Migrate away from contrib.HParams to the newly forked over one.

PiperOrigin-RevId: 231998023

Author: afrozenator

README.md

@@ -383,7 +383,7 @@ and are registered with
 **Hyperparameter sets** are defined and registered in code with
 [`@registry.register_hparams`](https://github.com/tensorflow/tensor2tensor/tree/master/tensor2tensor/utils/registry.py)
 and are encoded in
-[`tf.contrib.training.HParams`](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/training/python/training/hparam.py)
+[`HParams`](https://github.com/tensorflow/tensor2tensor/tree/master/tensor2tensor/utils/hparam.py)
 objects. The `HParams` are available to both the problem specification and the
 model. A basic set of hyperparameters are defined in
 [`common_hparams.py`](https://github.com/tensorflow/tensor2tensor/tree/master/tensor2tensor/layers/common_hparams.py)

docs/walkthrough.md

@@ -383,7 +383,7 @@ and are registered with
 **Hyperparameter sets** are defined and registered in code with
 [`@registry.register_hparams`](https://github.com/tensorflow/tensor2tensor/tree/master/tensor2tensor/utils/registry.py)
 and are encoded in
-[`tf.contrib.training.HParams`](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/training/python/training/hparam.py)
+[`HParams`](https://github.com/tensorflow/tensor2tensor/tree/master/tensor2tensor/utils/hparam.py)
 objects. The `HParams` are available to both the problem specification and the
 model. A basic set of hyperparameters are defined in
 [`common_hparams.py`](https://github.com/tensorflow/tensor2tensor/tree/master/tensor2tensor/layers/common_hparams.py)

tensor2tensor/data_generators/celeba_test.py

@@ -21,6 +21,7 @@
 
 from absl.testing import parameterized
 from tensor2tensor.data_generators import celeba
+from tensor2tensor.utils.hparam import HParams
 
 import tensorflow as tf
 
@@ -34,7 +35,7 @@ class CelebaTest(parameterized.TestCase, tf.test.TestCase):
   def testCelebaMultiResolutionPreprocessExample(self, resize_method):
     example = {"inputs": tf.random_uniform([218, 178, 3], minval=-1.)}
     mode = tf.estimator.ModeKeys.TRAIN
-    hparams = tf.contrib.training.HParams(resolutions=[8, 16, 32])
+    hparams = HParams(resolutions=[8, 16, 32])
     if resize_method is not None:
       hparams.resize_method = resize_method
 

tensor2tensor/data_generators/imagenet_test.py

@@ -21,6 +21,7 @@
 
 from absl.testing import parameterized
 from tensor2tensor.data_generators import imagenet
+from tensor2tensor.utils.hparam import HParams
 
 import tensorflow as tf
 
@@ -34,7 +35,7 @@ class ImagenetTest(parameterized.TestCase, tf.test.TestCase):
   def testImagenetMultiResolutionPreprocessExample(self, resize_method):
     example = {"inputs": tf.random_uniform([64, 64, 3], minval=-1.)}
     mode = tf.estimator.ModeKeys.TRAIN
-    hparams = tf.contrib.training.HParams(resolutions=[8, 16, 32])
+    hparams = HParams(resolutions=[8, 16, 32])
     if resize_method is not None:
       hparams.resize_method = resize_method
 

tensor2tensor/data_generators/mscoco_test.py

@@ -21,6 +21,7 @@
 
 from absl.testing import parameterized
 from tensor2tensor.data_generators import mscoco
+from tensor2tensor.utils.hparam import HParams
 
 import tensorflow as tf
 
@@ -34,7 +35,7 @@ class MscocoTest(parameterized.TestCase, tf.test.TestCase):
   def testMsCocoMultiResolutionPreprocessExample(self, resize_method):
     example = {"inputs": tf.random_uniform([400, 400, 3], minval=-1.)}
     mode = tf.estimator.ModeKeys.TRAIN
-    hparams = tf.contrib.training.HParams(resolutions=[8, 16, 32])
+    hparams = HParams(resolutions=[8, 16, 32])
     if resize_method is not None:
       hparams.resize_method = resize_method
 

tensor2tensor/data_generators/problem.py

@@ -29,6 +29,7 @@
 from tensor2tensor.utils import data_reader
 from tensor2tensor.utils import metrics
 from tensor2tensor.utils import mlperf_log
+from tensor2tensor.utils.hparam import HParams
 
 import tensorflow as tf
 from tensorflow.contrib.tpu.python.tpu import tpu_config
@@ -130,7 +131,7 @@ class TaskID(object):
 
 
 def default_model_hparams():
-  return tf.contrib.training.HParams(
+  return HParams(
       max_input_seq_length=0,
       max_target_seq_length=0,
       prepend_mode="none",
@@ -596,7 +597,7 @@ def dataset(self,
       output_buffer_size: int, how many elements to prefetch at end of pipeline.
       shuffle_files: whether to shuffle input files. Default behavior (i.e. when
         shuffle_files=None) is to shuffle if mode == TRAIN.
-      hparams: tf.contrib.training.HParams; hparams to be passed to
+      hparams: HParams; hparams to be passed to
         Problem.preprocess_example and Problem.hparams. If None, will use a
         default set that is a no-op.
       preprocess: bool, whether to map the Dataset through
@@ -992,9 +993,9 @@ def _create_modalities(problem_hparams, model_hparams):
   """Creates modalities and overrides any according to model hparams.
 
