tf.compat.v1.logging implemented with absl

official/benchmark/models/resnet_cifar_main.py

@@ -20,6 +20,7 @@
 
 from absl import app
 from absl import flags
+from absl import logging
 import numpy as np
 import tensorflow as tf
 from official.benchmark.models import resnet_cifar_model
@@ -100,7 +101,7 @@ def on_batch_begin(self, batch, logs=None):
     if lr != self.prev_lr:
       self.model.optimizer.learning_rate = lr  # lr should be a float here
       self.prev_lr = lr
-      tf.compat.v1.logging.debug(
+      logging.debug(
           'Epoch %05d Batch %05d: LearningRateBatchScheduler '
           'change learning rate to %s.', self.epochs, batch, lr)
 
@@ -280,6 +281,6 @@ def main(_):
 
 
 if __name__ == '__main__':
-  tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.INFO)
+  logging.set_verbosity(logging.INFO)
   define_cifar_flags()
   app.run(main)

official/benchmark/ncf_keras_benchmark.py

@@ -23,6 +23,7 @@
 
 from absl import flags
 from absl.testing import flagsaver
+from absl import logging
 import tensorflow as tf
 
 from official.recommendation import ncf_common
@@ -51,7 +52,7 @@ def __init__(self,
   def _setup(self):
     """Sets up and resets flags before each test."""
     assert tf.version.VERSION.startswith('2.')
-    tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.INFO)
+    logging.set_verbosity(logging.INFO)
     if NCFKerasBenchmarkBase.local_flags is None:
       ncf_common.define_ncf_flags()
       # Loads flags to get defaults to then override. List cannot be empty.

official/modeling/model_training_utils_test.py

@@ -22,6 +22,7 @@
 
 from absl.testing import parameterized
 from absl.testing.absltest import mock
+from absl import logging
 import numpy as np
 import tensorflow as tf
 
@@ -125,7 +126,7 @@ def summaries_with_matching_keyword(keyword, summary_dir):
     if event.summary is not None:
       for value in event.summary.value:
         if keyword in value.tag:
-          tf.compat.v1.logging.error(event)
+          logging.error(event)
           yield event.summary
 
 

official/nlp/transformer/translate.py

@@ -20,7 +20,7 @@
 
 import numpy as np
 import tensorflow as tf
-
+from absl import logging
 from official.nlp.transformer.utils import tokenizer
 
 _EXTRA_DECODE_LENGTH = 100
@@ -117,7 +117,7 @@ def input_generator():
           maxlen=params["decode_max_length"],
           dtype="int32",
           padding="post")
-      tf.compat.v1.logging.info("Decoding batch %d out of %d.", i,
+      logging.info("Decoding batch %d out of %d.", i,
                                 num_decode_batches)
       yield batch
 
@@ -172,7 +172,7 @@ def text_as_per_replica():
         translation = _trim_and_decode(val_outputs[j], subtokenizer)
         translations.append(translation)
         if print_all_translations:
-          tf.compat.v1.logging.info(
+          logging.info(
               "Translating:\n\tInput: %s\n\tOutput: %s" %
               (sorted_inputs[j + i * batch_size], translation))
 
@@ -181,7 +181,7 @@ def text_as_per_replica():
     if tf.io.gfile.isdir(output_file):
       raise ValueError("File output is a directory, will not save outputs to "
                        "file.")
-    tf.compat.v1.logging.info("Writing to file %s" % output_file)
+    logging.info("Writing to file %s" % output_file)
     with tf.compat.v1.gfile.Open(output_file, "w") as f:
       for i in sorted_keys:
         f.write("%s\n" % translations[i])
@@ -191,10 +191,10 @@ def translate_from_text(model, subtokenizer, txt):
   encoded_txt = _encode_and_add_eos(txt, subtokenizer)
   result = model.predict(encoded_txt)
   outputs = result["outputs"]
-  tf.compat.v1.logging.info("Original: \"%s\"" % txt)
+  logging.info("Original: \"%s\"" % txt)
   translate_from_input(outputs, subtokenizer)
 
