data_pipeline.py 36.4 KB
Newer Older
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
# Copyright 2018 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.
# ==============================================================================
"""Asynchronous data producer for the NCF pipeline."""

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import atexit
import functools
import os
import sys
import tempfile
import threading
import time
import timeit
import traceback
Taylor Robie's avatar
Taylor Robie committed
30
import typing
31
32
33
34
35

import numpy as np
import six
from six.moves import queue
import tensorflow as tf
36
from absl import logging
37
38

from official.recommendation import constants as rconst
39
from official.recommendation import movielens
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
from official.recommendation import popen_helper
from official.recommendation import stat_utils


SUMMARY_TEMPLATE = """General:
{spacer}Num users: {num_users}
{spacer}Num items: {num_items}

Training:
{spacer}Positive count:          {train_pos_ct}
{spacer}Batch size:              {train_batch_size} {multiplier}
{spacer}Batch count per epoch:   {train_batch_ct}

Eval:
{spacer}Positive count:          {eval_pos_ct}
{spacer}Batch size:              {eval_batch_size} {multiplier}
{spacer}Batch count per epoch:   {eval_batch_ct}"""


class DatasetManager(object):
  """Helper class for handling TensorFlow specific data tasks.

  This class takes the (relatively) framework agnostic work done by the data
  constructor classes and handles the TensorFlow specific portions (TFRecord
  management, tf.Dataset creation, etc.).
  """
66
67
68
69
70
71
72
73
74

  def __init__(self,
               is_training,
               stream_files,
               batches_per_epoch,
               shard_root=None,
               deterministic=False,
               num_train_epochs=None):
    # type: (bool, bool, int, typing.Optional[str], bool, int) -> None
Taylor Robie's avatar
Taylor Robie committed
75
76
77
78
79
80
81
82
83
84
    """Constructs a `DatasetManager` instance.
    Args:
      is_training: Boolean of whether the data provided is training or
        evaluation data. This determines whether to reuse the data
        (if is_training=False) and the exact structure to use when storing and
        yielding data.
      stream_files: Boolean indicating whether data should be serialized and
        written to file shards.
      batches_per_epoch: The number of batches in a single epoch.
      shard_root: The base directory to be used when stream_files=True.
85
      deterministic: Forgo non-deterministic speedups. (i.e. sloppy=True)
86
87
      num_train_epochs: Number of epochs to generate. If None, then each
        call to `get_dataset()` increments the number of epochs requested.
Taylor Robie's avatar
Taylor Robie committed
88
    """
89
    self._is_training = is_training
90
    self._deterministic = deterministic
91
92
93
94
95
96
    self._stream_files = stream_files
    self._writers = []
    self._write_locks = [threading.RLock() for _ in
                         range(rconst.NUM_FILE_SHARDS)] if stream_files else []
    self._batches_per_epoch = batches_per_epoch
    self._epochs_completed = 0
97
    self._epochs_requested = num_train_epochs if num_train_epochs else 0
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
    self._shard_root = shard_root

    self._result_queue = queue.Queue()
    self._result_reuse = []

  @property
  def current_data_root(self):
    subdir = (rconst.TRAIN_FOLDER_TEMPLATE.format(self._epochs_completed)
              if self._is_training else rconst.EVAL_FOLDER)
    return os.path.join(self._shard_root, subdir)

  def buffer_reached(self):
    # Only applicable for training.
    return (self._epochs_completed - self._epochs_requested >=
            rconst.CYCLES_TO_BUFFER and self._is_training)

  @staticmethod
115
  def serialize(data):
116
117
    """Convert NumPy arrays into a TFRecords entry."""

118
119
120
    def create_int_feature(values):
      return tf.train.Feature(int64_list=tf.train.Int64List(value=list(values)))

121
    feature_dict = {
122
123
        k: create_int_feature(v.astype(np.int64)) for k, v in data.items()
    }
124
125
126
127

    return tf.train.Example(
        features=tf.train.Features(feature=feature_dict)).SerializeToString()

128
129
  @staticmethod
  def deserialize(serialized_data, batch_size=None, is_training=True):
130
131
132
133
134
135
    """Convert serialized TFRecords into tensors.

    Args:
      serialized_data: A tensor containing serialized records.
      batch_size: The data arrives pre-batched, so batch size is needed to
        deserialize the data.
136
137
      is_training: Boolean, whether data to deserialize to training data
        or evaluation data.
138
139
    """

140
141
142
143
144
145
    def _get_feature_map(batch_size, is_training=True):
      """Returns data format of the serialized tf record file."""

