Trainer.py 11.6 KB
Newer Older
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
1
2
3
4
5
6
7
#!/usr/bin/env python
# vim: set fileencoding=utf-8 :
# @author: Tiago de Freitas Pereira <tiago.pereira@idiap.ch>
# @date: Tue 09 Aug 2016 15:25:22 CEST

import tensorflow as tf
from ..network import SequenceNetwork
8
9
10
import threading
import os
import bob.io.base
11
import bob.core
12
from ..analyzers import SoftmaxAnalizer
13
from tensorflow.core.framework import summary_pb2
14
import time
15
16
from bob.learn.tensorflow.datashuffler.OnlineSampling import OnLineSampling

17
os.environ["CUDA_VISIBLE_DEVICES"] = "1,2,3,0"
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
18

19
logger = bob.core.log.setup("bob.learn.tensorflow")
20

Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
21

22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
class Trainer(object):
    """
    One graph trainer.
    Use this trainer when your CNN is composed by one graph

    **Parameters**
      architecture: The architecture that you want to run. Should be a :py:class`bob.learn.tensorflow.network.SequenceNetwork`
      optimizer: One of the tensorflow optimizers https://www.tensorflow.org/versions/r0.10/api_docs/python/train.html
      use_gpu: Use GPUs in the training
      loss: Loss
      temp_dir: The output directory

      base_learning_rate: Initial learning rate
      weight_decay:
      convergence_threshold:

      iterations: Maximum number of iterations
      snapshot: Will take a snapshot of the network at every `n` iterations
      prefetch: Use extra Threads to deal with the I/O
      analizer: Neural network analizer :py:mod:`bob.learn.tensorflow.analyzers`
      verbosity_level:

    """
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
45
    def __init__(self,
46
47
                 architecture,
                 optimizer=tf.train.AdamOptimizer(),
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
48
49
                 use_gpu=False,
                 loss=None,
50
                 temp_dir="cnn",
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
51

52
53
54
55
                 # Learning rate
                 base_learning_rate=0.001,
                 weight_decay=0.9,

Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
56
                 ###### training options ##########
57
                 convergence_threshold=0.01,
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
58
                 iterations=5000,
59
60
                 snapshot=100,
                 prefetch=False,
61
62

                 ## Analizer
63
                 analizer=SoftmaxAnalizer(),
64

65
                 verbosity_level=2):
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
66

67
68
        if not isinstance(architecture, SequenceNetwork):
            raise ValueError("`architecture` should be instance of `SequenceNetwork`")
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
69
70

        self.architecture = architecture
71
        self.optimizer_class = optimizer
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
72
        self.use_gpu = use_gpu
73
74
75
76
77
        self.loss = loss
        self.temp_dir = temp_dir

        self.base_learning_rate = base_learning_rate
        self.weight_decay = weight_decay
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
78
79
80
81

        self.iterations = iterations
        self.snapshot = snapshot
        self.convergence_threshold = convergence_threshold
82
        self.prefetch = prefetch
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
83

84
85
86
87
88
89
90
91
        # Training variables used in the fit
        self.optimizer = None
        self.training_graph = None
        self.learning_rate = None
        self.training_graph = None
        self.train_data_shuffler = None
        self.summaries_train = None
        self.train_summary_writter = None
92
        self.thread_pool = None
93
94
95
96
97

        # Validation data
        self.validation_graph = None
        self.validation_summary_writter = None

98
99
100
101
102
103
        # Analizer
        self.analizer = analizer

        self.thread_pool = None
        self.enqueue_op = None

104
105
        bob.core.log.set_verbosity_level(logger, verbosity_level)

Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
106
107
108
    def __del__(self):
        tf.reset_default_graph()

109
    def compute_graph(self, data_shuffler, prefetch=False, name=""):
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
110
        """
111
112
        Computes the graph for the trainer.

