structured/main.py at master · elisaF/structured · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
from __future__ import division
from data_structure import DataSet
from predictor import InMemoryClient
import tensorflow as tf
import numpy as np
np.set_printoptions(threshold='nan')
import cPickle
import logging
from models import StructureModel
import subprocess
import tqdm
import time
import utils
from tensorflow.python import debug as tf_debug


def load_data(config):
    train, dev, test, embeddings, vocab = cPickle.load(open(config.data_file))
    trainset, devset, testset = DataSet(train), DataSet(dev), DataSet(test)
    vocab = dict([(v.index,k) for k,v in vocab.items()])
    trainset.sort()
    train_batches = trainset.get_batches(config.batch_size, config.epochs, rand=True)
    dev_batches = devset.get_batches(config.batch_size, 1, rand=False)
    test_batches = testset.get_batches(config.batch_size, 1, rand=False)
    dev_batches = [i for i in dev_batches]
    test_batches = [i for i in test_batches]
    return len(train), train_batches, dev_batches, test_batches, embeddings, vocab


def evaluate(sess, model, test_batches, logger):
    corr_count, all_count = 0, 0
    for ct, batch in test_batches:
        feed_dict = model.get_feed_dict(batch)
        feed_dict[model.t_variables['keep_prob']] = 1.0
        predictions = sess.run(model.final_output, feed_dict=feed_dict)
        predictions = np.argmax(predictions, 1)
        corr_count += np.sum(predictions == feed_dict[model.t_variables['gold_labels']])
        all_count += len(batch)
    acc_test = 1.0 * corr_count / all_count
    return acc_test


def run(config):
    logger = logging.getLogger()
    logger.setLevel(logging.DEBUG)
    time_log = str(time.time())
    ah = logging.FileHandler(time_log + '.log')
    ah.setLevel(logging.DEBUG)
    formatter = logging.Formatter('%(asctime)s - %(message)s')
    ah.setFormatter(formatter)
    logger.addHandler(ah)
    tf.set_random_seed(config.seed)
    initializer = utils.Initializer(config.init_seed)
    xavier_init = initializer.xavier_init()

    if config.model_dir:
        print(config)
        print("Tensorflow version: ", tf.__version__)
        print("Git version: ", get_git_revision_hash())
        logger.critical(str(config))
        logger.critical(get_git_revision_hash())

        evaluate_pretrained_model(config, logger)

    else:
        logger.debug("Going to load data")
        num_examples, train_batches, dev_batches, test_batches, embedding_matrix, vocab = load_data(config)
        logger.debug("Finished loading data.")
        # save vocab to file
        utils.save_dict(vocab, time_log +'.dict')
        print("Embedding matrix size: ", embedding_matrix.shape)
        config.n_embed, config.d_embed = embedding_matrix.shape

        config.dim_hidden = config.dim_sem + config.dim_str

        print(config)
        logger.critical(str(config))
        print("Tensorflow version: ", tf.__version__)
        print("Git version: ", get_git_revision_hash())
        logger.critical(get_git_revision_hash())

        model = StructureModel(config, xavier_init)
        model.build()
        model.get_loss()

        num_batches_per_epoch = int(num_examples / config.batch_size)
        num_steps = config.epochs * num_batches_per_epoch
        best_acc_dev = 0.0

        #with tf.Session(config=tf.ConfigProto(inter_op_parallelism_threads=1, intra_op_parallelism_threads=1)) as sess:
        with tf.Session() as sess:
            gvi = tf.global_variables_initializer()
            sess.run(gvi)
            sess.run(model.embeddings.assign(embedding_matrix.astype(np.float64)))
            if config.debug:
                sess = tf_debug.LocalCLIDebugWrapperSession(sess)
            loss = 0
            for ct, batch in tqdm.tqdm(train_batches, total=num_steps):
                feed_dict = model.get_feed_dict(batch)
                outputs, _, _loss = sess.run([model.final_output, model.opt, model.loss], feed_dict=feed_dict)
                loss+=_loss
                if(ct%config.log_period==0):
                    acc_test = evaluate(sess, model, test_batches, logger)
                    acc_dev = evaluate(sess, model, dev_batches, logger)
                    print('\nStep: {} Loss: {}'.format(ct, loss))
                    print('Test ACC: {}'.format(acc_test))
                    print('Dev  ACC: %s (%s)', acc_dev, best_acc_dev)
                    logger.debug('\nStep: {} Loss: {}'.format(ct, loss))
                    logger.debug('Test ACC: {}'.format(acc_test))
                    logger.debug('Dev  ACC: %s (%s)', acc_dev, best_acc_dev)
                    logger.handlers[0].flush()
                    loss = 0
                    if acc_dev > best_acc_dev:
                        best_acc_dev = acc_dev
                        save_model(sess, ct, model, logger, config)


