Keras TF API usage cleanup: Copy test_util.use_gpu into Keras and replace the usage within Keras.
PiperOrigin-RevId: 322436993 Change-Id: Ic5ed83a643999afec71054aba9484a4a5e4caf88
This commit is contained in:
Rick Chao
TensorFlower Gardener
parent
286e101e8f
commit
4807acc07e
tensorflow/python/keras
@ -20,10 +20,9 @@ from __future__ import print_function
|
||||
|
||||
from absl.testing import parameterized
|
||||
import numpy as np
|
||||
|
||||
from tensorflow.python.framework import test_util
|
||||
from tensorflow.python.keras import backend as K
|
||||
from tensorflow.python.keras import combinations
|
||||
from tensorflow.python.keras import testing_utils
|
||||
from tensorflow.python.keras.engine import input_layer
|
||||
from tensorflow.python.keras.engine import training
|
||||
from tensorflow.python.keras.layers.convolutional import Conv2D
|
||||
@ -71,7 +70,7 @@ class TrainingGPUTest(test.TestCase, parameterized.TestCase):
|
||||
return simple_model
|
||||
|
||||
if test.is_gpu_available(cuda_only=True):
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
losses_to_test = ['sparse_categorical_crossentropy',
|
||||
'categorical_crossentropy', 'binary_crossentropy']
|
||||
|
||||
|
||||
@ -435,7 +435,7 @@ class GroupedConvTest(keras_parameterized.TestCase):
|
||||
)
|
||||
def disable_test_group_conv(self, layer_cls, input_shape):
|
||||
if test.is_gpu_available(cuda_only=True):
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
inputs = random_ops.random_uniform(shape=input_shape)
|
||||
|
||||
layer = layer_cls(16, 3, groups=4, use_bias=False)
|
||||
@ -453,7 +453,7 @@ class GroupedConvTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_group_conv_depthwise(self):
|
||||
if test.is_gpu_available(cuda_only=True):
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
inputs = random_ops.random_uniform(shape=(3, 27, 27, 32))
|
||||
|
||||
layer = keras.layers.Conv2D(32, 3, groups=32, use_bias=False)
|
||||
@ -474,7 +474,7 @@ class Conv1DTransposeTest(keras_parameterized.TestCase):
|
||||
stack_size = 3
|
||||
num_col = 6
|
||||
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
testing_utils.layer_test(
|
||||
keras.layers.Conv1DTranspose,
|
||||
kwargs=kwargs,
|
||||
@ -509,7 +509,7 @@ class Conv3DTransposeTest(keras_parameterized.TestCase):
|
||||
num_col = 6
|
||||
depth = 5
|
||||
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
testing_utils.layer_test(
|
||||
keras.layers.Conv3DTranspose,
|
||||
kwargs=kwargs,
|
||||
|
||||
@ -473,6 +473,7 @@ cuda_py_test(
|
||||
deps = [
|
||||
":image_preprocessing",
|
||||
"//tensorflow/python:client_testlib",
|
||||
"//tensorflow/python/keras:testing_utils",
|
||||
"//third_party/py/numpy",
|
||||
"@absl_py//absl/testing:parameterized",
|
||||
],
|
||||
|
||||
@ -47,7 +47,7 @@ class ResizingTest(keras_parameterized.TestCase):
|
||||
orig_width = 8
|
||||
channels = 3
|
||||
kwargs.update({'height': expected_height, 'width': expected_width})
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
testing_utils.layer_test(
|
||||
image_preprocessing.Resizing,
|
||||
kwargs=kwargs,
|
||||
@ -79,7 +79,7 @@ class ResizingTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_down_sampling_numeric(self):
|
||||
for dtype in (np.int64, np.float32):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
input_image = np.reshape(np.arange(0, 16), (1, 4, 4, 1)).astype(dtype)
|
||||
layer = image_preprocessing.Resizing(
|
||||
height=2, width=2, interpolation='nearest')
|
||||
@ -95,7 +95,7 @@ class ResizingTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_up_sampling_numeric(self):
|
||||
for dtype in (np.int64, np.float32):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
input_image = np.reshape(np.arange(0, 4), (1, 2, 2, 1)).astype(dtype)
|
||||
layer = image_preprocessing.Resizing(
|
||||
height=4, width=4, interpolation='nearest')