   Args:
-    problem_hparams: tf.contrib.training.HParams for the Problem. It must have
+    problem_hparams: HParams for the Problem. It must have
       modality which is a dict of strings to ModalityTypes or Modality classes.
-    model_hparams: tf.contrib.training.HParams for the model. It may have
+    model_hparams: HParams for the model. It may have
       input_modalities and target_modality, which will override
       problem_hparams' modality input and target keys.
 
@@ -1024,7 +1025,7 @@ def _create_modalities(problem_hparams, model_hparams):
 
 def _default_hparams():
   """A set of basic model hyperparameters."""
-  return tf.contrib.training.HParams(
+  return HParams(
       # Use this parameter to get comparable perplexity numbers with different
       # tokenizations.  This value should be set to the ratio of the number of
       # tokens in the test set according to the tokenization used to the number

tensor2tensor/data_generators/video_utils.py

@@ -115,7 +115,7 @@ def convert_videos_to_summaries(input_videos, output_videos, target_videos,
     output_videos: 5-D NumPy array, (NTHWC) model predictions.
     target_videos: 5-D NumPy array, (NTHWC) target frames.
     tag: tf summary tag.
-    decode_hparams: tf.contrib.training.HParams.
+    decode_hparams: HParams.
     display_ground_truth: Whether or not to display ground truth videos.
   Returns:
     summaries: a list of tf frame-by-frame and video summaries.
@@ -274,7 +274,7 @@ def max_frames_per_video(self, hparams):
     hparams.video_num_target_frames.
 
     Args:
-      hparams: tf.contrib.training.HParams.
+      hparams: HParams.
     Returns:
       num_frames: int.
     """

tensor2tensor/layers/common_hparams.py

@@ -20,14 +20,15 @@
 from __future__ import print_function
 from six.moves import zip  # pylint: disable=redefined-builtin
 from tensor2tensor.utils import registry
+from tensor2tensor.utils.hparam import HParams
 
 import tensorflow as tf
 
 
 @registry.register_hparams("basic_1")
 def basic_params1():
   """A set of basic hyperparameters."""
-  return tf.contrib.training.HParams(
+  return HParams(
       # If the problem consists of variable-length sequences
       # (see problem.batch_size_means_tokens()), then this is the number
       # of tokens per batch per GPU or per TPU core.  Otherwise, this is

tensor2tensor/layers/common_image_attention.py

@@ -477,7 +477,7 @@ def postprocess_image(x, rows, cols, hparams):
       number of elements in x is batch * rows * cols * hparams.hidden_size.
     rows: Integer representing number of rows in a 2-D data point.
     cols: Integer representing number of columns in a 2-D data point.
-    hparams: tf.contrib.training.HParams set.
+    hparams: HParams set.
 
   Returns:
     Tensor of shape [batch, rows, cols, depth], where depth is
@@ -639,7 +639,7 @@ def create_output(decoder_output, rows, cols, targets, hparams):
     cols: Integer representing number of columns in a 2-D data point.
     targets: Tensor of shape [batch, hparams.img_len, hparams.img_len,
       hparams.num_channels].
-    hparams: tf.contrib.training.HParams set.
+    hparams: HParams set.
 
   Returns:
     Tensor of shape [batch, hparams.img_len, hparams.img_len,

tensor2tensor/layers/common_image_attention_test.py

@@ -22,6 +22,7 @@
 from absl.testing import parameterized
 from tensor2tensor.layers import common_hparams
 from tensor2tensor.layers import common_image_attention
+from tensor2tensor.utils.hparam import HParams
 
 import tensorflow as tf
 
@@ -36,7 +37,7 @@ def testPostProcessImageTrainMode(self, likelihood, num_mixtures, depth):
     batch = 1
     rows = 8
     cols = 24
-    hparams = tf.contrib.training.HParams(
+    hparams = HParams(
         hidden_size=2,
         likelihood=likelihood,
         mode=tf.estimator.ModeKeys.TRAIN,
@@ -58,7 +59,7 @@ def testPostProcessImageInferMode(self, likelihood, num_mixtures, depth):
     cols = 24
     block_length = 4
     block_width = 2
-    hparams = tf.contrib.training.HParams(
+    hparams = HParams(
         block_raster_scan=True,
         hidden_size=2,
         likelihood=likelihood,
@@ -90,7 +91,7 @@ def testCreateOutputTrainMode(self, likelihood, num_mixtures, depth):
       cols = channels * width
     else:
       cols = width
-    hparams = tf.contrib.training.HParams(
+    hparams = HParams(
         hidden_size=2,
         likelihood=likelihood,
         mode=tf.estimator.ModeKeys.TRAIN,