internal merge from github

PiperOrigin-RevId: 160607416

Author: Lukasz Kaiser

.gitignore

@@ -7,8 +7,3 @@
 # PyPI distribution artificats
 build/
 dist/
-
-# Sublime project files
-*.sublime-project
-*.sublime-workspace
-

tensor2tensor/data_generators/text_encoder.py

@@ -321,10 +321,10 @@ def bisect(min_val, max_val):
         other_subtokenizer = bisect(present_count + 1, max_val)
       else:
         other_subtokenizer = bisect(min_val, present_count - 1)
-      if (abs(other_subtokenizer.vocab_size - target_size) <
-          abs(subtokenizer.vocab_size - target_size)):
-        return other_subtokenizer
-      return subtokenizer
+        if (abs(other_subtokenizer.vocab_size - target_size) <
+            abs(subtokenizer.vocab_size - target_size)):
+          return other_subtokenizer
+        return subtokenizer
 
     return bisect(min_val, max_val)
 

tensor2tensor/models/common_layers.py

@@ -135,9 +135,14 @@ def image_augmentation(images, do_colors=False):
 def cifar_image_augmentation(images):
   """Image augmentation suitable for CIFAR-10/100.
 
-  As described in https://arxiv.org/pdf/1608.06993v3.pdf (page 5)."""
-  images = tf.image.resize_image_with_crop_or_pad(
-      images, 40, 40)
+  As described in https://arxiv.org/pdf/1608.06993v3.pdf (page 5).
+
+  Args:
+    images: a Tensor.
+  Returns:
+    Tensor of the same shape as images.
+  """
+  images = tf.image.resize_image_with_crop_or_pad(images, 40, 40)
   images = tf.random_crop(images, [32, 32, 3])
   images = tf.image.random_flip_left_right(images)
   return images

tensor2tensor/utils/data_reader.py

@@ -203,12 +203,10 @@ def preprocess(img):
             lambda img=inputs: resize(img))
       else:
         examples["inputs"] = tf.to_int64(resize(inputs))
-
     elif ("image_cifar10" in data_file_pattern
-        and mode == tf.contrib.learn.ModeKeys.TRAIN):
+          and mode == tf.contrib.learn.ModeKeys.TRAIN):
       examples["inputs"] = common_layers.cifar_image_augmentation(
           examples["inputs"])
-
   elif "audio" in data_file_pattern:
     # Reshape audio to proper shape
     sample_count = tf.to_int32(examples.pop("audio/sample_count"))