|
||||
@ -152,7 +152,7 @@ class CenterCropTest(keras_parameterized.TestCase):
|
||||
(num_samples, orig_height, orig_width, channels)).astype(np.float32)
|
||||
expected_output = get_numpy_center_crop(
|
||||
input_images, expected_height, expected_width)
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
testing_utils.layer_test(
|
||||
image_preprocessing.CenterCrop,
|
||||
kwargs=kwargs,
|
||||
@ -209,7 +209,7 @@ class RandomCropTest(keras_parameterized.TestCase):
|
||||
orig_width = 8
|
||||
channels = 3
|
||||
kwargs = {'height': expected_height, 'width': expected_width}
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
testing_utils.layer_test(
|
||||
image_preprocessing.RandomCrop,
|
||||
kwargs=kwargs,
|
||||
@ -240,7 +240,7 @@ class RandomCropTest(keras_parameterized.TestCase):
|
||||
with test.mock.patch.object(
|
||||
stateless_random_ops, 'stateless_random_uniform',
|
||||
return_value=mock_offset):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
layer = image_preprocessing.RandomCrop(height, width)
|
||||
inp = np.random.random((12, 5, 8, 3))
|
||||
actual_output = layer(inp, training=1)
|
||||
@ -270,7 +270,7 @@ class RandomCropTest(keras_parameterized.TestCase):
|
||||
np.random.seed(1337)
|
||||
height, width = 8, 16
|
||||
inp = np.random.random((12, 8, 16, 3))
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
layer = image_preprocessing.RandomCrop(height, width)
|
||||
actual_output = layer(inp, training=0)
|
||||
self.assertAllClose(inp, actual_output)
|
||||
@ -279,7 +279,7 @@ class RandomCropTest(keras_parameterized.TestCase):
|
||||
np.random.seed(1337)
|
||||
height, width = 3, 3
|
||||
inp = np.random.random((12, 10, 6, 3))
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
layer = image_preprocessing.RandomCrop(height, width)
|
||||
actual_output = layer(inp, training=0)
|
||||
resized_inp = image_ops.resize_images_v2(
|
||||
@ -291,7 +291,7 @@ class RandomCropTest(keras_parameterized.TestCase):
|
||||
np.random.seed(1337)
|
||||
height, width = 4, 6
|
||||
inp = np.random.random((12, 8, 16, 3))
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
layer = image_preprocessing.RandomCrop(height, width)
|
||||
actual_output = layer(inp, training=0)
|
||||
resized_inp = image_ops.resize_images_v2(inp, size=[4, 8])
|
||||
@ -359,7 +359,7 @@ class RandomFlipTest(keras_parameterized.TestCase):
|
||||
expected_output = np.flip(expected_output, axis=1)
|
||||
with test.mock.patch.object(
|
||||
random_ops, 'random_uniform', return_value=mock_random):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
layer = image_preprocessing.RandomFlip(mode)
|
||||
actual_output = layer(inp, training=1)
|
||||
self.assertAllClose(expected_output, actual_output)
|
||||
@ -396,7 +396,7 @@ class RandomFlipTest(keras_parameterized.TestCase):
|
||||
with CustomObjectScope({'RandomFlip': image_preprocessing.RandomFlip}):
|
||||
input_images = np.random.random((2, 5, 8, 3)).astype(np.float32)
|
||||
expected_output = input_images
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
layer = image_preprocessing.RandomFlip()
|
||||
actual_output = layer(input_images, training=0)
|
||||
self.assertAllClose(expected_output, actual_output)
|
||||
@ -446,7 +446,7 @@ class RandomContrastTest(keras_parameterized.TestCase):
|
||||
expected_output = (inp - inp_mean) * mock_random + inp_mean
|
||||
with test.mock.patch.object(
|
||||
random_ops, 'random_uniform', return_value=mock_random):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
layer = image_preprocessing.RandomContrast((lower, upper))
|
||||
actual_output = layer(inp, training=True)
|
||||
self.assertAllClose(expected_output, actual_output)
|
||||
@ -467,7 +467,7 @@ class RandomContrastTest(keras_parameterized.TestCase):
|
||||
with CustomObjectScope(
|
||||