 
 def translate_from_input(outputs, subtokenizer):
   translation = _trim_and_decode(outputs, subtokenizer)
-  tf.compat.v1.logging.info("Translation: \"%s\"" % translation)
+  logging.info("Translation: \"%s\"" % translation)

official/nlp/transformer/utils/tokenizer.py

@@ -22,6 +22,7 @@
 import re
 import sys
 import unicodedata
+from absl import logging
 
 import numpy as np
 import six
@@ -63,7 +64,7 @@ class Subtokenizer(object):
 
   def __init__(self, vocab_file, reserved_tokens=None):
     """Initializes class, creating a vocab file if data_files is provided."""
-    tf.compat.v1.logging.info("Initializing Subtokenizer from file %s." %
+    logging.info("Initializing Subtokenizer from file %s." %
                               vocab_file)
 
     if reserved_tokens is None:
@@ -109,15 +110,15 @@ def init_from_files(
       reserved_tokens = RESERVED_TOKENS
 
     if tf.io.gfile.exists(vocab_file):
-      tf.compat.v1.logging.info("Vocab file already exists (%s)" % vocab_file)
+      logging.info("Vocab file already exists (%s)" % vocab_file)
     else:
-      tf.compat.v1.logging.info("Begin steps to create subtoken vocabulary...")
+      logging.info("Begin steps to create subtoken vocabulary...")
       token_counts = _count_tokens(files, file_byte_limit, correct_strip)
       alphabet = _generate_alphabet_dict(token_counts)
       subtoken_list = _generate_subtokens_with_target_vocab_size(
           token_counts, alphabet, target_vocab_size, threshold, min_count,
           reserved_tokens)
-      tf.compat.v1.logging.info("Generated vocabulary with %d subtokens." %
+      logging.info("Generated vocabulary with %d subtokens." %
                                 len(subtoken_list))
       _save_vocab_file(vocab_file, subtoken_list)
     return Subtokenizer(vocab_file)
@@ -402,7 +403,7 @@ def _generate_subtokens_with_target_vocab_size(
     reserved_tokens = RESERVED_TOKENS
 
   if min_count is not None:
-    tf.compat.v1.logging.info(
+    logging.info(
         "Using min_count=%d to generate vocab with target size %d" %
         (min_count, target_size))
     return _generate_subtokens(
@@ -411,13 +412,13 @@ def _generate_subtokens_with_target_vocab_size(
   def bisect(min_val, max_val):
     """Recursive function to binary search for subtoken vocabulary."""
     cur_count = (min_val + max_val) // 2
-    tf.compat.v1.logging.info("Binary search: trying min_count=%d (%d %d)" %
+    logging.info("Binary search: trying min_count=%d (%d %d)" %
                               (cur_count, min_val, max_val))
     subtoken_list = _generate_subtokens(
         token_counts, alphabet, cur_count, reserved_tokens=reserved_tokens)
 
     val = len(subtoken_list)
-    tf.compat.v1.logging.info(
+    logging.info(
         "Binary search: min_count=%d resulted in %d tokens" % (cur_count, val))
 
     within_threshold = abs(val - target_size) < threshold
@@ -434,7 +435,7 @@ def bisect(min_val, max_val):
       return other_subtoken_list
     return subtoken_list
 
-  tf.compat.v1.logging.info("Finding best min_count to get target size of %d" %
+  logging.info("Finding best min_count to get target size of %d" %
                             target_size)
   return bisect(_MIN_MIN_COUNT, _MAX_MIN_COUNT)
 
@@ -603,7 +604,7 @@ def _generate_subtokens(
   # subtoken_dict, count how often the resulting subtokens appear, and update
   # the dictionary with subtokens w/ high enough counts.
   for i in xrange(num_iterations):
-    tf.compat.v1.logging.info("\tGenerating subtokens: iteration %d" % i)
+    logging.info("\tGenerating subtokens: iteration %d" % i)
     # Generate new subtoken->id dictionary using the new subtoken list.
     subtoken_dict = _list_to_index_dict(subtoken_list)
 