113

114
115
116
        ** Parameters **

            data_shuffler: Data shuffler
117
            prefetch:
118
119
120
121
            name: Name of the graph
        """

        # Defining place holders
122
        if prefetch:
123
            [placeholder_data, placeholder_labels] = data_shuffler.get_placeholders_forprefetch(name=name)
124
125
126
127
128
129
130

            # Defining a placeholder queue for prefetching
            queue = tf.FIFOQueue(capacity=10,
                                 dtypes=[tf.float32, tf.int64],
                                 shapes=[placeholder_data.get_shape().as_list()[1:], []])

            # Fetching the place holders from the queue
131
            self.enqueue_op = queue.enqueue_many([placeholder_data, placeholder_labels])
132
133
134
135
136
137
138
            feature_batch, label_batch = queue.dequeue_many(data_shuffler.batch_size)

            # Creating the architecture for train and validation
            if not isinstance(self.architecture, SequenceNetwork):
                raise ValueError("The variable `architecture` must be an instance of "
                                 "`bob.learn.tensorflow.network.SequenceNetwork`")
        else:
139
            [feature_batch, label_batch] = data_shuffler.get_placeholders(name=name)
140
141
142
143
144
145
146
147
148

        # Creating graphs and defining the loss
        network_graph = self.architecture.compute_graph(feature_batch)
        graph = self.loss(network_graph, label_batch)

        return graph

    def get_feed_dict(self, data_shuffler):
        """
149
        Given a data shuffler prepared the dictionary to be injected in the graph
150
151
152
153

        ** Parameters **
            data_shuffler:

Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
154
        """
155
156
        [data, labels] = data_shuffler.get_batch()
        [data_placeholder, label_placeholder] = data_shuffler.get_placeholders()
157
158
159
160
161

        feed_dict = {data_placeholder: data,
                     label_placeholder: labels}
        return feed_dict

162
163
164
165
166
167
168
169
170
171
    def fit(self, session, step):
        """
        Run one iteration (`forward` and `backward`)

        ** Parameters **
            session: Tensorflow session
            step: Iteration number

        """

172
        if self.prefetch:
173
174
            _, l, lr, summary = session.run([self.optimizer, self.training_graph,
                                             self.learning_rate, self.summaries_train])
175
176
177
178
179
        else:
            feed_dict = self.get_feed_dict(self.train_data_shuffler)
            _, l, lr, summary = session.run([self.optimizer, self.training_graph,
                                             self.learning_rate, self.summaries_train], feed_dict=feed_dict)

180
181
        logger.info("Loss training set step={0} = {1}".format(step, l))
        self.train_summary_writter.add_summary(summary, step)
182

183
    def compute_validation(self,  session, data_shuffler, step):
184
185
186
187
188
189
190
191
192
        """
        Computes the loss in the validation set

        ** Parameters **
            session: Tensorflow session
            data_shuffler: The data shuffler to be used
            step: Iteration number

        """
193
        # Opening a new session for validation
194
195
196
197
        self.validation_graph = self.compute_graph(data_shuffler, name="validation")
        feed_dict = self.get_feed_dict(data_shuffler)
        l = session.run(self.validation_graph, feed_dict=feed_dict)

198
199
200
        if self.validation_summary_writter is None:
            self.validation_summary_writter = tf.train.SummaryWriter(os.path.join(self.temp_dir, 'validation'), session.graph)

201
202
203
204
205
        summaries = []
        summaries.append(summary_pb2.Summary.Value(tag="loss", simple_value=float(l)))
        self.validation_summary_writter.add_summary(summary_pb2.Summary(value=summaries), step)
        logger.info("Loss VALIDATION set step={0} = {1}".format(step, l))

206
207
208
209
210
    def create_general_summary(self):
        """
        Creates a simple tensorboard summary with the value of the loss and learning rate
        """

211
212
213
214
215
        # Train summary
        tf.scalar_summary('loss', self.training_graph, name="train")
        tf.scalar_summary('lr', self.learning_rate, name="train")
        return tf.merge_all_summaries()

216
    def start_thread(self, session):
217
218
219
220
221
222
223
        """
        Start pool of threads for pre-fetching