{'RandomContrast': image_preprocessing.RandomContrast}):
|
||||
input_images = np.random.random((2, 5, 8, 3))
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
layer = image_preprocessing.RandomContrast(amplitude)
|
||||
layer(input_images)
|
||||
|
||||
@ -476,7 +476,7 @@ class RandomContrastTest(keras_parameterized.TestCase):
|
||||
{'RandomContrast': image_preprocessing.RandomContrast}):
|
||||
input_images = np.random.random((2, 5, 8, 3)).astype(np.float32)
|
||||
expected_output = input_images
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
layer = image_preprocessing.RandomContrast((0.1, 0.2))
|
||||
actual_output = layer(input_images, training=False)
|
||||
self.assertAllClose(expected_output, actual_output)
|
||||
@ -485,7 +485,7 @@ class RandomContrastTest(keras_parameterized.TestCase):
|
||||
with CustomObjectScope(
|
||||
{'RandomContrast': image_preprocessing.RandomContrast}):
|
||||
input_images = np.random.randint(low=0, high=255, size=(2, 5, 8, 3))
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
layer = image_preprocessing.RandomContrast((0.1, 0.2))
|
||||
layer(input_images)
|
||||
|
||||
@ -517,7 +517,7 @@ class RandomTranslationTest(keras_parameterized.TestCase):
|
||||
orig_width = 8
|
||||
channels = 3
|
||||
kwargs = {'height_factor': height_factor, 'width_factor': width_factor}
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
testing_utils.layer_test(
|
||||
image_preprocessing.RandomTranslation,
|
||||
kwargs=kwargs,
|
||||
@ -532,7 +532,7 @@ class RandomTranslationTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_random_translation_up_numeric_reflect(self):
|
||||
for dtype in (np.int64, np.float32):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
input_image = np.reshape(np.arange(0, 25), (1, 5, 5, 1)).astype(dtype)
|
||||
# Shifting by -.2 * 5 = 1 pixel.
|
||||
layer = image_preprocessing.RandomTranslation(
|
||||
@ -552,7 +552,7 @@ class RandomTranslationTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_random_translation_up_numeric_constant(self):
|
||||
for dtype in (np.int64, np.float32):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
input_image = np.reshape(np.arange(0, 25), (1, 5, 5, 1)).astype(dtype)
|
||||
# Shifting by -.2 * 5 = 1 pixel.
|
||||
layer = image_preprocessing.RandomTranslation(
|
||||
@ -572,7 +572,7 @@ class RandomTranslationTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_random_translation_down_numeric_reflect(self):
|
||||
for dtype in (np.int64, np.float32):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
input_image = np.reshape(np.arange(0, 25), (1, 5, 5, 1)).astype(dtype)
|
||||
# Shifting by .2 * 5 = 1 pixel.
|
||||
layer = image_preprocessing.RandomTranslation(
|
||||
@ -592,7 +592,7 @@ class RandomTranslationTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_random_translation_asymmetric_size_numeric_reflect(self):
|
||||
for dtype in (np.int64, np.float32):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
input_image = np.reshape(np.arange(0, 16), (1, 8, 2, 1)).astype(dtype)
|
||||
# Shifting by .5 * 8 = 1 pixel.
|
||||
layer = image_preprocessing.RandomTranslation(
|
||||
@ -615,7 +615,7 @@ class RandomTranslationTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_random_translation_down_numeric_constant(self):
|
||||
for dtype in (np.int64, np.float32):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
input_image = np.reshape(np.arange(0, 25), (1, 5, 5, 1)).astype(dtype)
|
||||
# Shifting by -.2 * 5 = 1 pixel.
|
||||
layer = image_preprocessing.RandomTranslation(
|
||||
@ -635,7 +635,7 @@ class RandomTranslationTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_random_translation_left_numeric_reflect(self):
|
||||
for dtype in (np.int64, np.float32):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
input_image = np.reshape(np.arange(0, 25), (1, 5, 5, 1)).astype(dtype)
