Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

tf_upgrade_v2 on resnet and utils folders. #6154

Merged
merged 15 commits into from
Feb 5, 2019
Merged

tf_upgrade_v2 on resnet and utils folders. #6154

Show file tree
Hide file tree
Changes from 13 commits
Commits
Show all changes
15 commits
Select commit Hold shift + click to select a range
4b921e5
Add resnet56 short tests. (#6101)
tfboyd Jan 31, 2019
ce73f81
Use core mirrored strategy in official models (#6126)
guptapriya Feb 1, 2019
39b1273
Imagenet short tests (#6132)
tfboyd Feb 1, 2019
3ab2869
points default data_dir to child folder. (#6131)
tfboyd Feb 1, 2019
c4ece8a
Imagenet short tests (#6136)
tfboyd Feb 1, 2019
17a777d
Flag (#6121)
seemuch Feb 1, 2019
e57a082
Export benchmark stats using tf.test.Benchmark.report_benchmark() (#6…
lindong28 Feb 1, 2019
9a4f074
Typos. (#6120)
dynamicwebpaige Feb 2, 2019
d1f5d1a
log verbosity=2 logs every epoch no progress bars (#6142)
tfboyd Feb 3, 2019
8b18b5b
tf_upgrade_v2 on resnet and utils folder.
goldiegadde Feb 4, 2019
69ce0a2
Merge branch 'master' into gadde-v2-convert
goldiegadde Feb 4, 2019
682736c
tf_upgrade_v2 on resnet and utils folder.
goldiegadde Feb 4, 2019
020e374
Merge branch 'gadde-v2-convert' of github.com:goldiegadde/models into…
goldiegadde Feb 5, 2019
bab39b4
Merge branch 'gadde-v2-convert' of github.com:goldiegadde/models into…
goldiegadde Feb 5, 2019
9ca3ce7
Merge branch 'gadde-v2-convert' of github.com:goldiegadde/models into…
goldiegadde Feb 5, 2019
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion official/resnet/cifar10_download_and_extract.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,4 +60,4 @@ def _progress(count, block_size, total_size):

if __name__ == '__main__':
FLAGS, unparsed = parser.parse_known_args()
tf.app.run(argv=[sys.argv[0]] + unparsed)
tf.compat.v1.app.run(argv=[sys.argv[0]] + unparsed)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

can't we use absl`s app.run here? (simiar to cifar10_main.py)?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We can ask and then see if they can fix it. We can also manually change it.

15 changes: 8 additions & 7 deletions official/resnet/cifar10_main.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,7 +52,7 @@
###############################################################################
def get_filenames(is_training, data_dir):
"""Returns a list of filenames."""
assert tf.gfile.Exists(data_dir), (
assert tf.io.gfile.exists(data_dir), (
'Run cifar10_download_and_extract.py first to download and extract the '
'CIFAR-10 data.')

Expand All @@ -68,7 +68,7 @@ def get_filenames(is_training, data_dir):
def parse_record(raw_record, is_training, dtype):
"""Parse CIFAR-10 image and label from a raw record."""
# Convert bytes to a vector of uint8 that is record_bytes long.
record_vector = tf.decode_raw(raw_record, tf.uint8)
record_vector = tf.io.decode_raw(raw_record, tf.uint8)

# The first byte represents the label, which we convert from uint8 to int32
# and then to one-hot.
Expand All @@ -81,7 +81,7 @@ def parse_record(raw_record, is_training, dtype):

# Convert from [depth, height, width] to [height, width, depth], and cast as
# float32.
image = tf.cast(tf.transpose(depth_major, [1, 2, 0]), tf.float32)
image = tf.cast(tf.transpose(a=depth_major, perm=[1, 2, 0]), tf.float32)

image = preprocess_image(image, is_training)
image = tf.cast(image, dtype)
Expand All @@ -97,7 +97,7 @@ def preprocess_image(image, is_training):
image, HEIGHT + 8, WIDTH + 8)