      if is_training:
        return {
            movielens.USER_COLUMN:
146
                tf.io.FixedLenFeature([batch_size, 1], dtype=tf.int64),
147
            movielens.ITEM_COLUMN:
148
                tf.io.FixedLenFeature([batch_size, 1], dtype=tf.int64),
149
            rconst.VALID_POINT_MASK:
150
                tf.io.FixedLenFeature([batch_size, 1], dtype=tf.int64),
151
            "labels":
152
                tf.io.FixedLenFeature([batch_size, 1], dtype=tf.int64)
153
154
155
156
        }
      else:
        return {
            movielens.USER_COLUMN:
157
                tf.io.FixedLenFeature([batch_size, 1], dtype=tf.int64),
158
            movielens.ITEM_COLUMN:
159
                tf.io.FixedLenFeature([batch_size, 1], dtype=tf.int64),
160
            rconst.DUPLICATE_MASK:
161
                tf.io.FixedLenFeature([batch_size, 1], dtype=tf.int64)
162
163
        }

164
    features = tf.io.parse_single_example(
165
        serialized_data, _get_feature_map(batch_size, is_training=is_training))
166
167
    users = tf.cast(features[movielens.USER_COLUMN], rconst.USER_DTYPE)
    items = tf.cast(features[movielens.ITEM_COLUMN], rconst.ITEM_DTYPE)
168

169
    if is_training:
170
171
      valid_point_mask = tf.cast(features[rconst.VALID_POINT_MASK], tf.bool)
      fake_dup_mask = tf.zeros_like(users)
172
173
174
175
      return {
          movielens.USER_COLUMN: users,
          movielens.ITEM_COLUMN: items,
          rconst.VALID_POINT_MASK: valid_point_mask,
176
177
178
179
180
181
          rconst.TRAIN_LABEL_KEY:
              tf.reshape(tf.cast(features["labels"], tf.bool),
                         (batch_size, 1)),
          rconst.DUPLICATE_MASK: fake_dup_mask
      }
    else:
182
183
      labels = tf.cast(tf.zeros_like(users), tf.bool)
      fake_valid_pt_mask = tf.cast(tf.zeros_like(users), tf.bool)
184
185
186
187
188
189
      return {
          movielens.USER_COLUMN:
              users,
          movielens.ITEM_COLUMN:
              items,
          rconst.DUPLICATE_MASK:
190
              tf.cast(features[rconst.DUPLICATE_MASK], tf.bool),
191
192
193
194
195
          rconst.VALID_POINT_MASK:
              fake_valid_pt_mask,
          rconst.TRAIN_LABEL_KEY:
              labels
      }
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210

  def put(self, index, data):
    # type: (int, dict) -> None
    """Store data for later consumption.

    Because there are several paths for storing and yielding data (queues,
    lists, files) the data producer simply provides the data in a standard
    format at which point the dataset manager handles storing it in the correct
    form.

    Args:
      index: Used to select shards when writing to files.
      data: A dict of the data to be stored. This method mutates data, and
        therefore expects to be the only consumer.
    """
211
212
213
    if self._is_training:
      mask_start_index = data.pop(rconst.MASK_START_INDEX)
      batch_size = data[movielens.ITEM_COLUMN].shape[0]
214
215
      data[rconst.VALID_POINT_MASK] = np.expand_dims(
          np.less(np.arange(batch_size), mask_start_index), -1)
216
217

    if self._stream_files:
218
      example_bytes = self.serialize(data)
219
220
221
222
      with self._write_locks[index % rconst.NUM_FILE_SHARDS]:
        self._writers[index % rconst.NUM_FILE_SHARDS].write(example_bytes)

    else:
223
224
      self._result_queue.put((
          data, data.pop("labels")) if self._is_training else data)
225
226
227

  def start_construction(self):
    if self._stream_files:
228
      tf.io.gfile.makedirs(self.current_data_root)
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
      template = os.path.join(self.current_data_root, rconst.SHARD_TEMPLATE)
      self._writers = [tf.io.TFRecordWriter(template.format(i))
                       for i in range(rconst.NUM_FILE_SHARDS)]

  def end_construction(self):
    if self._stream_files:
      [writer.close() for writer in self._writers]
      self._writers = []
      self._result_queue.put(self.current_data_root)

    self._epochs_completed += 1

  def data_generator(self, epochs_between_evals):
    """Yields examples during local training."""
    assert not self._stream_files
Taylor Robie's avatar
Taylor Robie committed
244
    assert self._is_training or epochs_between_evals == 1
245
246
247
248
249
250

    if self._is_training:
      for _ in range(self._batches_per_epoch * epochs_between_evals):
        yield self._result_queue.get(timeout=300)

    else:
Taylor Robie's avatar
Taylor Robie committed
251
252
253
254
255
256
257
258
259
260
261
262
      if self._result_reuse:
        assert len(self._result_reuse) == self._batches_per_epoch

        for i in self._result_reuse:
          yield i
      else:
        # First epoch.
        for _ in range(self._batches_per_epoch * epochs_between_evals):
          result = self._result_queue.get(timeout=300)
          self._result_reuse.append(result)
          yield result

Shining Sun's avatar
Shining Sun committed
263
264
  def increment_request_epoch(self):
    self._epochs_requested += 1
265
266
267
268
269
270
271
272
273

  def get_dataset(self, batch_size, epochs_between_evals):
    """Construct the dataset to be used for training and eval.