@@ -616,5 +617,5 @@ def _generate_subtokens(
     subtoken_list, max_subtoken_length = _gen_new_subtoken_list(
         subtoken_counts, min_count, alphabet, reserved_tokens)
 
-    tf.compat.v1.logging.info("\tVocab size: %d" % len(subtoken_list))
+    logging.info("\tVocab size: %d" % len(subtoken_list))
   return subtoken_list

official/r1/mnist/mnist.py

@@ -17,6 +17,7 @@
 from __future__ import division
 from __future__ import print_function
 
+from absl import logging
 from absl import app as absl_app
 from absl import flags
 from six.moves import range
@@ -243,6 +244,6 @@ def main(_):
 
 
 if __name__ == '__main__':
-  tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.INFO)
+  logging.set_verbosity(logging.INFO)
   define_mnist_flags()
   absl_app.run(main)

official/r1/mnist/mnist_test.py

@@ -21,7 +21,7 @@
 import unittest
 
 import tensorflow as tf  # pylint: disable=g-bad-import-order
-
+from absl import logging
 from official.r1.mnist import mnist
 from official.utils.misc import keras_utils
 
@@ -143,5 +143,5 @@ def benchmark_train_step_time(self):
 
 
 if __name__ == '__main__':
-  tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)
+  logging.set_verbosity(logging.ERROR)
   tf.test.main()

official/r1/resnet/cifar10_main.py

@@ -20,6 +20,7 @@
 
 import os
 
+from absl import logging
 from absl import app as absl_app
 from absl import flags
 from six.moves import range
@@ -139,7 +140,7 @@ def input_fn(is_training,
   dataset = tf.data.FixedLengthRecordDataset(filenames, _RECORD_BYTES)
 
   if input_context:
-    tf.compat.v1.logging.info(
+    logging.info(
         'Sharding the dataset: input_pipeline_id=%d num_input_pipelines=%d' % (
             input_context.input_pipeline_id, input_context.num_input_pipelines))
     dataset = dataset.shard(input_context.num_input_pipelines,
@@ -270,7 +271,7 @@ def run_cifar(flags_obj):
     Dictionary of results. Including final accuracy.
   """
   if flags_obj.image_bytes_as_serving_input:
-    tf.compat.v1.logging.fatal(
+    logging.fatal(
         '--image_bytes_as_serving_input cannot be set to True for CIFAR. '
         'This flag is only applicable to ImageNet.')
     return
@@ -291,6 +292,6 @@ def main(_):
 
 
 if __name__ == '__main__':
-  tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.INFO)
+  logging.set_verbosity(logging.INFO)
   define_cifar_flags()
   absl_app.run(main)

official/r1/resnet/cifar10_test.py

@@ -21,12 +21,13 @@
 
 import numpy as np
 import tensorflow as tf  # pylint: disable=g-bad-import-order
+from absl import logging
 
 from official.r1.resnet import cifar10_main
 from official.utils.misc import keras_utils
 from official.utils.testing import integration
 
-tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)
+logging.set_verbosity(logging.ERROR)
 
 _BATCH_SIZE = 128
 _HEIGHT = 32

official/r1/resnet/estimator_benchmark.py

@@ -21,6 +21,7 @@
 import time
 
 from absl import flags
+from absl import logging
 from absl.testing import flagsaver
 import tensorflow as tf  # pylint: disable=g-bad-import-order
 
@@ -56,7 +57,7 @@ def _get_model_dir(self, folder_name):
 
   def _setup(self):
     """Sets up and resets flags before each test."""
-    tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.INFO)
+    logging.set_verbosity(logging.INFO)
     if EstimatorBenchmark.local_flags is None:
       for flag_method in self.flag_methods:
         flag_method()

official/r1/resnet/imagenet_main.py

@@ -22,6 +22,7 @@
 
 from absl import app as absl_app
 from absl import flags
+from absl import logging
 from six.moves import range
 import tensorflow as tf
 
