entity-tagging/5.residual-char-ngrams.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from bs4 import BeautifulSoup\n",
    "import re\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from tqdm import tqdm\n",
    "import itertools"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def process_string(string):\n",
    "    string = re.sub('[^A-Za-z0-9\\-\\/ ]+', ' ', string).split()\n",
    "    return [y.strip() for y in string]\n",
    "\n",
    "def to_title(string):\n",
    "    if string.isupper():\n",
    "        string = string.title()\n",
    "    return string"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def parse_raw(filename):\n",
    "    with open(filename, 'r') as fopen:\n",
    "        entities = fopen.read()\n",
    "    soup = BeautifulSoup(entities, 'html.parser')\n",
    "    inside_tag = ''\n",
    "    texts, labels = [], []\n",
    "    for sentence in soup.prettify().split('\\n'):\n",
    "        if len(inside_tag):\n",
    "            splitted = process_string(sentence)\n",
    "            texts += splitted\n",
    "            labels += [inside_tag] * len(splitted)\n",
    "            inside_tag = ''\n",
    "        else:\n",
    "            if not sentence.find('</'):\n",
    "                pass\n",
    "            elif not sentence.find('<'):\n",
    "                inside_tag = sentence.split('>')[0][1:]\n",
    "            else:\n",
    "                splitted = process_string(sentence)\n",
    "                texts += splitted\n",
    "                labels += ['OTHER'] * len(splitted)\n",
    "    assert (len(texts)==len(labels)), \"length texts and labels are not same\"\n",
    "    print('len texts and labels: ', len(texts))\n",
    "    return texts,labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "len texts and labels:  34012\n",
      "len texts and labels:  9249\n"
     ]
    }
   ],
   "source": [
    "train_texts, train_labels = parse_raw('data_train.txt')\n",
    "test_texts, test_labels = parse_raw('data_test.txt')\n",
    "train_texts += test_texts\n",
    "train_labels += test_labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open('entities-bm-normalize-v3.txt','r') as fopen:\n",
    "    entities_bm = fopen.read().split('\\n')[:-1]\n",
    "entities_bm = [i.split() for i in entities_bm]\n",
    "entities_bm = [[i[0],'TIME' if i[0] in 'jam' else i[1]] for i in entities_bm]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "replace_by = {'organizaiton':'organization','orgnization':'organization',\n",
    "             'othoer': 'OTHER'}\n",
    "\n",
    "with open('NER-part1.txt','r') as fopen:\n",
    "    nexts = fopen.read().split('\\n')[:-1]\n",
    "nexts = [i.split() for i in nexts]\n",
    "for i in nexts:\n",
    "    if len(i) == 2:\n",
    "        label = i[1].lower()\n",
    "        if 'other' in label:\n",
    "            label = label.upper()\n",
    "        if label in replace_by:\n",
    "            label = replace_by[label]\n",
    "        train_labels.append(label)\n",
    "        train_texts.append(i[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "'KN'\n",
      "'KA'\n"
     ]
    }
   ],
   "source": [
    "replace_by = {'LOC':'location','PRN':'person','NORP':'organization','ORG':'organization','LAW':'law',\n",
    "             'EVENT':'event','FAC':'organization','TIME':'time','O':'OTHER','ART':'person','DOC':'law'}\n",
    "for i in entities_bm:\n",
    "    try:\n",
    "        string = process_string(i[0])\n",
    "        if len(string):\n",
    "            train_labels.append(replace_by[i[1]])\n",
    "            train_texts.append(process_string(i[0])[0])  \n",
    "    except Exception as e:\n",
    "        print(e)\n",
    "        \n",
    "assert (len(train_texts)==len(train_labels)), \"length texts and labels are not same\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array(['OTHER', 'event', 'law', 'location', 'organization', 'person',\n",
       "        'quantity', 'time'], dtype='<U12'),\n",
       " array([49712,   234,   185,  2056,  2596,  4397,  1341,  1296]))"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.unique(train_labels,return_counts=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "def _pad_sequence(\n",
    "    sequence,\n",
    "    n,\n",
    "    pad_left = False,\n",
    "    pad_right = False,\n",
    "    left_pad_symbol = None,\n",
    "    right_pad_symbol = None,\n",
    "):\n",
    "    sequence = iter(sequence)\n",
