local_mnist.py

# Copyright 2018 Google Inc. 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.
"""Reading MNIST dataset locally.

This library contains functions used to read MNIST-family data such as vanilla
MNIST or Fashion-MNIST. Typical usage is:

  data_dir = ...
  train, validation, test = read_data_sets(data_dir)
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import gzip
import os

import numpy as np
import tensorflow as tf

gfile = tf.gfile


def _read32(bytestream):
  dt = np.dtype(np.uint32).newbyteorder('>')
  return np.frombuffer(bytestream.read(4), dtype=dt)[0]


def extract_images(f):
  """Extract the images into a 4D uint8 np array [index, y, x, depth].

  Args:
    f: A file object that can be passed into a gzip reader.

  Returns:
    data: A 4D uint8 np array [index, y, x, depth].

  Raises:
    ValueError: If the bytestream does not start with 2051.
  """
  tf.logging.info('Extracting %s', f.name)
  with gzip.GzipFile(fileobj=f) as bytestream:
    magic = _read32(bytestream)
    if magic != 2051:
      raise ValueError(
          'Invalid magic number %d in MNIST image file: %s' % (magic, f.name))
    num_images = _read32(bytestream)
    rows = _read32(bytestream)
    cols = _read32(bytestream)
    buf = bytestream.read(rows * cols * num_images)
    data = np.frombuffer(buf, dtype=np.uint8)
    data = data.reshape(num_images, rows, cols, 1)
    return data


def dense_to_one_hot(labels_dense, num_classes):
  """Convert class labels from scalars to one-hot vectors."""
  num_labels = labels_dense.shape[0]
  index_offset = np.arange(num_labels) * num_classes
  labels_one_hot = np.zeros((num_labels, num_classes))
  labels_one_hot.flat[index_offset + labels_dense.ravel()] = 1
  return labels_one_hot


def extract_labels(f, one_hot=False, num_classes=10):
  """Extract the labels into a 1D uint8 np array [index].

  Args:
    f: A file object that can be passed into a gzip reader.
    one_hot: Does one hot encoding for the result.
    num_classes: Number of classes for the one hot encoding.

  Returns:
    labels: a 1D uint8 np array.

  Raises:
    ValueError: If the bystream doesn't start with 2049.
  """
  tf.logging.info('Extracting %s', f.name)
  with gzip.GzipFile(fileobj=f) as bytestream:
    magic = _read32(bytestream)
    if magic != 2049:
      raise ValueError(
          'Invalid magic number %d in MNIST label file: %s' % (magic, f.name))
    num_items = _read32(bytestream)
    buf = bytestream.read(num_items)
    labels = np.frombuffer(buf, dtype=np.uint8)
    if one_hot:
      return dense_to_one_hot(labels, num_classes)
    return labels


class DataSet(object):
  """A dataset for MNIST."""

  def __init__(
      self,
      images,
      labels,
      fake_data=False,  # pylint:disable=unused-argument
      one_hot=False,  # pylint:disable=unused-argument
      dtype=np.float32,
      reshape=True,
      seed=None,  # pylint:disable=unused-argument
  ):
    """Construct a DataSet.

    one_hot arg is used only if fake_data is true.  `dtype` can be either
    `uint8` to leave the input as `[0, 255]`, or `float32` to rescale into
    `[0, 1]`.  Seed arg provides for convenient deterministic testing.
    """
    # If op level seed is not set, use whatever graph level seed is
    # returned.
    np.random.seed()
    assert images.shape[0] == labels.shape[0], (
        'images.shape: %s labels.shape: %s' % (images.shape, labels.shape))
    self._num_examples = images.shape[0]

    # Convert shape from [num examples, rows, columns, depth]
    # to [num examples, rows*columns] (assuming depth == 1)
    if reshape:
      assert images.shape[3] == 1
      images = images.reshape(images.shape[0],
                              images.shape[1] * images.shape[2])
    if dtype == np.float32:  # Convert from [0, 255] -> [0.0, 1.0].
      images = images.astype(np.float32)
      images = np.multiply(images, 1.0 / 255.0)
    self._images = images
    self._labels = labels
    self._epochs_completed = 0
    self._index_in_epoch = 0