def save_model(sess, step, model, logger, config):
    export_path = config.model_dir_prefix + "-" + str(step)
    print('Exporting trained model to %s' % export_path)
    logger.info('Exporting trained model to %s' % export_path)
    builder = tf.saved_model.builder.SavedModelBuilder(export_path)
    input_token_idxs = tf.saved_model.utils.build_tensor_info(model.t_variables['token_idxs'])
    input_sent_l = tf.saved_model.utils.build_tensor_info(model.t_variables['sent_l'])
    input_mask_tokens = tf.saved_model.utils.build_tensor_info(model.t_variables['mask_tokens'])
    input_mask_sents = tf.saved_model.utils.build_tensor_info(model.t_variables['mask_sents'])
    input_doc_l = tf.saved_model.utils.build_tensor_info(model.t_variables['doc_l'])
    input_gold_labels = tf.saved_model.utils.build_tensor_info(model.t_variables['gold_labels'])
    input_doc_ids = tf.saved_model.utils.build_tensor_info(model.t_variables['doc_ids'])
    input_max_sent_l = tf.saved_model.utils.build_tensor_info(model.t_variables['max_sent_l'])
    input_max_doc_l = tf.saved_model.utils.build_tensor_info(model.t_variables['max_doc_l'])
    input_mask_parser_1 = tf.saved_model.utils.build_tensor_info(model.t_variables['mask_parser_1'])
    input_mask_parser_2 = tf.saved_model.utils.build_tensor_info(model.t_variables['mask_parser_2'])
    input_batch_l = tf.saved_model.utils.build_tensor_info(model.t_variables['batch_l'])
    input_keep_prob = tf.saved_model.utils.build_tensor_info(model.t_variables['keep_prob'])

    output = tf.saved_model.utils.build_tensor_info(model.final_output)
    if config.skip_doc_attention:
        str_scores = tf.saved_model.utils.build_tensor_info(tf.convert_to_tensor(np.empty([1,1]), np.float64))
    else:
        str_scores = tf.saved_model.utils.build_tensor_info(model.str_scores)
    prediction_signature = (
        tf.saved_model.signature_def_utils.build_signature_def(
            inputs={'input_token_idxs': input_token_idxs, 'input_sent_l': input_sent_l,
                    'input_mask_tokens': input_mask_tokens, 'input_mask_sents': input_mask_sents,
                    'input_doc_l': input_doc_l, 'input_gold_labels': input_gold_labels,
                    'input_doc_ids': input_doc_ids,
                    'input_max_sent_l': input_max_sent_l, 'input_max_doc_l': input_max_doc_l,
                    'input_mask_parser_1': input_mask_parser_1, 'input_mask_parser_2': input_mask_parser_2,
                    'input_batch_l': input_batch_l, 'input_keep_prob': input_keep_prob},
            outputs={'output': output, 'str_scores': str_scores},
            method_name=tf.saved_model.signature_constants.PREDICT_METHOD_NAME))

    builder.add_meta_graph_and_variables(
        sess, [tf.saved_model.tag_constants.SERVING],
        signature_def_map={
            tf.saved_model.signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY:
                prediction_signature
        },
    )
    builder.save()
    print('Done exporting!')
    logger.info('Done exporting!')


def evaluate_pretrained_model(config, logger):
    client = InMemoryClient(config.model_dir, config.vocab_file, config.data_output_file, logger, config.skip_doc_attention)
    test_batches = client.load_data(config, config.evaluate_split)
    client.predict(test_batches, config.skip_doc_attention, config.evaluate_split)


def get_git_revision_hash():
    return subprocess.check_output(['git', 'rev-parse', 'HEAD'])