# Randomly crop a [HEIGHT, WIDTH] section of the image.
image = tf.random_crop(image, [HEIGHT, WIDTH, NUM_CHANNELS])
image = tf.image.random_crop(image, [HEIGHT, WIDTH, NUM_CHANNELS])

# Randomly flip the image horizontally.
image = tf.image.random_flip_left_right(image)
Expand Down Expand Up @@ -253,8 +253,9 @@ def run_cifar(flags_obj):
Dictionary of results. Including final accuracy.
"""
if flags_obj.image_bytes_as_serving_input:
tf.logging.fatal('--image_bytes_as_serving_input cannot be set to True '
'for CIFAR. This flag is only applicable to ImageNet.')
tf.compat.v1.logging.fatal(
'--image_bytes_as_serving_input cannot be set to True for CIFAR. '
'This flag is only applicable to ImageNet.')
return

input_function = (flags_obj.use_synthetic_data and
Expand All @@ -273,6 +274,6 @@ def main(_):


if __name__ == '__main__':
tf.logging.set_verbosity(tf.logging.INFO)
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.INFO)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was checking the docs and I think we are supposed to use absl for logging in 2.0?
https://github.com/tensorflow/docs/blob/master/site/en/r2/guide/effective_tf2.md#api-cleanup

Looks like the upgrade script doesn't do this automatically. Can we do this in a subsequent PR perhaps?

from absl import logging

logging.info(...)
logging.error(...)

https://abseil.io/docs/python/guides/logging

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, this is on our list. I was shocked that no logging moves forward in TF 2.0 without a rewrite.

define_cifar_flags()
absl_app.run(main)
11 changes: 6 additions & 5 deletions official/resnet/cifar10_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,7 @@
from official.resnet import cifar10_main
from official.utils.testing import integration

tf.logging.set_verbosity(tf.logging.ERROR)
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)

_BATCH_SIZE = 128
_HEIGHT = 32
Expand All @@ -44,7 +44,7 @@ def setUpClass(cls): # pylint: disable=invalid-name

def tearDown(self):
super(BaseTest, self).tearDown()
tf.gfile.DeleteRecursively(self.get_temp_dir())
tf.io.gfile.rmtree(self.get_temp_dir())

def test_dataset_input_fn(self):
fake_data = bytearray()
Expand All @@ -62,7 +62,8 @@ def test_dataset_input_fn(self):
filename, cifar10_main._RECORD_BYTES) # pylint: disable=protected-access
fake_dataset = fake_dataset.map(
lambda val: cifar10_main.parse_record(val, False, tf.float32))
image, label = fake_dataset.make_one_shot_iterator().get_next()
image, label = tf.compat.v1.data.make_one_shot_iterator(
fake_dataset).get_next()

self.assertAllEqual(label.shape, ())
self.assertAllEqual(image.shape, (_HEIGHT, _WIDTH, _NUM_CHANNELS))
Expand All @@ -79,7 +80,7 @@ def test_dataset_input_fn(self):
def cifar10_model_fn_helper(self, mode, resnet_version, dtype):
input_fn = cifar10_main.get_synth_input_fn(dtype)
dataset = input_fn(True, '', _BATCH_SIZE)
iterator = dataset.make_initializable_iterator()
iterator = tf.compat.v1.data.make_initializable_iterator(dataset)
features, labels = iterator.get_next()
spec = cifar10_main.cifar10_model_fn(
features, labels, mode, {
Expand Down Expand Up @@ -142,7 +143,7 @@ def _test_cifar10model_shape(self, resnet_version):
model = cifar10_main.Cifar10Model(32, data_format='channels_last',
num_classes=num_classes,
resnet_version=resnet_version)
fake_input = tf.random_uniform([batch_size, _HEIGHT, _WIDTH, _NUM_CHANNELS])
fake_input = tf.random.uniform([batch_size, _HEIGHT, _WIDTH, _NUM_CHANNELS])
output = model(fake_input, training=True)