    "    if pad_left:\n",
    "        sequence = itertools.chain((left_pad_symbol,) * (n - 1), sequence)\n",
    "    if pad_right:\n",
    "        sequence = itertools.chain(sequence, (right_pad_symbol,) * (n - 1))\n",
    "    return sequence\n",
    "\n",
    "\n",
    "def ngrams(\n",
    "    sequence,\n",
    "    n,\n",
    "    pad_left = False,\n",
    "    pad_right = False,\n",
    "    left_pad_symbol = None,\n",
    "    right_pad_symbol = None,\n",
    "):\n",
    "    \"\"\"\n",
    "    generate ngrams\n",
    "\n",
    "    Parameters\n",
    "    ----------\n",
    "    sequence : list of str\n",
    "        list of tokenize words\n",
    "    n : int\n",
    "        ngram size\n",
    "\n",
    "    Returns\n",
    "    -------\n",
    "    ngram: list\n",
    "    \"\"\"\n",
    "    sequence = _pad_sequence(\n",
    "        sequence, n, pad_left, pad_right, left_pad_symbol, right_pad_symbol\n",
    "    )\n",
    "\n",
    "    history = []\n",
    "    while n > 1:\n",
    "        try:\n",
    "            next_item = next(sequence)\n",
    "        except StopIteration:\n",
    "            return\n",
    "        history.append(next_item)\n",
    "        n -= 1\n",
    "    for item in sequence:\n",
    "        history.append(item)\n",
    "        yield tuple(history)\n",
    "        del history[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_ngrams(s, grams=(2,3,4)):\n",
    "    return [''.join(i) for k in grams for i in list(ngrams(s,k))]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "word2idx = {'PAD': 0,'NUM':1,'UNK':2}\n",
    "tag2idx = {'PAD': 0}\n",
    "char2idx = {'PAD': 0,'NUM':1,'UNK':2}\n",
    "word_idx = 3\n",
    "tag_idx = 1\n",
    "char_idx = 3\n",
    "\n",
    "def parse_XY(texts, labels):\n",
    "    global word2idx, tag2idx, char2idx, word_idx, tag_idx, char_idx\n",
    "    X, Y = [], []\n",
    "    for no, text in enumerate(texts):\n",
    "        text = text.lower()\n",
    "        if len(text) < 2:\n",
    "            continue\n",
    "        tag = labels[no]\n",
    "        for c in get_ngrams(text):\n",
    "            if c not in char2idx:\n",
    "                char2idx[c] = char_idx\n",
    "                char_idx += 1\n",
    "        if tag not in tag2idx:\n",
    "            tag2idx[tag] = tag_idx\n",
    "            tag_idx += 1\n",
    "        Y.append(tag2idx[tag])\n",
    "        if text not in word2idx:\n",
    "            word2idx[text] = word_idx\n",
    "            word_idx += 1\n",
    "        X.append(word2idx[text])\n",
    "    return X, np.array(Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "X, Y = parse_XY(train_texts, train_labels)\n",
    "idx2word={idx: tag for tag, idx in word2idx.items()}\n",
    "idx2tag = {i: w for w, i in tag2idx.items()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "seq_len = 50\n",
    "def iter_seq(x):\n",
    "    return np.array([x[i: i+seq_len] for i in range(0, len(x)-seq_len, 1)])\n",
    "\n",
    "def to_train_seq(*args):\n",
    "    return [iter_seq(x) for x in args]\n",
    "\n",
    "def generate_char_seq(batch, maxlen):\n",
    "    temp = np.zeros((batch.shape[0],batch.shape[1], maxlen),dtype=np.int32)\n",
    "    for i in range(batch.shape[0]):\n",
    "        for k in range(batch.shape[1]):\n",
    "            for no, c in enumerate(get_ngrams(idx2word[batch[i,k]])):\n",
    "                temp[i,k,no] = char2idx[c]\n",
    "    return temp"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(61353, 50)"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_seq, Y_seq = to_train_seq(X, Y)\n",
    "X_char_seq = generate_char_seq(X_seq, seq_len * 2)\n",
    "X_seq.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(61353, 50)"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Y_seq.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/dist-packages/sklearn/cross_validation.py:41: DeprecationWarning: This module was deprecated in version 0.18 in favor of the model_selection module into which all the refactored classes and functions are moved. Also note that the interface of the new CV iterators are different from that of this module. This module will be removed in 0.20.\n",
      "  \"This module will be removed in 0.20.\", DeprecationWarning)\n"
     ]
    }
   ],
   "source": [
    "from sklearn.cross_validation import train_test_split\n",
    "train_Y, test_Y, train_X, test_X = train_test_split(Y_seq, X_char_seq,test_size=0.2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Attention:\n",