    For local training, data is provided through Dataset.from_generator. For
    remote training (TPUs) the data is first serialized to files and then sent
    to the TPU through a StreamingFilesDataset.

    Args:
274
      batch_size: The per-replica batch size of the dataset.
275
276
277
      epochs_between_evals: How many epochs worth of data to yield.
        (Generator mode only.)
    """
Shining Sun's avatar
Shining Sun committed
278
    self.increment_request_epoch()
279
280
281
282
283
284
285
286
287
288
    if self._stream_files:
      if epochs_between_evals > 1:
        raise ValueError("epochs_between_evals > 1 not supported for file "
                         "based dataset.")
      epoch_data_dir = self._result_queue.get(timeout=300)
      if not self._is_training:
        self._result_queue.put(epoch_data_dir)  # Eval data is reused.

      file_pattern = os.path.join(
          epoch_data_dir, rconst.SHARD_TEMPLATE.format("*"))
Shining Sun's avatar
Shining Sun committed
289
      # TODO(seemuch): remove this contrib import
290
291
292
      # pylint: disable=line-too-long
      from tensorflow.contrib.tpu.python.tpu.datasets import StreamingFilesDataset
      # pylint: enable=line-too-long
293
      dataset = StreamingFilesDataset(
294
          files=file_pattern, worker_job=popen_helper.worker_job(),
295
296
          num_parallel_reads=rconst.NUM_FILE_SHARDS, num_epochs=1,
          sloppy=not self._deterministic)
297
298
299
300
      map_fn = functools.partial(
          self.deserialize,
          batch_size=batch_size,
          is_training=self._is_training)
301
302
303
304
305
      dataset = dataset.map(map_fn, num_parallel_calls=16)

    else:
      types = {movielens.USER_COLUMN: rconst.USER_DTYPE,
               movielens.ITEM_COLUMN: rconst.ITEM_DTYPE}
306
307
308
309
      shapes = {
          movielens.USER_COLUMN: tf.TensorShape([batch_size, 1]),
          movielens.ITEM_COLUMN: tf.TensorShape([batch_size, 1])
      }
310
311
312

      if self._is_training:
        types[rconst.VALID_POINT_MASK] = np.bool
313
        shapes[rconst.VALID_POINT_MASK] = tf.TensorShape([batch_size, 1])
314
315

        types = (types, np.bool)
316
        shapes = (shapes, tf.TensorShape([batch_size, 1]))
317
318
319

      else:
        types[rconst.DUPLICATE_MASK] = np.bool
320
        shapes[rconst.DUPLICATE_MASK] = tf.TensorShape([batch_size, 1])
321
322
323
324
325
326
327
328
329
330
331
332
333

      data_generator = functools.partial(
          self.data_generator, epochs_between_evals=epochs_between_evals)
      dataset = tf.data.Dataset.from_generator(
          generator=data_generator, output_types=types,
          output_shapes=shapes)

    return dataset.prefetch(16)

  def make_input_fn(self, batch_size):
    """Create an input_fn which checks for batch size consistency."""

    def input_fn(params):
334
335
336
337
      """Returns batches for training."""

      # Estimator passes batch_size during training and eval_batch_size during
      # eval. TPUEstimator only passes batch_size.
338
      param_batch_size = (params["batch_size"] if self._is_training else
339
                          params.get("eval_batch_size") or params["batch_size"])
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
      if batch_size != param_batch_size:
        raise ValueError("producer batch size ({}) differs from params batch "
                         "size ({})".format(batch_size, param_batch_size))

      epochs_between_evals = (params.get("epochs_between_evals", 1)
                              if self._is_training else 1)
      return self.get_dataset(batch_size=batch_size,
                              epochs_between_evals=epochs_between_evals)

    return input_fn


class BaseDataConstructor(threading.Thread):
  """Data constructor base class.

  This class manages the control flow for constructing data. It is not meant
  to be used directly, but instead subclasses should implement the following
  two methods:

    self.construct_lookup_variables
    self.lookup_negative_items

  """
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385

  def __init__(
      self,
      maximum_number_epochs,  # type: int
      num_users,  # type: int
      num_items,  # type: int
      user_map,  # type: dict
      item_map,  # type: dict
      train_pos_users,  # type: np.ndarray
      train_pos_items,  # type: np.ndarray
      train_batch_size,  # type: int
      batches_per_train_step,  # type: int
      num_train_negatives,  # type: int
      eval_pos_users,  # type: np.ndarray
      eval_pos_items,  # type: np.ndarray
      eval_batch_size,  # type: int
      batches_per_eval_step,  # type: int
      stream_files,  # type: bool
      deterministic=False,  # type: bool
      epoch_dir=None,  # type: str
      num_train_epochs=None,  # type: int
      create_data_offline=False  # type: bool
  ):
386
387
388
389
390
391
392
393
394
395
396
397
398
399
    # General constants
    self._maximum_number_epochs = maximum_number_epochs
    self._num_users = num_users
    self._num_items = num_items
    self.user_map = user_map
    self.item_map = item_map
    self._train_pos_users = train_pos_users
    self._train_pos_items = train_pos_items
    self.train_batch_size = train_batch_size
    self._num_train_negatives = num_train_negatives
    self._batches_per_train_step = batches_per_train_step
    self._eval_pos_users = eval_pos_users
    self._eval_pos_items = eval_pos_items
    self.eval_batch_size = eval_batch_size
400
401
    self.num_train_epochs = num_train_epochs
    self.create_data_offline = create_data_offline
402
403
404
405
406
407
408

    # Training
    if self._train_pos_users.shape != self._train_pos_items.shape:
      raise ValueError(
          "User positives ({}) is different from item positives ({})".format(
              self._train_pos_users.shape, self._train_pos_items.shape))

Taylor Robie's avatar
Taylor Robie committed
409
    (self._train_pos_count,) = self._train_pos_users.shape
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
    self._elements_in_epoch = (1 + num_train_negatives) * self._train_pos_count
    self.train_batches_per_epoch = self._count_batches(
        self._elements_in_epoch, train_batch_size, batches_per_train_step)

    # Evaluation
    if eval_batch_size % (1 + rconst.NUM_EVAL_NEGATIVES):
      raise ValueError("Eval batch size {} is not divisible by {}".format(
          eval_batch_size, 1 + rconst.NUM_EVAL_NEGATIVES))
    self._eval_users_per_batch = int(
        eval_batch_size // (1 + rconst.NUM_EVAL_NEGATIVES))
    self._eval_elements_in_epoch = num_users * (1 + rconst.NUM_EVAL_NEGATIVES)
    self.eval_batches_per_epoch = self._count_batches(
        self._eval_elements_in_epoch, eval_batch_size, batches_per_eval_step)

    # Intermediate artifacts
    self._current_epoch_order = np.empty(shape=(0,))
    self._shuffle_iterator = None

Taylor Robie's avatar
Taylor Robie committed
428
    self._shuffle_with_forkpool = not stream_files
429
    if stream_files:
430
      self._shard_root = epoch_dir or tempfile.mkdtemp(prefix="ncf_")
431
      atexit.register(tf.io.gfile.rmtree, dirname=self._shard_root)
432
433
434
    else:
      self._shard_root = None

435
436
437
438
439
440
441
442
    self._train_dataset = DatasetManager(True, stream_files,
                                         self.train_batches_per_epoch,
                                         self._shard_root, deterministic,
                                         num_train_epochs)
    self._eval_dataset = DatasetManager(False, stream_files,
                                        self.eval_batches_per_epoch,
                                        self._shard_root, deterministic,
                                        num_train_epochs)
443
444
445
446
447
448

    # Threading details
    super(BaseDataConstructor, self).__init__()
    self.daemon = True
    self._stop_loop = False
    self._fatal_exception = None
449
    self.deterministic = deterministic
450

Taylor Robie's avatar
Taylor Robie committed
451
  def __str__(self):
452
453
454
455
456
457
458
459
460
    multiplier = ("(x{} devices)".format(self._batches_per_train_step)
                  if self._batches_per_train_step > 1 else "")
    summary = SUMMARY_TEMPLATE.format(
        spacer="  ", num_users=self._num_users, num_items=self._num_items,
        train_pos_ct=self._train_pos_count,
        train_batch_size=self.train_batch_size,
        train_batch_ct=self.train_batches_per_epoch,
        eval_pos_ct=self._num_users, eval_batch_size=self.eval_batch_size,
        eval_batch_ct=self.eval_batches_per_epoch, multiplier=multiplier)
Taylor Robie's avatar
Taylor Robie committed
461
    return super(BaseDataConstructor, self).__str__() + "\n" + summary
462
463
464

  @staticmethod
  def _count_batches(example_count, batch_size, batches_per_step):
Taylor Robie's avatar
Taylor Robie committed
465
    """Determine the number of batches, rounding up to fill all devices."""
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
    x = (example_count + batch_size - 1) // batch_size
    return (x + batches_per_step - 1) // batches_per_step * batches_per_step

  def stop_loop(self):
    self._stop_loop = True

  def construct_lookup_variables(self):
    """Perform any one time pre-compute work."""
    raise NotImplementedError

  def lookup_negative_items(self, **kwargs):
    """Randomly sample negative items for given users."""
    raise NotImplementedError

  def _run(self):
    atexit.register(self.stop_loop)
    self._start_shuffle_iterator()
    self.construct_lookup_variables()
    self._construct_training_epoch()
    self._construct_eval_epoch()
    for _ in range(self._maximum_number_epochs - 1):
      self._construct_training_epoch()
488
    self.stop_loop()
489
490
491
492
493
494
495