self.assertAllEqual(output.shape, (batch_size, num_classes))
Expand Down
2 changes: 1 addition & 1 deletion official/resnet/estimator_cifar_benchmark.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -144,7 +144,7 @@ def _get_model_dir(self, folder_name):
return os.path.join(self.output_dir, folder_name)

def _setup(self):
tf.logging.set_verbosity(tf.logging.DEBUG)
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.DEBUG)
if EstimatorCifar10BenchmarkTests.local_flags is None:
cifar_main.define_cifar_flags()
# Loads flags to get defaults to then override.
Expand Down
21 changes: 11 additions & 10 deletions official/resnet/imagenet_main.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -95,22 +95,23 @@ def _parse_example_proto(example_serialized):
"""
# Dense features in Example proto.
feature_map = {
'image/encoded': tf.FixedLenFeature([], dtype=tf.string,
default_value=''),
'image/class/label': tf.FixedLenFeature([], dtype=tf.int64,
default_value=-1),
'image/class/text': tf.FixedLenFeature([], dtype=tf.string,
'image/encoded': tf.io.FixedLenFeature([], dtype=tf.string,
default_value=''),
'image/class/label': tf.io.FixedLenFeature([], dtype=tf.int64,
default_value=-1),
'image/class/text': tf.io.FixedLenFeature([], dtype=tf.string,
default_value=''),
}
sparse_float32 = tf.VarLenFeature(dtype=tf.float32)
sparse_float32 = tf.io.VarLenFeature(dtype=tf.float32)
# Sparse features in Example proto.
feature_map.update(
{k: sparse_float32 for k in ['image/object/bbox/xmin',
'image/object/bbox/ymin',
'image/object/bbox/xmax',
'image/object/bbox/ymax']})

features = tf.parse_single_example(example_serialized, feature_map)
features = tf.io.parse_single_example(serialized=example_serialized,
features=feature_map)
label = tf.cast(features['image/class/label'], dtype=tf.int32)

xmin = tf.expand_dims(features['image/object/bbox/xmin'].values, 0)
Expand All @@ -124,7 +125,7 @@ def _parse_example_proto(example_serialized):
# Force the variable number of bounding boxes into the shape
# [1, num_boxes, coords].
bbox = tf.expand_dims(bbox, 0)
bbox = tf.transpose(bbox, [0, 2, 1])
bbox = tf.transpose(a=bbox, perm=[0, 2, 1])

return features['image/encoded'], label, bbox

Expand Down Expand Up @@ -188,7 +189,7 @@ def input_fn(is_training, data_dir, batch_size, num_epochs=1,
# This number is low enough to not cause too much contention on small systems
# but high enough to provide the benefits of parallelization. You may want
# to increase this number if you have a large number of CPU cores.
dataset = dataset.apply(tf.contrib.data.parallel_interleave(
dataset = dataset.apply(tf.data.experimental.parallel_interleave(
tf.data.TFRecordDataset, cycle_length=10))

return resnet_run_loop.process_record_dataset(
Expand Down Expand Up @@ -352,6 +353,6 @@ def main(_):


if __name__ == '__main__':
tf.logging.set_verbosity(tf.logging.INFO)
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.INFO)
define_imagenet_flags()
absl_app.run(main)
6 changes: 3 additions & 3 deletions official/resnet/imagenet_preprocessing.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -108,7 +108,7 @@ def _central_crop(image, crop_height, crop_width):
Returns:
3-D tensor with cropped image.
"""
shape = tf.shape(image)
shape = tf.shape(input=image)
height, width = shape[0], shape[1]

amount_to_be_cropped_h = (height - crop_height)
Expand Down Expand Up @@ -195,7 +195,7 @@ def _aspect_preserving_resize(image, resize_min):
Returns:
resized_image: A 3-D tensor containing the resized image.
"""
shape = tf.shape(image)
shape = tf.shape(input=image)
height, width = shape[0], shape[1]

new_height, new_width = _smallest_size_at_least(height, width, resize_min)
Expand All @@ -218,7 +218,7 @@ def _resize_image(image, height, width):
resized_image: A 3-D tensor containing the resized image. The first two
dimensions have the shape [height, width].
"""
return tf.image.resize_images(
return tf.image.resize(
image, [height, width], method=tf.image.ResizeMethod.BILINEAR,
align_corners=False)