  @property
  def images(self):
    return self._images

  @property
  def labels(self):
    return self._labels

  @property
  def num_examples(self):
    return self._num_examples

  @property
  def epochs_completed(self):
    return self._epochs_completed

  def next_batch(self, batch_size, fake_data=False, shuffle=True):
    """Return the next `batch_size` examples from this data set."""
    if fake_data:
      fake_image = [1] * 784
      if self.one_hot:
        fake_label = [1] + [0] * 9
      else:
        fake_label = 0
      return [fake_image for _ in range(batch_size)], [
          fake_label for _ in range(batch_size)
      ]
    start = self._index_in_epoch
    # Shuffle for the first epoch
    if self._epochs_completed == 0 and start == 0 and shuffle:
      perm0 = np.arange(self._num_examples)
      np.random.shuffle(perm0)
      self._images = self.images[perm0]
      self._labels = self.labels[perm0]
    # Go to the next epoch
    if start + batch_size > self._num_examples:
      # Finished epoch
      self._epochs_completed += 1
      # Get the rest examples in this epoch
      rest_num_examples = self._num_examples - start
      images_rest_part = self._images[start:self._num_examples]
      labels_rest_part = self._labels[start:self._num_examples]
      # Shuffle the data
      if shuffle:
        perm = np.arange(self._num_examples)
        np.random.shuffle(perm)
        self._images = self.images[perm]
        self._labels = self.labels[perm]
      # Start next epoch
      start = 0
      self._index_in_epoch = batch_size - rest_num_examples
      end = self._index_in_epoch
      images_new_part = self._images[start:end]
      labels_new_part = self._labels[start:end]
      return np.concatenate(
          (images_rest_part, images_new_part), axis=0), np.concatenate(
              (labels_rest_part, labels_new_part), axis=0)
    else:
      self._index_in_epoch += batch_size
      end = self._index_in_epoch
      return self._images[start:end], self._labels[start:end]


def read_data_sets(
    train_dir,
    fake_data=False,  # pylint:disable=unused-argument
    one_hot=False,
    dtype=np.float32,
    reshape=True,
    validation_size=5000,
    seed=None,
):
  """Read multiple datasets."""
  # pylint:disable=invalid-name
  TRAIN_IMAGES = 'train-images-idx3-ubyte.gz'
  TRAIN_LABELS = 'train-labels-idx1-ubyte.gz'
  TEST_IMAGES = 't10k-images-idx3-ubyte.gz'
  TEST_LABELS = 't10k-labels-idx1-ubyte.gz'

  local_file = os.path.join(train_dir, TRAIN_IMAGES)
  with gfile.Open(local_file, 'rb') as f:
    train_images = extract_images(f)

  local_file = os.path.join(train_dir, TRAIN_LABELS)
  with gfile.Open(local_file, 'rb') as f:
    train_labels = extract_labels(f, one_hot=one_hot)

  local_file = os.path.join(train_dir, TEST_IMAGES)
  with gfile.Open(local_file, 'rb') as f:
    test_images = extract_images(f)

  local_file = os.path.join(train_dir, TEST_LABELS)
  with gfile.Open(local_file, 'rb') as f:
    test_labels = extract_labels(f, one_hot=one_hot)

  if not 0 <= validation_size <= len(train_images):
    raise ValueError(
        'Validation size should be between 0 and {}. Received: {}.'.format(
            len(train_images), validation_size))

  validation_images = train_images[:validation_size]
  validation_labels = train_labels[:validation_size]
  train_images = train_images[validation_size:]
  train_labels = train_labels[validation_size:]

  options = dict(dtype=dtype, reshape=reshape, seed=seed)

  train = DataSet(train_images, train_labels, **options)
  validation = DataSet(validation_images, validation_labels, **options)
  test = DataSet(test_images, test_labels, **options)

  return train, validation, test


def load_mnist(train_dir='MNIST-data'):
  return read_data_sets(train_dir)