  def run(self):
    try:
      self._run()
    except Exception as e:
      # The Thread base class swallows stack traces, so unfortunately it is
      # necessary to catch and re-raise to get debug output
Taylor Robie's avatar
Taylor Robie committed
496
      traceback.print_exc()
497
498
499
500
501
      self._fatal_exception = e
      sys.stderr.flush()
      raise

  def _start_shuffle_iterator(self):
502
503
504
505
    if self._shuffle_with_forkpool:
      pool = popen_helper.get_forkpool(3, closing=False)
    else:
      pool = popen_helper.get_threadpool(1, closing=False)
506
507
508
    atexit.register(pool.close)
    args = [(self._elements_in_epoch, stat_utils.random_int32())
            for _ in range(self._maximum_number_epochs)]
509
510
    imap = pool.imap if self.deterministic else pool.imap_unordered
    self._shuffle_iterator = imap(stat_utils.permutation, args)
511
512
513
514
515
516
517
518

  def _get_training_batch(self, i):
    """Construct a single batch of training data.

    Args:
      i: The index of the batch. This is used when stream_files=True to assign
        data to file shards.
    """
Taylor Robie's avatar
Taylor Robie committed
519
520
521
    batch_indices = self._current_epoch_order[i * self.train_batch_size:
                                              (i + 1) * self.train_batch_size]
    (mask_start_index,) = batch_indices.shape
522
523
524
525
526
527
528
529
530
531
532
533

    batch_ind_mod = np.mod(batch_indices, self._train_pos_count)
    users = self._train_pos_users[batch_ind_mod]

    negative_indices = np.greater_equal(batch_indices, self._train_pos_count)
    negative_users = users[negative_indices]

    negative_items = self.lookup_negative_items(negative_users=negative_users)

    items = self._train_pos_items[batch_ind_mod]
    items[negative_indices] = negative_items

Taylor Robie's avatar
Taylor Robie committed
534
    labels = np.logical_not(negative_indices)
535
536
537
538
539
540
541
542
543
544
545
546
547
548

    # Pad last partial batch
    pad_length = self.train_batch_size - mask_start_index
    if pad_length:
      # We pad with arange rather than zeros because the network will still
      # compute logits for padded examples, and padding with zeros would create
      # a very "hot" embedding key which can have performance implications.
      user_pad = np.arange(pad_length, dtype=users.dtype) % self._num_users
      item_pad = np.arange(pad_length, dtype=items.dtype) % self._num_items
      label_pad = np.zeros(shape=(pad_length,), dtype=labels.dtype)
      users = np.concatenate([users, user_pad])
      items = np.concatenate([items, item_pad])
      labels = np.concatenate([labels, label_pad])

549
550
551
552
553
554
555
556
557
558
559
    self._train_dataset.put(
        i, {
            movielens.USER_COLUMN:
                np.reshape(users, (self.train_batch_size, 1)),
            movielens.ITEM_COLUMN:
                np.reshape(items, (self.train_batch_size, 1)),
            rconst.MASK_START_INDEX:
                np.array(mask_start_index, dtype=np.int32),
            "labels":
                np.reshape(labels, (self.train_batch_size, 1)),
        })
560
561
562
563
564
565
566

  def _wait_to_construct_train_epoch(self):
    count = 0
    while self._train_dataset.buffer_reached() and not self._stop_loop:
      time.sleep(0.01)
      count += 1
      if count >= 100 and np.log10(count) == np.round(np.log10(count)):
567
        logging.info(
568
569
570
571
            "Waited {} times for training data to be consumed".format(count))

  def _construct_training_epoch(self):
    """Loop to construct a batch of training data."""
572
573
574
    if not self.create_data_offline:
      self._wait_to_construct_train_epoch()

575
576
577
578
579
580
    start_time = timeit.default_timer()
    if self._stop_loop:
      return

    self._train_dataset.start_construction()
    map_args = list(range(self.train_batches_per_epoch))
Taylor Robie's avatar
Taylor Robie committed
581
    self._current_epoch_order = next(self._shuffle_iterator)
582

583
584
585
    get_pool = (popen_helper.get_fauxpool if self.deterministic else
                popen_helper.get_threadpool)
    with get_pool(6) as pool:
586
587
588
      pool.map(self._get_training_batch, map_args)
    self._train_dataset.end_construction()

589
    logging.info("Epoch construction complete. Time: {:.1f} seconds".format(
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
        timeit.default_timer() - start_time))

  @staticmethod
  def _assemble_eval_batch(users, positive_items, negative_items,
                           users_per_batch):
    """Construct duplicate_mask and structure data accordingly.

    The positive items should be last so that they lose ties. However, they
    should not be masked out if the true eval positive happens to be
    selected as a negative. So instead, the positive is placed in the first
    position, and then switched with the last element after the duplicate
    mask has been computed.

