%|██████████| 120/120 [00:05<00:00, 20.38it/s, accuracy=0.938, accuracy_depends=0.821, cost=0.457]\n",
      "train minibatch loop:   0%|          | 1/375 [00:00<01:07,  5.53it/s, accuracy=0.915, accuracy_depends=0.711, cost=0.767]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 8, training loss: 0.486278, training acc: 0.927081, training depends: 0.774812, valid loss: 0.932128, valid acc: 0.904604, valid depends: 0.722158\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 375/375 [01:06<00:00,  5.67it/s, accuracy=0.925, accuracy_depends=0.787, cost=0.485] \n",
      "test minibatch loop: 100%|██████████| 120/120 [00:05<00:00, 20.50it/s, accuracy=0.958, accuracy_depends=0.825, cost=0.524]\n",
      "train minibatch loop:   0%|          | 1/375 [00:00<01:06,  5.61it/s, accuracy=0.924, accuracy_depends=0.735, cost=0.633]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 9, training loss: 0.447538, training acc: 0.931575, training depends: 0.781835, valid loss: 0.943356, valid acc: 0.905484, valid depends: 0.722892\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 375/375 [01:06<00:00,  5.66it/s, accuracy=0.942, accuracy_depends=0.806, cost=0.424] \n",
      "test minibatch loop: 100%|██████████| 120/120 [00:05<00:00, 20.55it/s, accuracy=0.935, accuracy_depends=0.815, cost=0.496]\n",
      "train minibatch loop:   0%|          | 1/375 [00:00<01:06,  5.67it/s, accuracy=0.896, accuracy_depends=0.748, cost=0.611]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 10, training loss: 0.413205, training acc: 0.932623, training depends: 0.789132, valid loss: 0.954858, valid acc: 0.903540, valid depends: 0.724419\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 375/375 [01:06<00:00,  5.66it/s, accuracy=0.943, accuracy_depends=0.788, cost=0.442] \n",
      "test minibatch loop: 100%|██████████| 120/120 [00:05<00:00, 20.28it/s, accuracy=0.945, accuracy_depends=0.795, cost=0.602]\n",
      "train minibatch loop:   0%|          | 1/375 [00:00<01:05,  5.69it/s, accuracy=0.92, accuracy_depends=0.761, cost=0.558]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 11, training loss: 0.389162, training acc: 0.934991, training depends: 0.793624, valid loss: 0.962155, valid acc: 0.910515, valid depends: 0.726305\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 375/375 [01:06<00:00,  5.65it/s, accuracy=0.943, accuracy_depends=0.806, cost=0.433] \n",
      "test minibatch loop: 100%|██████████| 120/120 [00:05<00:00, 20.37it/s, accuracy=0.942, accuracy_depends=0.828, cost=0.454]\n",
      "train minibatch loop:   0%|          | 1/375 [00:00<01:06,  5.67it/s, accuracy=0.919, accuracy_depends=0.759, cost=0.538]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 12, training loss: 0.368160, training acc: 0.940245, training depends: 0.797881, valid loss: 0.978189, valid acc: 0.906123, valid depends: 0.726453\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 375/375 [01:06<00:00,  5.66it/s, accuracy=0.942, accuracy_depends=0.807, cost=0.404] \n",
      "test minibatch loop: 100%|██████████| 120/120 [00:05<00:00, 20.48it/s, accuracy=0.951, accuracy_depends=0.844, cost=0.43] \n",
      "train minibatch loop:   0%|          | 1/375 [00:00<01:06,  5.63it/s, accuracy=0.934, accuracy_depends=0.759, cost=0.563]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 13, training loss: 0.356211, training acc: 0.941396, training depends: 0.800658, valid loss: 0.964498, valid acc: 0.910670, valid depends: 0.727217\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 375/375 [01:06<00:00,  5.67it/s, accuracy=0.943, accuracy_depends=0.814, cost=0.378] \n",