|
||||
# Shifting by .2 * 5 = 1 pixel.
|
||||
layer = image_preprocessing.RandomTranslation(
|
||||
@ -655,7 +655,7 @@ class RandomTranslationTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_random_translation_left_numeric_constant(self):
|
||||
for dtype in (np.int64, np.float32):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
input_image = np.reshape(np.arange(0, 25), (1, 5, 5, 1)).astype(dtype)
|
||||
# Shifting by -.2 * 5 = 1 pixel.
|
||||
layer = image_preprocessing.RandomTranslation(
|
||||
@ -678,7 +678,7 @@ class RandomTranslationTest(keras_parameterized.TestCase):
|
||||
{'RandomTranslation': image_preprocessing.RandomTranslation}):
|
||||
input_images = np.random.random((2, 5, 8, 3)).astype(np.float32)
|
||||
expected_output = input_images
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
layer = image_preprocessing.RandomTranslation(.5, .5)
|
||||
actual_output = layer(input_images, training=0)
|
||||
self.assertAllClose(expected_output, actual_output)
|
||||
@ -996,7 +996,7 @@ class RandomRotationTest(keras_parameterized.TestCase):
|
||||
orig_width = 8
|
||||
channels = 3
|
||||
kwargs = {'factor': factor}
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
testing_utils.layer_test(
|
||||
image_preprocessing.RandomRotation,
|
||||
kwargs=kwargs,
|
||||
@ -1014,7 +1014,7 @@ class RandomRotationTest(keras_parameterized.TestCase):
|
||||
{'RandomTranslation': image_preprocessing.RandomRotation}):
|
||||
input_images = np.random.random((2, 5, 8, 3)).astype(np.float32)
|
||||
expected_output = input_images
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
layer = image_preprocessing.RandomRotation(.5)
|
||||
actual_output = layer(input_images, training=0)
|
||||
self.assertAllClose(expected_output, actual_output)
|
||||
@ -1025,7 +1025,7 @@ class RandomRotationTest(keras_parameterized.TestCase):
|
||||
And that replicas got the same random result.
|
||||
"""
|
||||
input_images = np.random.random((2, 5, 8, 3)).astype(np.float32)
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
strat = MirroredStrategy(devices=['cpu', 'gpu'])
|
||||
with strat.scope():
|
||||
layer = image_preprocessing.RandomRotation(.5)
|
||||
@ -1052,7 +1052,7 @@ class RandomZoomTest(keras_parameterized.TestCase):
|
||||
orig_width = 8
|
||||
channels = 3
|
||||
kwargs = {'height_factor': height_factor, 'width_factor': width_factor}
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
testing_utils.layer_test(
|
||||
image_preprocessing.RandomZoom,
|
||||
kwargs=kwargs,
|
||||
@ -1073,7 +1073,7 @@ class RandomZoomTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_random_zoom_in_numeric(self):
|
||||
for dtype in (np.int64, np.float32):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
input_image = np.reshape(np.arange(0, 25), (5, 5, 1)).astype(dtype)
|
||||
layer = image_preprocessing.RandomZoom((-.5, -.5), (-.5, -.5),
|
||||
interpolation='nearest')
|
||||
@ -1092,7 +1092,7 @@ class RandomZoomTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_random_zoom_out_numeric(self):
|
||||
for dtype in (np.int64, np.float32):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
input_image = np.reshape(np.arange(0, 25), (5, 5, 1)).astype(dtype)
|
||||
layer = image_preprocessing.RandomZoom((.5, .5), (.8, .8),
|
||||
fill_mode='constant',
|
||||
@ -1112,7 +1112,7 @@ class RandomZoomTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_random_zoom_out_numeric_preserve_aspect_ratio(self):
|
||||
for dtype in (np.int64, np.float32):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
input_image = np.reshape(np.arange(0, 25), (5, 5, 1)).astype(dtype)
|
||||
layer = image_preprocessing.RandomZoom((.5, .5),
|
||||
fill_mode='constant',
|
||||
@ -1135,7 +1135,7 @@ class RandomZoomTest(keras_parameterized.TestCase):
|
||||
{'RandomZoom': image_preprocessing.RandomZoom}):
|
||||
input_images = np.random.random((2, 5, 8, 3)).astype(np.float32)
|
||||
expected_output = input_images
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
layer = image_preprocessing.RandomZoom(.5, .5)
|
||||
actual_output = layer(input_images, training=0)
|
||||
self.assertAllClose(expected_output, actual_output)
|
||||
@ -1157,7 +1157,7 @@ class RandomHeightTest(keras_parameterized.TestCase):
|
||||
orig_height = 5
|
||||
orig_width = 8
|
||||
channels = 3
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
img = np.random.random((num_samples, orig_height, orig_width, channels))
|
||||
layer = image_preprocessing.RandomHeight(factor)
|
||||
img_out = layer(img, training=True)
|
||||
@ -1176,7 +1176,7 @@ class RandomHeightTest(keras_parameterized.TestCase):
|
||||
mock_factor = 0
|
||||
with test.mock.patch.object(
|
||||
gen_stateful_random_ops, 'stateful_uniform', return_value=mock_factor):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
img = np.random.random((12, 5, 8, 3))
|
||||
layer = image_preprocessing.RandomHeight(.4)
|
||||
img_out = layer(img, training=True)
|
||||
@ -1184,7 +1184,7 @@ class RandomHeightTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_random_height_longer_numeric(self):
|
||||
for dtype in (np.int64, np.float32):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
input_image = np.reshape(np.arange(0, 6), (2, 3, 1)).astype(dtype)
|
||||
layer = image_preprocessing.RandomHeight(factor=(1., 1.))