    Args:
      users: An array of users in a batch. (should be identical along axis 1)
      positive_items: An array (batch_size x 1) of positive item indices.
      negative_items: An array of negative item indices.
      users_per_batch: How many users should be in the batch. This is passed
        as an argument so that ncf_test.py can use this method.

    Returns:
      User, item, and duplicate_mask arrays.
    """
    items = np.concatenate([positive_items, negative_items], axis=1)

    # We pad the users and items here so that the duplicate mask calculation
Taylor Robie's avatar
Taylor Robie committed
616
    # will include padding. The metric function relies on all padded elements
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
    # except the positive being marked as duplicate to mask out padded points.
    if users.shape[0] < users_per_batch:
      pad_rows = users_per_batch - users.shape[0]
      padding = np.zeros(shape=(pad_rows, users.shape[1]), dtype=np.int32)
      users = np.concatenate([users, padding.astype(users.dtype)], axis=0)
      items = np.concatenate([items, padding.astype(items.dtype)], axis=0)

    duplicate_mask = stat_utils.mask_duplicates(items, axis=1).astype(np.bool)

    items[:, (0, -1)] = items[:, (-1, 0)]
    duplicate_mask[:, (0, -1)] = duplicate_mask[:, (-1, 0)]

    assert users.shape == items.shape == duplicate_mask.shape
    return users, items, duplicate_mask

  def _get_eval_batch(self, i):
    """Construct a single batch of evaluation data.

    Args:
      i: The index of the batch.
    """
    low_index = i * self._eval_users_per_batch
    high_index = (i + 1) * self._eval_users_per_batch
    users = np.repeat(self._eval_pos_users[low_index:high_index, np.newaxis],
                      1 + rconst.NUM_EVAL_NEGATIVES, axis=1)
    positive_items = self._eval_pos_items[low_index:high_index, np.newaxis]
    negative_items = (self.lookup_negative_items(negative_users=users[:, :-1])
                      .reshape(-1, rconst.NUM_EVAL_NEGATIVES))

    users, items, duplicate_mask = self._assemble_eval_batch(
        users, positive_items, negative_items, self._eval_users_per_batch)

649
650
651
652
653
654
655
656
657
    self._eval_dataset.put(
        i, {
            movielens.USER_COLUMN:
                np.reshape(users.flatten(), (self.eval_batch_size, 1)),
            movielens.ITEM_COLUMN:
                np.reshape(items.flatten(), (self.eval_batch_size, 1)),
            rconst.DUPLICATE_MASK:
                np.reshape(duplicate_mask.flatten(), (self.eval_batch_size, 1)),
        })
658
659
660
661
662
663
664
665
666
667

  def _construct_eval_epoch(self):
    """Loop to construct data for evaluation."""
    if self._stop_loop:
      return

    start_time = timeit.default_timer()

    self._eval_dataset.start_construction()
    map_args = [i for i in range(self.eval_batches_per_epoch)]
668
669
670
671

    get_pool = (popen_helper.get_fauxpool if self.deterministic else
                popen_helper.get_threadpool)
    with get_pool(6) as pool:
672
673
674
      pool.map(self._get_eval_batch, map_args)
    self._eval_dataset.end_construction()

675
    logging.info("Eval construction complete. Time: {:.1f} seconds".format(
676
677
678
        timeit.default_timer() - start_time))

  def make_input_fn(self, is_training):
Taylor Robie's avatar
Taylor Robie committed
679
680
    # It isn't feasible to provide a foolproof check, so this is designed to
    # catch most failures rather than provide an exhaustive guard.
681
682
683
684
685
686
687
688
    if self._fatal_exception is not None:
      raise ValueError("Fatal exception in the data production loop: {}"
                       .format(self._fatal_exception))

    return (
        self._train_dataset.make_input_fn(self.train_batch_size) if is_training
        else self._eval_dataset.make_input_fn(self.eval_batch_size))

Shining Sun's avatar
Shining Sun committed
689
690
691
  def increment_request_epoch(self):
    self._train_dataset.increment_request_epoch()

692
693
694

class DummyConstructor(threading.Thread):
  """Class for running with synthetic data."""
695

696
697
698
699
700
  def __init__(self, *args, **kwargs):
    super(DummyConstructor, self).__init__(*args, **kwargs)
    self.train_batches_per_epoch = rconst.SYNTHETIC_BATCHES_PER_EPOCH
    self.eval_batches_per_epoch = rconst.SYNTHETIC_BATCHES_PER_EPOCH

701
702
703
704
705
706
  def run(self):
    pass

  def stop_loop(self):
    pass

Shining Sun's avatar
Shining Sun committed
707
708
709
  def increment_request_epoch(self):
    pass

710
711
712
713
714
  @staticmethod
  def make_input_fn(is_training):
    """Construct training input_fn that uses synthetic data."""

    def input_fn(params):
715
716
717
718
      """Returns dummy input batches for training."""