Expand Down
12 changes: 6 additions & 6 deletions official/resnet/imagenet_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,7 +24,7 @@
from official.resnet import imagenet_main
from official.utils.testing import integration

tf.logging.set_verbosity(tf.logging.ERROR)
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)

_BATCH_SIZE = 32
_LABEL_CLASSES = 1001
Expand All @@ -39,7 +39,7 @@ def setUpClass(cls): # pylint: disable=invalid-name

def tearDown(self):
super(BaseTest, self).tearDown()
tf.gfile.DeleteRecursively(self.get_temp_dir())
tf.io.gfile.rmtree(self.get_temp_dir())

def _tensor_shapes_helper(self, resnet_size, resnet_version, dtype, with_gpu):
"""Checks the tensor shapes after each phase of the ResNet model."""
Expand All @@ -62,7 +62,7 @@ def reshape(shape):
resnet_version=resnet_version,
dtype=dtype
)
inputs = tf.random_uniform([1, 224, 224, 3])
inputs = tf.random.uniform([1, 224, 224, 3])
output = model(inputs, training=True)

initial_conv = graph.get_tensor_by_name('resnet_model/initial_conv:0')
Expand Down Expand Up @@ -189,11 +189,11 @@ def test_tensor_shapes_resnet_200_with_gpu_v2(self):

def resnet_model_fn_helper(self, mode, resnet_version, dtype):
"""Tests that the EstimatorSpec is given the appropriate arguments."""
tf.train.create_global_step()
tf.compat.v1.train.create_global_step()

input_fn = imagenet_main.get_synth_input_fn(dtype)
dataset = input_fn(True, '', _BATCH_SIZE)
iterator = dataset.make_initializable_iterator()
iterator = tf.compat.v1.data.make_initializable_iterator(dataset)
features, labels = iterator.get_next()
spec = imagenet_main.imagenet_model_fn(
features, labels, mode, {
Expand Down Expand Up @@ -257,7 +257,7 @@ def _test_imagenetmodel_shape(self, resnet_version):
50, data_format='channels_last', num_classes=num_classes,
resnet_version=resnet_version)

fake_input = tf.random_uniform([batch_size, 224, 224, 3])
fake_input = tf.random.uniform([batch_size, 224, 224, 3])
output = model(fake_input, training=True)

self.assertAllEqual(output.shape, (batch_size, num_classes))
Expand Down
2 changes: 1 addition & 1 deletion official/resnet/keras/keras_benchmark.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,7 +43,7 @@ def _get_model_dir(self, folder_name):

def _setup(self):
"""Sets up and resets flags before each test."""
tf.logging.set_verbosity(tf.logging.DEBUG)
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.DEBUG)
if KerasBenchmark.local_flags is None:
for flag_method in self.flag_methods:
flag_method()
Expand Down
6 changes: 3 additions & 3 deletions official/resnet/keras/keras_cifar_main.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -81,7 +81,7 @@ def parse_record_keras(raw_record, is_training, dtype):
Tuple with processed image tensor and one-hot-encoded label tensor.
"""
image, label = cifar_main.parse_record(raw_record, is_training, dtype)
label = tf.sparse_to_dense(label, (cifar_main.NUM_CLASSES,), 1)
label = tf.compat.v1.sparse_to_dense(label, (cifar_main.NUM_CLASSES,), 1)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

can you file a ticket to fix this? We should not need to do this anyway and it might be better to just figure out the right way to do this.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We are going to go back and file tickets on anything compat.v1 in the keras code path.