      "test minibatch loop: 100%|██████████| 120/120 [00:05<00:00, 20.28it/s, accuracy=0.945, accuracy_depends=0.805, cost=0.468]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 14, training loss: 0.346428, training acc: 0.943538, training depends: 0.802292, valid loss: 0.971327, valid acc: 0.908659, valid depends: 0.727845\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "from tqdm import tqdm\n",
    "\n",
    "batch_size = 32\n",
    "epoch = 15\n",
    "\n",
    "for e in range(epoch):\n",
    "    train_acc, train_loss = [], []\n",
    "    test_acc, test_loss = [], []\n",
    "    train_acc_depends, test_acc_depends = [], []\n",
    "    \n",
    "    pbar = tqdm(\n",
    "        range(0, len(train_X), batch_size), desc = 'train minibatch loop'\n",
    "    )\n",
    "    for i in pbar:\n",
    "        index = min(i + batch_size, len(train_X))\n",
    "        batch_x = train_X[i: index]\n",
    "        batch_x = pad_sequences(batch_x,padding='post')\n",
    "        batch_y = train_Y[i: index]\n",
    "        batch_y = pad_sequences(batch_y,padding='post')\n",
    "        batch_depends = train_depends[i: index]\n",
    "        batch_depends = pad_sequences(batch_depends,padding='post')\n",
    "        \n",
    "        acc_depends, acc, cost, _ = sess.run(\n",
    "            [model.accuracy_depends, model.accuracy, model.cost, model.optimizer],\n",
    "            feed_dict = {\n",
    "                model.words: batch_x,\n",
    "                model.types: batch_y,\n",
    "                model.heads: batch_depends\n",
    "            },\n",
    "        )\n",
    "        train_loss.append(cost)\n",
    "        train_acc.append(acc)\n",
    "        train_acc_depends.append(acc_depends)\n",
    "        pbar.set_postfix(cost = cost, accuracy = acc, accuracy_depends = acc_depends)\n",
    "        \n",
    "    pbar = tqdm(\n",
    "        range(0, len(test_X), batch_size), desc = 'test minibatch loop'\n",
    "    )\n",
    "    for i in pbar:\n",
    "        index = min(i + batch_size, len(test_X))\n",
    "        batch_x = test_X[i: index]\n",
    "        batch_x = pad_sequences(batch_x,padding='post')\n",
    "        batch_y = test_Y[i: index]\n",
    "        batch_y = pad_sequences(batch_y,padding='post')\n",
    "        batch_depends = test_depends[i: index]\n",
    "        batch_depends = pad_sequences(batch_depends,padding='post')\n",
    "        \n",
    "        acc_depends, acc, cost = sess.run(\n",
    "            [model.accuracy_depends, model.accuracy, model.cost],\n",
    "            feed_dict = {\n",
    "                model.words: batch_x,\n",
    "                model.types: batch_y,\n",
    "                model.heads: batch_depends\n",
    "            },\n",
    "        )\n",
    "        test_loss.append(cost)\n",
    "        test_acc.append(acc)\n",
    "        test_acc_depends.append(acc_depends)\n",
    "        pbar.set_postfix(cost = cost, accuracy = acc, accuracy_depends = acc_depends)\n",
    "    \n",
    "    \n",
    "    print(\n",
    "    'epoch: %d, training loss: %f, training acc: %f, training depends: %f, valid loss: %f, valid acc: %f, valid depends: %f\\n'\n",
    "    % (e, np.mean(train_loss), \n",
    "       np.mean(train_acc), \n",
    "       np.mean(train_acc_depends), \n",
    "       np.mean(test_loss), \n",
    "       np.mean(test_acc), \n",
    "       np.mean(test_acc_depends)\n",
    "    ))\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([ 0, 40,  6, 22, 26, 23, 18, 16,  1,  1,  5,  3, 13, 10, 11,  6, 12,\n",
       "        13, 10, 16,  7,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,\n",
       "         0,  0,  0,  0,  0,  0,  0,  0]),\n",