    "    def __init__(self,hidden_size):\n",
    "        self.hidden_size = hidden_size\n",
    "        self.dense_layer = tf.layers.Dense(hidden_size)\n",
    "        self.v = tf.random_normal([hidden_size],mean=0,stddev=1/np.sqrt(hidden_size))\n",
    "        \n",
    "    def score(self, hidden_tensor, encoder_outputs):\n",
    "        energy = tf.nn.tanh(self.dense_layer(tf.concat([hidden_tensor,encoder_outputs],2)))\n",
    "        energy = tf.transpose(energy,[0,2,1])\n",
    "        batch_size = tf.shape(encoder_outputs)[0]\n",
    "        v = tf.expand_dims(tf.tile(tf.expand_dims(self.v,0),[batch_size,1]),1)\n",
    "        energy = tf.matmul(v,energy)\n",
    "        return tf.squeeze(energy,1)\n",
    "    \n",
    "    def __call__(self, hidden, encoder_outputs):\n",
    "        seq_len = tf.shape(encoder_outputs)[1]\n",
    "        batch_size = tf.shape(encoder_outputs)[0]\n",
    "        H = tf.tile(tf.expand_dims(hidden, 1),[1,seq_len,1])\n",
    "        attn_energies = self.score(H,encoder_outputs)\n",
    "        return tf.expand_dims(tf.nn.softmax(attn_energies),1)\n",
    "\n",
    "class Model:\n",
    "    def __init__(\n",
    "        self,\n",
    "        dict_size,\n",
    "        size_layers,\n",
    "        learning_rate,\n",
    "        maxlen,\n",
    "        num_blocks = 3,\n",
    "        block_size = 128,\n",
    "    ):\n",
    "        self.word_ids = tf.placeholder(tf.int32, shape = [None, maxlen, maxlen * 2])\n",
    "        self.labels = tf.placeholder(tf.int32, shape = [None, maxlen])\n",
    "        embeddings = tf.Variable(tf.random_uniform([dict_size, size_layers], -1, 1))\n",
    "        embedded = tf.nn.embedding_lookup(embeddings, self.word_ids)\n",
    "        embedded = tf.reduce_mean(embedded,axis=2)\n",
    "        self.attention = Attention(size_layers)\n",
    "        self.maxlen = tf.shape(self.word_ids)[1]\n",
    "        self.lengths = tf.count_nonzero(tf.reduce_sum(self.word_ids,axis=2), 1)\n",
    "\n",
    "        def residual_block(x, size, rate, block):\n",
    "            with tf.variable_scope(\n",
    "                'block_%d_%d' % (block, rate), reuse = False\n",
    "            ):\n",
    "                attn_weights = self.attention(tf.reduce_sum(x,axis=1), x)\n",
    "                conv_filter = tf.layers.conv1d(\n",
    "                    attn_weights,\n",
    "                    x.shape[2] // 4,\n",
    "                    kernel_size = size,\n",
    "                    strides = 1,\n",
    "                    padding = 'same',\n",
    "                    dilation_rate = rate,\n",
    "                    activation = tf.nn.tanh,\n",
    "                )\n",
    "                conv_gate = tf.layers.conv1d(\n",
    "                    x,\n",
    "                    x.shape[2] // 4,\n",
    "                    kernel_size = size,\n",
    "                    strides = 1,\n",
    "                    padding = 'same',\n",
    "                    dilation_rate = rate,\n",
    "                    activation = tf.nn.sigmoid,\n",
    "                )\n",
    "                out = tf.multiply(conv_filter, conv_gate)\n",
    "                out = tf.layers.conv1d(\n",
    "                    out,\n",
    "                    block_size,\n",
    "                    kernel_size = 1,\n",
    "                    strides = 1,\n",
    "                    padding = 'same',\n",
    "                    activation = tf.nn.tanh,\n",
    "                )\n",
    "                return tf.add(x, out), out\n",
    "\n",
    "        forward = tf.layers.conv1d(\n",
    "            embedded, block_size, kernel_size = 1, strides = 1, padding = 'SAME'\n",
    "        )\n",
    "        zeros = tf.zeros_like(forward)\n",
    "        for i in range(num_blocks):\n",
    "            for r in [1, 2, 4, 8, 16]:\n",
    "                forward, s = residual_block(\n",
    "                    forward, size = 7, rate = r, block = i\n",
    "                )\n",
    "                zeros = tf.add(zeros, s)\n",
    "        logits = tf.layers.conv1d(\n",
    "            zeros, len(idx2tag), kernel_size = 1, strides = 1, padding = 'SAME'\n",
    "        )\n",
    "        log_likelihood, transition_params = tf.contrib.crf.crf_log_likelihood(\n",
    "            logits, self.labels, self.lengths\n",
    "        )\n",