return image, label


Expand All @@ -98,7 +98,7 @@ def run(flags_obj):
Dictionary of training and eval stats.
"""
if flags_obj.enable_eager:
tf.enable_eager_execution()
tf.compat.v1.enable_eager_execution()
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

i think maybe we should also add an else clause here for the opposite case so it can work in 2.0 too?

else:
  tf.compat.v1.disable_eager_execution()

wdyt?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We will make a 2.0 branch as some things are in 2.0 that cannot be done in 1.0 like saving a model in 2.0 format, the call just does not exist.


dtype = flags_core.get_tf_dtype(flags_obj)
if dtype == 'fp16':
Expand Down Expand Up @@ -194,7 +194,7 @@ def main(_):


if __name__ == '__main__':
tf.logging.set_verbosity(tf.logging.INFO)
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.INFO)
cifar_main.define_cifar_flags()
keras_common.define_keras_flags()
absl_app.run(main)
30 changes: 14 additions & 16 deletions official/resnet/keras/keras_common.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -80,7 +80,7 @@ def on_batch_end(self, batch, logs=None):
if batch != 0:
self.record_batch = True
self.timestamp_log.append(BatchTimestamp(batch, timestamp))
tf.logging.info("BenchmarkMetric: {'num_batches':%d, 'time_taken': %f,"
tf.compat.v1.logging.info("BenchmarkMetric: {'num_batches':%d, 'time_taken': %f,"
"'images_per_second': %f}" %
(batch, elapsed_time, examples_per_second))

Expand Down Expand Up @@ -120,7 +120,7 @@ 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.logging.debug('Epoch %05d Batch %05d: LearningRateBatchScheduler '
tf.compat.v1.logging.debug('Epoch %05d Batch %05d: LearningRateBatchScheduler '
'change learning rate to %s.', self.epochs, batch, lr)


Expand Down Expand Up @@ -226,22 +226,20 @@ def get_synth_input_fn(height, width, num_channels, num_classes,
def input_fn(is_training, data_dir, batch_size, *args, **kwargs):
"""Returns dataset filled with random data."""
# Synthetic input should be within [0, 255].
inputs = tf.truncated_normal(
[height, width, num_channels],
dtype=dtype,
mean=127,
stddev=60,
name='synthetic_inputs')

labels = tf.random_uniform(
[1],
minval=0,
maxval=num_classes - 1,
dtype=tf.int32,
name='synthetic_labels')
inputs = tf.random.truncated_normal([height, width, num_channels],
dtype=dtype,
mean=127,
stddev=60,
name='synthetic_inputs')

labels = tf.random.uniform([1],
minval=0,
maxval=num_classes - 1,
dtype=tf.int32,
name='synthetic_labels')
data = tf.data.Dataset.from_tensors((inputs, labels)).repeat()
data = data.batch(batch_size)
data = data.prefetch(buffer_size=tf.contrib.data.AUTOTUNE)
data = data.prefetch(buffer_size=tf.data.experimental.AUTOTUNE)
return data

return input_fn
Expand Down
2 changes: 1 addition & 1 deletion official/resnet/keras/keras_common_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,7 @@

from official.resnet.keras import keras_common

tf.logging.set_verbosity(tf.logging.ERROR)
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)


class KerasCommonTests(tf.test.TestCase):
Expand Down
4 changes: 2 additions & 2 deletions official/resnet/keras/keras_imagenet_main.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -88,7 +88,7 @@ def run(flags_obj):
ValueError: If fp16 is passed as it is not currently supported.
"""
if flags_obj.enable_eager:
tf.enable_eager_execution()
tf.compat.v1.enable_eager_execution()
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

same here re: disable eager in else case


dtype = flags_core.get_tf_dtype(flags_obj)
if dtype == 'fp16':
Expand Down Expand Up @@ -187,7 +187,7 @@ def main(_):


if __name__ == '__main__':
tf.logging.set_verbosity(tf.logging.INFO)
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.INFO)
imagenet_main.define_imagenet_flags()
keras_common.define_keras_flags()
absl_app.run(main)
Loading