        **Parameters**
          session: Tensorflow session
        """

224
        threads = []
225
226
        for n in range(3):
            t = threading.Thread(target=self.load_and_enqueue, args=(session,))
227
228
229
230
            t.daemon = True  # thread will close when parent quits
            t.start()
            threads.append(t)
        return threads
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
231

232
233
    def load_and_enqueue(self, session):
        """
234
        Injecting data in the place holder queue
235
236
237

        **Parameters**
          session: Tensorflow session
238
        """
239

240
        while not self.thread_pool.should_stop():
241
242
            [train_data, train_labels] = self.train_data_shuffler.get_batch()
            [train_placeholder_data, train_placeholder_labels] = self.train_data_shuffler.get_placeholders()
243

244
245
246
            feed_dict = {train_placeholder_data: train_data,
                         train_placeholder_labels: train_labels}

247
            session.run(self.enqueue_op, feed_dict=feed_dict)
248
249
250

    def train(self, train_data_shuffler, validation_data_shuffler=None):
        """
251
        Train the network
252
253
254
255
256
        """

        # Creating directory
        bob.io.base.create_directories_safe(self.temp_dir)
        self.train_data_shuffler = train_data_shuffler
257

258
        # TODO: find an elegant way to provide this as a parameter of the trainer
259
        self.learning_rate = tf.train.exponential_decay(
260
261
262
263
264
265
            self.base_learning_rate,  # Learning rate
            train_data_shuffler.batch_size,
            train_data_shuffler.n_samples,
            self.weight_decay  # Decay step
        )

266
        self.training_graph = self.compute_graph(train_data_shuffler, prefetch=self.prefetch, name="train")
267

268
        # Preparing the optimizer
269
        self.optimizer_class._learning_rate = self.learning_rate
270
271
        #self.optimizer = self.optimizer_class.minimize(self.training_graph, global_step=tf.Variable(0))
        self.optimizer = self.optimizer_class.minimize(self.training_graph)
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
272

273
        # Train summary
274
        self.summaries_train = self.create_general_summary()
275
276

        logger.info("Initializing !!")
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
277
        # Training
278
        hdf5 = bob.io.base.HDF5File(os.path.join(self.temp_dir, 'model.hdf5'), 'w')
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
279

280
281
282
        config = tf.ConfigProto(log_device_placement=True)
        config.gpu_options.allow_growth = True
        with tf.Session(config=True) as session:
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
283

Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
284
            tf.initialize_all_variables().run()
285

286
287
288
            if isinstance(train_data_shuffler, OnLineSampling):
                train_data_shuffler.set_feature_extractor(self.architecture, session=session)

289
            # Start a thread to enqueue data asynchronously, and hide I/O latency.
290
291
292
293
            if self.prefetch:
                self.thread_pool = tf.train.Coordinator()
                tf.train.start_queue_runners(coord=self.thread_pool)
                threads = self.start_thread(session)
294

295
            # TENSOR BOARD SUMMARY
296
            self.train_summary_writter = tf.train.SummaryWriter(os.path.join(self.temp_dir, 'train'), session.graph)
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
297
            for step in range(self.iterations):
298
299
300
301
302
303
304

                start = time.time()
                self.fit(session, step)
                end = time.time()
                summary = summary_pb2.Summary.Value(tag="elapsed_time", simple_value=float(end-start))
                self.train_summary_writter.add_summary(summary_pb2.Summary(value=[summary]), step)

305
                if validation_data_shuffler is not None and step % self.snapshot == 0:
306
                    self.compute_validation(session, validation_data_shuffler, step)
307

308
309
                    if self.analizer is not None:
                        self.validation_summary_writter.add_summary(self.analizer(
310
                             validation_data_shuffler, self.architecture, session), step)
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
311

312
313
314
315
316
            logger.info("Training finally finished")

            self.train_summary_writter.close()
            if validation_data_shuffler is not None:
                self.validation_summary_writter.close()
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
317

318
319
320
            self.architecture.save(hdf5)
            del hdf5

321
322
323
324
            if self.prefetch:
                # now they should definetely stop
                self.thread_pool.request_stop()
                self.thread_pool.join(threads)
325

326
            session.close() # For some reason the session is not closed after the context manager finishes