    "        self.cost = tf.reduce_mean(-log_likelihood)\n",
    "        self.optimizer = tf.train.AdamOptimizer(\n",
    "            learning_rate = learning_rate\n",
    "        ).minimize(self.cost)\n",
    "        \n",
    "        mask = tf.sequence_mask(self.lengths, maxlen = self.maxlen)\n",
    "        \n",
    "        self.tags_seq, _ = tf.contrib.crf.crf_decode(\n",
    "            logits, transition_params, self.lengths\n",
    "        )\n",
    "\n",
    "        self.prediction = tf.boolean_mask(self.tags_seq, mask)\n",
    "        mask_label = tf.boolean_mask(self.labels, mask)\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": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/gradients_impl.py:100: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.\n",
      "  \"Converting sparse IndexedSlices to a dense Tensor of unknown shape. \"\n"
     ]
    }
   ],
   "source": [
    "tf.reset_default_graph()\n",
    "sess = tf.InteractiveSession()\n",
    "\n",
    "dim = 256\n",
    "dropout = 1\n",
    "learning_rate = 1e-3\n",
    "batch_size = 32\n",
    "\n",
    "model = Model(len(char2idx), dim, learning_rate, seq_len)\n",
    "sess.run(tf.global_variables_initializer())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 1534/1534 [03:02<00:00,  7.28it/s, accuracy=0.899, cost=12.6]\n",
      "test minibatch loop: 100%|██████████| 384/384 [00:17<00:00, 21.86it/s, accuracy=0.887, cost=10.8]\n",
      "train minibatch loop:   0%|          | 1/1534 [00:00<03:04,  8.32it/s, accuracy=0.914, cost=11.4]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "time taken: 199.92683672904968\n",
      "epoch: 0, training loss: 30.030219, training acc: 0.823603, valid loss: 11.642038, valid acc: 0.907485\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "train minibatch loop: 100%|██████████| 1534/1534 [02:59<00:00,  8.68it/s, accuracy=0.985, cost=1.88]\n",
      "test minibatch loop: 100%|██████████| 384/384 [00:17<00:00, 22.36it/s, accuracy=0.981, cost=2.21]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "time taken: 197.2810423374176\n",
      "epoch: 1, training loss: 6.219392, training acc: 0.951056, valid loss: 2.520342, valid acc: 0.982579\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "import time\n",
    "\n",
    "for e in range(2):\n",
    "    lasttime = time.time()\n",
    "    train_acc, train_loss, test_acc, test_loss = 0, 0, 0, 0\n",
    "    pbar = tqdm(\n",
    "        range(0, len(train_X), batch_size), desc = 'train minibatch loop'\n",
    "    )\n",
    "    for i in pbar:\n",
    "        batch_x = train_X[i : min(i + batch_size, train_X.shape[0])]\n",
    "        batch_y = train_Y[i : min(i + batch_size, train_X.shape[0])]\n",
    "        acc, cost, _ = sess.run(\n",
    "            [model.accuracy, model.cost, model.optimizer],\n",
    "            feed_dict = {\n",
    "                model.word_ids: batch_x,\n",
    "                model.labels: batch_y\n",
    "            },\n",
    "        )\n",
    "        assert not np.isnan(cost)\n",
    "        train_loss += cost\n",
    "        train_acc += acc\n",
    "        pbar.set_postfix(cost = cost, accuracy = acc)\n",
    "    \n",
    "    pbar = tqdm(\n",
    "        range(0, len(test_X), batch_size), desc = 'test minibatch loop'\n",
    "    )\n",
    "    for i in pbar:\n",
    "        batch_x = test_X[i : min(i + batch_size, test_X.shape[0])]\n",
    "        batch_y = test_Y[i : min(i + batch_size, test_X.shape[0])]\n",
    "        acc, cost = sess.run(\n",
    "            [model.accuracy, model.cost],\n",
    "            feed_dict = {\n",
    "                model.word_ids: batch_x,\n",
    "                model.labels: batch_y\n",
    "            },\n",
    "        )\n",
    "        assert not np.isnan(cost)\n",
    "        test_loss += cost\n",
    "        test_acc += acc\n",
    "        pbar.set_postfix(cost = cost, accuracy = acc)\n",
    "    \n",
    "    train_loss /= len(train_X) / batch_size\n",
    "    train_acc /= len(train_X) / batch_size\n",
    "    test_loss /= len(test_X) / batch_size\n",
    "    test_acc /= len(test_X) / batch_size\n",
    "\n",
    "    print('time taken:', time.time() - lasttime)\n",
    "    print(\n",
    "        'epoch: %d, training loss: %f, training acc: %f, valid loss: %f, valid acc: %f\\n'\n",
    "        % (e, train_loss, train_acc, test_loss, test_acc)\n",
    "    )"
   ]
  },
  {
   "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
}