|
||||
# Return type of RandomHeight() is float32 if `interpolation` is not
|
||||
@ -1204,7 +1204,7 @@ class RandomHeightTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_random_height_shorter_numeric(self):
|
||||
for dtype in (np.int64, np.float32):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
input_image = np.reshape(np.arange(0, 8), (4, 2, 1)).astype(dtype)
|
||||
layer = image_preprocessing.RandomHeight(
|
||||
factor=(-.5, -.5), interpolation='nearest')
|
||||
@ -1226,7 +1226,7 @@ class RandomHeightTest(keras_parameterized.TestCase):
|
||||
with CustomObjectScope({'RandomHeight': image_preprocessing.RandomHeight}):
|
||||
input_images = np.random.random((2, 5, 8, 3)).astype(np.float32)
|
||||
expected_output = input_images
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
layer = image_preprocessing.RandomHeight(.5)
|
||||
actual_output = layer(input_images, training=0)
|
||||
self.assertAllClose(expected_output, actual_output)
|
||||
@ -1248,7 +1248,7 @@ class RandomWidthTest(keras_parameterized.TestCase):
|
||||
orig_height = 5
|
||||
orig_width = 8
|
||||
channels = 3
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
img = np.random.random((num_samples, orig_height, orig_width, channels))
|
||||
layer = image_preprocessing.RandomWidth(factor)
|
||||
img_out = layer(img, training=True)
|
||||
@ -1267,7 +1267,7 @@ class RandomWidthTest(keras_parameterized.TestCase):
|
||||
mock_factor = 0
|
||||
with test.mock.patch.object(
|
||||
gen_stateful_random_ops, 'stateful_uniform', return_value=mock_factor):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
img = np.random.random((12, 8, 5, 3))
|
||||
layer = image_preprocessing.RandomWidth(.4)
|
||||
img_out = layer(img, training=True)
|
||||
@ -1275,7 +1275,7 @@ class RandomWidthTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_random_width_longer_numeric(self):
|
||||
for dtype in (np.int64, np.float32):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
input_image = np.reshape(np.arange(0, 6), (3, 2, 1)).astype(dtype)
|
||||
layer = image_preprocessing.RandomWidth(factor=(1., 1.))
