sentence_prediction_test.py 6.37 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
# Lint as: python3
# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for official.nlp.tasks.sentence_prediction."""
17
import functools
18
import os
19
20

from absl.testing import parameterized
21
22
23
24
25
26
import tensorflow as tf

from official.nlp.bert import configs
from official.nlp.bert import export_tfhub
from official.nlp.configs import bert
from official.nlp.configs import encoders
Chen Chen's avatar
Chen Chen committed
27
from official.nlp.data import sentence_prediction_dataloader
28
29
30
from official.nlp.tasks import sentence_prediction


31
class SentencePredictionTaskTest(tf.test.TestCase, parameterized.TestCase):
32

A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
33
34
  def setUp(self):
    super(SentencePredictionTaskTest, self).setUp()
Chen Chen's avatar
Chen Chen committed
35
36
37
    self._train_data_config = (
        sentence_prediction_dataloader.SentencePredictionDataConfig(
            input_path="dummy", seq_length=128, global_batch_size=1))
38

Pengchong Jin's avatar
Pengchong Jin committed
39
  def get_model_config(self, num_classes):
40
    return bert.BertPretrainerConfig(
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
41
42
43
44
45
        encoder=encoders.TransformerEncoderConfig(
            vocab_size=30522, num_layers=1),
        num_masked_tokens=0,
        cls_heads=[
            bert.ClsHeadConfig(
46
47
48
                inner_dim=10,
                num_classes=num_classes,
                name="sentence_prediction")
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
49
50
        ])

51
52
53
54
55
56
  def _run_task(self, config):
    task = sentence_prediction.SentencePredictionTask(config)
    model = task.build_model()
    metrics = task.build_metrics()

    strategy = tf.distribute.get_strategy()
57
58
    dataset = strategy.experimental_distribute_datasets_from_function(
        functools.partial(task.build_inputs, config.train_data))
59
60
61
62
63
64
65
66

    iterator = iter(dataset)
    optimizer = tf.keras.optimizers.SGD(lr=0.1)
    task.train_step(next(iterator), model, optimizer, metrics=metrics)
    task.validation_step(next(iterator), model, metrics=metrics)

  def test_task(self):
    config = sentence_prediction.SentencePredictionConfig(
Hongkun Yu's avatar
Hongkun Yu committed
67
        init_checkpoint=self.get_temp_dir(),
Pengchong Jin's avatar
Pengchong Jin committed
68
        model=self.get_model_config(2),
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
69
        train_data=self._train_data_config)
70
71
72
73
74
75
76
77
78
79
    task = sentence_prediction.SentencePredictionTask(config)
    model = task.build_model()
    metrics = task.build_metrics()
    dataset = task.build_inputs(config.train_data)

    iterator = iter(dataset)
    optimizer = tf.keras.optimizers.SGD(lr=0.1)
    task.train_step(next(iterator), model, optimizer, metrics=metrics)
    task.validation_step(next(iterator), model, metrics=metrics)

Hongkun Yu's avatar
Hongkun Yu committed
80
81
82
83
84
85
86
87
88
    # Saves a checkpoint.
    pretrain_cfg = bert.BertPretrainerConfig(
        encoder=encoders.TransformerEncoderConfig(
            vocab_size=30522, num_layers=1),
        num_masked_tokens=20,
        cls_heads=[
            bert.ClsHeadConfig(
                inner_dim=10, num_classes=3, name="next_sentence")
        ])
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
89
    pretrain_model = bert.instantiate_bertpretrainer_from_cfg(pretrain_cfg)
Hongkun Yu's avatar
Hongkun Yu committed
90
91
92
93
94
    ckpt = tf.train.Checkpoint(
        model=pretrain_model, **pretrain_model.checkpoint_items)
    ckpt.save(config.init_checkpoint)
    task.initialize(model)

95
96
97
  @parameterized.parameters(("matthews_corrcoef", 2),
                            ("pearson_spearman_corr", 1))
  def test_np_metrics(self, metric_type, num_classes):
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
98
    config = sentence_prediction.SentencePredictionConfig(
99
100
        metric_type=metric_type,
        init_checkpoint=self.get_temp_dir(),
Pengchong Jin's avatar
Pengchong Jin committed
101
        model=self.get_model_config(num_classes),
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
102
103
104
        train_data=self._train_data_config)
    task = sentence_prediction.SentencePredictionTask(config)
    model = task.build_model()
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
    dataset = task.build_inputs(config.train_data)

    iterator = iter(dataset)
    strategy = tf.distribute.get_strategy()
    distributed_outputs = strategy.run(
        functools.partial(task.validation_step, model=model),
        args=(next(iterator),))
    outputs = tf.nest.map_structure(strategy.experimental_local_results,
                                    distributed_outputs)
    aggregated = task.aggregate_logs(step_outputs=outputs)
    aggregated = task.aggregate_logs(state=aggregated, step_outputs=outputs)
    self.assertIn(metric_type, task.reduce_aggregated_logs(aggregated))

  def test_task_with_fit(self):
    config = sentence_prediction.SentencePredictionConfig(
Pengchong Jin's avatar
Pengchong Jin committed
120
        model=self.get_model_config(2), train_data=self._train_data_config)
121
122
    task = sentence_prediction.SentencePredictionTask(config)
    model = task.build_model()
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
123
124
125
126
127
128
129
130
131
    model = task.compile_model(
        model,
        optimizer=tf.keras.optimizers.SGD(lr=0.1),
        train_step=task.train_step,
        metrics=task.build_metrics())
    dataset = task.build_inputs(config.train_data)
    logs = model.fit(dataset, epochs=1, steps_per_epoch=2)
    self.assertIn("loss", logs.history)

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
  def _export_bert_tfhub(self):
    bert_config = configs.BertConfig(
        vocab_size=30522,
        hidden_size=16,
        intermediate_size=32,
        max_position_embeddings=128,
        num_attention_heads=2,
        num_hidden_layers=1)
    _, encoder = export_tfhub.create_bert_model(bert_config)
    model_checkpoint_dir = os.path.join(self.get_temp_dir(), "checkpoint")
    checkpoint = tf.train.Checkpoint(model=encoder)
    checkpoint.save(os.path.join(model_checkpoint_dir, "test"))
    model_checkpoint_path = tf.train.latest_checkpoint(model_checkpoint_dir)

    vocab_file = os.path.join(self.get_temp_dir(), "uncased_vocab.txt")
    with tf.io.gfile.GFile(vocab_file, "w") as f:
      f.write("dummy content")

    hub_destination = os.path.join(self.get_temp_dir(), "hub")
    export_tfhub.export_bert_tfhub(bert_config, model_checkpoint_path,
                                   hub_destination, vocab_file)
    return hub_destination

  def test_task_with_hub(self):
    hub_module_url = self._export_bert_tfhub()
    config = sentence_prediction.SentencePredictionConfig(
        hub_module_url=hub_module_url,
Pengchong Jin's avatar
Pengchong Jin committed
159
        model=self.get_model_config(2),
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
160
        train_data=self._train_data_config)
161
162
163
164
165
    self._run_task(config)


if __name__ == "__main__":
  tf.test.main()