      # Estimator passes batch_size during training and eval_batch_size during
      # eval. TPUEstimator only passes batch_size.
719
      batch_size = (params["batch_size"] if is_training else
720
                    params.get("eval_batch_size") or params["batch_size"])
721
722
723
      num_users = params["num_users"]
      num_items = params["num_items"]

724
725
726
      users = tf.random.uniform([batch_size, 1],
                                dtype=tf.int32,
                                minval=0,
727
                                maxval=num_users)
728
729
730
      items = tf.random.uniform([batch_size, 1],
                                dtype=tf.int32,
                                minval=0,
731
732
733
                                maxval=num_items)

      if is_training:
734
735
736
737
738
739
740
741
742
743
        valid_point_mask = tf.cast(
            tf.random.uniform([batch_size, 1],
                              dtype=tf.int32,
                              minval=0,
                              maxval=2), tf.bool)
        labels = tf.cast(
            tf.random.uniform([batch_size, 1],
                              dtype=tf.int32,
                              minval=0,
                              maxval=2), tf.bool)
744
745
746
747
748
749
        data = {
            movielens.USER_COLUMN: users,
            movielens.ITEM_COLUMN: items,
            rconst.VALID_POINT_MASK: valid_point_mask,
        }, labels
      else:
750
751
752
753
754
        dupe_mask = tf.cast(
            tf.random.uniform([batch_size, 1],
                              dtype=tf.int32,
                              minval=0,
                              maxval=2), tf.bool)
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
        data = {
            movielens.USER_COLUMN: users,
            movielens.ITEM_COLUMN: items,
            rconst.DUPLICATE_MASK: dupe_mask,
        }

      dataset = tf.data.Dataset.from_tensors(data).repeat(
          rconst.SYNTHETIC_BATCHES_PER_EPOCH * params["batches_per_step"])
      dataset = dataset.prefetch(32)
      return dataset

    return input_fn


class MaterializedDataConstructor(BaseDataConstructor):
  """Materialize a table of negative examples for fast negative generation.

  This class creates a table (num_users x num_items) containing all of the
  negative examples for each user. This table is conceptually ragged; that is to
Taylor Robie's avatar
Taylor Robie committed
774
  say the items dimension will have a number of unused elements at the end equal
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
  to the number of positive elements for a given user. For instance:

  num_users = 3
  num_items = 5
  positives = [[1, 3], [0], [1, 2, 3, 4]]

  will generate a negative table:
  [
    [0         2         4         int32max  int32max],
    [1         2         3         4         int32max],
    [0         int32max  int32max  int32max  int32max],
  ]

  and a vector of per-user negative counts, which in this case would be:
    [3, 4, 1]

  When sampling negatives, integers are (nearly) uniformly selected from the
  range [0, per_user_neg_count[user]) which gives a column_index, at which
  point the negative can be selected as:
    negative_table[user, column_index]

  This technique will not scale; however MovieLens is small enough that even
  a pre-compute which is quadratic in problem size will still fit in memory. A
  more scalable lookup method is in the works.
  """
  def __init__(self, *args, **kwargs):
    super(MaterializedDataConstructor, self).__init__(*args, **kwargs)
    self._negative_table = None
    self._per_user_neg_count = None

  def construct_lookup_variables(self):
    # Materialize negatives for fast lookup sampling.
    start_time = timeit.default_timer()
    inner_bounds = np.argwhere(self._train_pos_users[1:] -
                               self._train_pos_users[:-1])[:, 0] + 1
Taylor Robie's avatar
Taylor Robie committed
810
    (upper_bound,) = self._train_pos_users.shape
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
    index_bounds = [0] + inner_bounds.tolist() + [upper_bound]
    self._negative_table = np.zeros(shape=(self._num_users, self._num_items),
                                    dtype=rconst.ITEM_DTYPE)

    # Set the table to the max value to make sure the embedding lookup will fail
    # if we go out of bounds, rather than just overloading item zero.
    self._negative_table += np.iinfo(rconst.ITEM_DTYPE).max
    assert self._num_items < np.iinfo(rconst.ITEM_DTYPE).max

    # Reuse arange during generation. np.delete will make a copy.
    full_set = np.arange(self._num_items, dtype=rconst.ITEM_DTYPE)

    self._per_user_neg_count = np.zeros(
        shape=(self._num_users,), dtype=np.int32)

    # Threading does not improve this loop. For some reason, the np.delete
    # call does not parallelize well. Multiprocessing incurs too much
    # serialization overhead to be worthwhile.
    for i in range(self._num_users):
      positives = self._train_pos_items[index_bounds[i]:index_bounds[i+1]]
      negatives = np.delete(full_set, positives)
      self._per_user_neg_count[i] = self._num_items - positives.shape[0]
      self._negative_table[i, :self._per_user_neg_count[i]] = negatives

835
    logging.info("Negative sample table built. Time: {:.1f} seconds".format(
836
837
838
839
840
841
        timeit.default_timer() - start_time))

  def lookup_negative_items(self, negative_users, **kwargs):
    negative_item_choice = stat_utils.very_slightly_biased_randint(
        self._per_user_neg_count[negative_users])
    return self._negative_table[negative_users, negative_item_choice]
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887


class BisectionDataConstructor(BaseDataConstructor):
  """Use bisection to index within positive examples.