@@ -194,7 +195,7 @@ def input_fn(is_training,
   dataset = tf.data.Dataset.from_tensor_slices(filenames)
 
   if input_context:
-    tf.compat.v1.logging.info(
+    logging.info(
         'Sharding the dataset: input_pipeline_id=%d num_input_pipelines=%d' % (
             input_context.input_pipeline_id, input_context.num_input_pipelines))
     dataset = dataset.shard(input_context.num_input_pipelines,
@@ -387,6 +388,6 @@ def main(_):
 
 
 if __name__ == '__main__':
-  tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.INFO)
+  logging.set_verbosity(logging.INFO)
   define_imagenet_flags()
   absl_app.run(main)

official/r1/resnet/imagenet_test.py

@@ -25,7 +25,7 @@
 from official.utils.misc import keras_utils
 from official.utils.testing import integration
 
-tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)
+logging.set_verbosity(logging.ERROR)
 
 _BATCH_SIZE = 32
 _LABEL_CLASSES = 1001

official/r1/resnet/resnet_run_loop.py

@@ -28,6 +28,7 @@
 import multiprocessing
 import os
 
+from absl import logging
 from absl import flags
 import tensorflow as tf
 
@@ -83,7 +84,7 @@ def process_record_dataset(dataset,
     options.experimental_threading.private_threadpool_size = (
         datasets_num_private_threads)
     dataset = dataset.with_options(options)
-    tf.compat.v1.logging.info('datasets_num_private_threads: %s',
+    logging.info('datasets_num_private_threads: %s',
                               datasets_num_private_threads)
 
   # Disable intra-op parallelism to optimize for throughput instead of latency.
@@ -205,16 +206,16 @@ def override_flags_and_set_envars_for_gpu_thread_pool(flags_obj):
     what has been set by the user on the command-line.
   """
   cpu_count = multiprocessing.cpu_count()
-  tf.compat.v1.logging.info('Logical CPU cores: %s', cpu_count)
+  logging.info('Logical CPU cores: %s', cpu_count)
 
   # Sets up thread pool for each GPU for op scheduling.
   per_gpu_thread_count = 1
   total_gpu_thread_count = per_gpu_thread_count * flags_obj.num_gpus
   os.environ['TF_GPU_THREAD_MODE'] = flags_obj.tf_gpu_thread_mode
   os.environ['TF_GPU_THREAD_COUNT'] = str(per_gpu_thread_count)
-  tf.compat.v1.logging.info('TF_GPU_THREAD_COUNT: %s',
+  logging.info('TF_GPU_THREAD_COUNT: %s',
                             os.environ['TF_GPU_THREAD_COUNT'])
-  tf.compat.v1.logging.info('TF_GPU_THREAD_MODE: %s',
+  logging.info('TF_GPU_THREAD_MODE: %s',
                             os.environ['TF_GPU_THREAD_MODE'])
 
   # Reduces general thread pool by number of threads used for GPU pool.
@@ -648,7 +649,7 @@ def input_fn_eval():
         hooks=train_hooks,
         max_steps=flags_obj.max_train_steps)
     eval_spec = tf.estimator.EvalSpec(input_fn=input_fn_eval)
-    tf.compat.v1.logging.info('Starting to train and evaluate.')
+    logging.info('Starting to train and evaluate.')
     tf.estimator.train_and_evaluate(classifier, train_spec, eval_spec)
     # tf.estimator.train_and_evalute doesn't return anything in multi-worker
     # case.
@@ -671,7 +672,7 @@ def input_fn_eval():
       schedule[-1] = train_epochs - sum(schedule[:-1])  # over counting.
 
     for cycle_index, num_train_epochs in enumerate(schedule):
-      tf.compat.v1.logging.info('Starting cycle: %d/%d', cycle_index,
+      logging.info('Starting cycle: %d/%d', cycle_index,
                                 int(n_loops))
 
       if num_train_epochs:
@@ -691,7 +692,7 @@ def input_fn_eval():
       # allows the eval (which is generally unimportant in those circumstances)
       # to terminate.  Note that eval will run for max_train_steps each loop,
       # regardless of the global_step count.
-      tf.compat.v1.logging.info('Starting to evaluate.')
+      logging.info('Starting to evaluate.')
       eval_results = classifier.evaluate(input_fn=input_fn_eval,
                                          steps=flags_obj.max_train_steps)