       " array([ 3,  2,  8,  5,  5,  2,  8,  8, -1, -1,  0, 11, 10,  8, 14, 13,  8,\n",
       "        15, 14, 14,  8, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,\n",
       "        -1, -1, -1, -1, -1, -1, -1, -1]),\n",
       " array([-1,  2,  8,  5,  5,  2,  8,  8, -1, -1,  0, 11, 11,  8, 14, 14,  8,\n",
       "        16, 14, 14,  8, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,\n",
       "        -1, -1, -1, -1, -1, -1, -1, -1], dtype=int32))"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tags_seq, heads = sess.run(\n",
    "    [model.tags_seq, model.heads_seq],\n",
    "    feed_dict = {\n",
    "        model.words: batch_x,\n",
    "    },\n",
    ")\n",
    "tags_seq[0], heads[0] - 1, batch_depends[0] - 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "def evaluate(heads_pred, types_pred, heads, types, lengths,\n",
    "             symbolic_root=False, symbolic_end=False):\n",
    "    batch_size, _ = heads_pred.shape\n",
    "    ucorr = 0.\n",
    "    lcorr = 0.\n",
    "    total = 0.\n",
    "    ucomplete_match = 0.\n",
    "    lcomplete_match = 0.\n",
    "\n",
    "    corr_root = 0.\n",
    "    total_root = 0.\n",
    "    start = 1 if symbolic_root else 0\n",
    "    end = 1 if symbolic_end else 0\n",
    "    for i in range(batch_size):\n",
    "        ucm = 1.\n",
    "        lcm = 1.\n",
    "        for j in range(start, lengths[i] - end):\n",
    "\n",
    "            total += 1\n",
    "            if heads[i, j] == heads_pred[i, j]:\n",
    "                ucorr += 1\n",
    "                if types[i, j] == types_pred[i, j]:\n",
    "                    lcorr += 1\n",
    "                else:\n",
    "                    lcm = 0\n",
    "            else:\n",
    "                ucm = 0\n",
    "                lcm = 0\n",
    "\n",
    "            if heads[i, j] == 0:\n",
    "                total_root += 1\n",
    "                corr_root += 1 if heads_pred[i, j] == 0 else 0\n",
    "\n",
    "        ucomplete_match += ucm\n",
    "        lcomplete_match += lcm\n",
    "    \n",
    "    return ucorr / total, lcorr / total, corr_root / total_root"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "arcs, types, roots = [], [], []\n",
    "\n",
    "for i in range(0, len(test_X), batch_size):\n",
    "    index = min(i + batch_size, len(test_X))\n",
    "    batch_x = test_X[i: index]\n",
    "    batch_x = pad_sequences(batch_x,padding='post')\n",
    "    batch_y = test_Y[i: index]\n",
    "    batch_y = pad_sequences(batch_y,padding='post')\n",
    "    batch_depends = test_depends[i: index]\n",
    "    batch_depends = pad_sequences(batch_depends,padding='post')\n",
    "    \n",
    "    tags_seq, heads = sess.run(\n",
    "        [model.tags_seq, model.heads_seq],\n",
    "        feed_dict = {\n",
    "            model.words: batch_x,\n",
    "        },\n",
    "    )\n",
    "    \n",
    "    arc_accuracy, type_accuracy, root_accuracy = evaluate(heads - 1, tags_seq, batch_depends - 1, batch_y, \n",
    "            np.count_nonzero(batch_x, axis = 1))\n",
    "    arcs.append(arc_accuracy)\n",
    "    types.append(type_accuracy)\n",
    "    roots.append(root_accuracy)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "arc accuracy: 0.728543873570515\n",
      "types accuracy: 0.6711201611430444\n",
      "root accuracy: 0.7393229166666667\n"
     ]
    }
   ],
   "source": [
    "print('arc accuracy:', np.mean(arcs))\n",
    "print('types accuracy:', np.mean(types))\n",
    "print('root accuracy:', np.mean(roots))"
   ]
  },
  {
   "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
}