  This class tallies the number of negative items which appear before each
  positive item for a user. This means that in order to select the ith negative
  item for a user, it only needs to determine which two positive items bound
  it at which point the item id for the ith negative is a simply algebraic
  expression.
  """
  def __init__(self, *args, **kwargs):
    super(BisectionDataConstructor, self).__init__(*args, **kwargs)
    self.index_bounds = None
    self._sorted_train_pos_items = None
    self._total_negatives = None

  def _index_segment(self, user):
    lower, upper = self.index_bounds[user:user+2]
    items = self._sorted_train_pos_items[lower:upper]

    negatives_since_last_positive = np.concatenate(
        [items[0][np.newaxis], items[1:] - items[:-1] - 1])

    return np.cumsum(negatives_since_last_positive)

  def construct_lookup_variables(self):
    start_time = timeit.default_timer()
    inner_bounds = np.argwhere(self._train_pos_users[1:] -
                               self._train_pos_users[:-1])[:, 0] + 1
    (upper_bound,) = self._train_pos_users.shape
    self.index_bounds = np.array([0] + inner_bounds.tolist() + [upper_bound])

    # Later logic will assume that the users are in sequential ascending order.
    assert np.array_equal(self._train_pos_users[self.index_bounds[:-1]],
                          np.arange(self._num_users))

    self._sorted_train_pos_items = self._train_pos_items.copy()

    for i in range(self._num_users):
      lower, upper = self.index_bounds[i:i+2]
      self._sorted_train_pos_items[lower:upper].sort()

    self._total_negatives = np.concatenate([
        self._index_segment(i) for i in range(self._num_users)])

888
    logging.info("Negative total vector built. Time: {:.1f} seconds".format(
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
        timeit.default_timer() - start_time))

  def lookup_negative_items(self, negative_users, **kwargs):
    output = np.zeros(shape=negative_users.shape, dtype=rconst.ITEM_DTYPE) - 1

    left_index = self.index_bounds[negative_users]
    right_index = self.index_bounds[negative_users + 1] - 1

    num_positives = right_index - left_index + 1
    num_negatives = self._num_items - num_positives
    neg_item_choice = stat_utils.very_slightly_biased_randint(num_negatives)

    # Shortcuts:
    # For points where the negative is greater than or equal to the tally before
    # the last positive point there is no need to bisect. Instead the item id
    # corresponding to the negative item choice is simply:
    #   last_postive_index + 1 + (neg_choice - last_negative_tally)
    # Similarly, if the selection is less than the tally at the first positive
    # then the item_id is simply the selection.
    #
    # Because MovieLens organizes popular movies into low integers (which is
    # preserved through the preprocessing), the first shortcut is very
    # efficient, allowing ~60% of samples to bypass the bisection. For the same
    # reason, the second shortcut is rarely triggered (<0.02%) and is therefore
    # not worth implementing.
    use_shortcut = neg_item_choice >= self._total_negatives[right_index]
    output[use_shortcut] = (
        self._sorted_train_pos_items[right_index] + 1 +
        (neg_item_choice - self._total_negatives[right_index])
    )[use_shortcut]

920
921
922
923
    if np.all(use_shortcut):
      # The bisection code is ill-posed when there are no elements.
      return output

924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
    not_use_shortcut = np.logical_not(use_shortcut)
    left_index = left_index[not_use_shortcut]
    right_index = right_index[not_use_shortcut]
    neg_item_choice = neg_item_choice[not_use_shortcut]

    num_loops = np.max(
        np.ceil(np.log2(num_positives[not_use_shortcut])).astype(np.int32))

    for i in range(num_loops):
      mid_index = (left_index + right_index) // 2
      right_criteria = self._total_negatives[mid_index] > neg_item_choice
      left_criteria = np.logical_not(right_criteria)

      right_index[right_criteria] = mid_index[right_criteria]
      left_index[left_criteria] = mid_index[left_criteria]

    # Expected state after bisection pass:
    #   The right index is the smallest index whose tally is greater than the
    #   negative item choice index.

    assert np.all((right_index - left_index) <= 1)

    output[not_use_shortcut] = (
        self._sorted_train_pos_items[right_index] -
        (self._total_negatives[right_index] - neg_item_choice)
    )

    assert np.all(output >= 0)

    return output


def get_constructor(name):
  if name == "bisection":
    return BisectionDataConstructor
  if name == "materialized":
    return MaterializedDataConstructor
  raise ValueError("Unrecognized constructor: {}".format(name))