|
||||
# Return type of RandomWidth() is float32 if `interpolation` is not
|
||||
@ -1294,7 +1294,7 @@ class RandomWidthTest(keras_parameterized.TestCase):
|
||||
|
||||
def test_random_width_shorter_numeric(self):
|
||||
for dtype in (np.int64, np.float32):
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
input_image = np.reshape(np.arange(0, 8), (2, 4, 1)).astype(dtype)
|
||||
layer = image_preprocessing.RandomWidth(
|
||||
factor=(-.5, -.5), interpolation='nearest')
|
||||
@ -1316,7 +1316,7 @@ class RandomWidthTest(keras_parameterized.TestCase):
|
||||
with CustomObjectScope({'RandomWidth': image_preprocessing.RandomWidth}):
|
||||
input_images = np.random.random((2, 5, 8, 3)).astype(np.float32)
|
||||
expected_output = input_images
|
||||
with tf_test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
layer = image_preprocessing.RandomWidth(.5)
|
||||
actual_output = layer(input_images, training=0)
|
||||
self.assertAllClose(expected_output, actual_output)
|
||||
|
||||
@ -73,7 +73,7 @@ class OptimizerTest(test.TestCase, parameterized.TestCase):
|
||||
@combinations.generate(combinations.combine(mode=['graph', 'eager']))
|
||||
def testBasic(self):
|
||||
for dtype in _DATA_TYPES:
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
var0 = variables.Variable([1.0, 2.0], dtype=dtype)
|
||||
var1 = variables.Variable([3.0, 4.0], dtype=dtype)
|
||||
loss = lambda: 5 * var0 + 3 * var1 # pylint: disable=cell-var-from-loop
|
||||
@ -138,7 +138,7 @@ class OptimizerTest(test.TestCase, parameterized.TestCase):
|
||||
@combinations.generate(combinations.combine(mode=['graph', 'eager']))
|
||||
def testPrecomputedGradient(self):
|
||||
for dtype in _DATA_TYPES:
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
var0 = variables.Variable([1.0, 2.0], dtype=dtype)
|
||||
var1 = variables.Variable([3.0, 4.0], dtype=dtype)
|
||||
loss = lambda: 5 * var0 + 3 * var1 # pylint: disable=cell-var-from-loop
|
||||
@ -162,7 +162,7 @@ class OptimizerTest(test.TestCase, parameterized.TestCase):
|
||||
@combinations.generate(combinations.combine(mode=['graph', 'eager']))
|
||||
def testNoGradients(self):
|
||||
for dtype in _DATA_TYPES:
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
var0 = variables.Variable([1.0, 2.0], dtype=dtype)
|
||||
var1 = variables.Variable([3.0, 4.0], dtype=dtype)
|
||||
loss = lambda: 5 * var0 # pylint: disable=cell-var-from-loop
|
||||
@ -174,7 +174,7 @@ class OptimizerTest(test.TestCase, parameterized.TestCase):
|
||||
@combinations.generate(combinations.combine(mode=['graph', 'eager']))
|
||||
def testNoGradientsForAnyVariables_Minimize(self):
|
||||
for dtype in _DATA_TYPES:
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
var0 = variables.Variable([1.0, 2.0], dtype=dtype)
|
||||
var1 = variables.Variable([3.0, 4.0], dtype=dtype)
|
||||
loss = lambda: constant_op.constant(5.0)
|
||||
@ -187,7 +187,7 @@ class OptimizerTest(test.TestCase, parameterized.TestCase):
|
||||
@combinations.generate(combinations.combine(mode=['graph', 'eager']))
|
||||
def testNoGradientsForAnyVariables_ApplyGradients(self):
|
||||
for dtype in _DATA_TYPES:
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
var0 = variables.Variable([1.0, 2.0], dtype=dtype)
|
||||
var1 = variables.Variable([3.0, 4.0], dtype=dtype)
|
||||
sgd_op = gradient_descent.SGD(3.0)
|
||||
@ -198,7 +198,7 @@ class OptimizerTest(test.TestCase, parameterized.TestCase):
|
||||
@combinations.generate(combinations.combine(mode=['graph', 'eager']))
|
||||
def testGradientsAsVariables(self):
|
||||
for i, dtype in enumerate(_DATA_TYPES):
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
var0 = variables.Variable([1.0, 2.0], dtype=dtype)
|
||||
var1 = variables.Variable([3.0, 4.0], dtype=dtype)
|
||||
loss = lambda: 5 * var0 + 3 * var1 # pylint: disable=cell-var-from-loop
|
||||
@ -236,7 +236,7 @@ class OptimizerTest(test.TestCase, parameterized.TestCase):
|
||||
|
||||
@combinations.generate(combinations.combine(mode=['graph', 'eager']))
|
||||
def testComputeGradientsWithTensors(self):
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
x = ops.convert_to_tensor_v2(1.0)
|
||||
|
||||
def f():
|
||||
@ -256,7 +256,7 @@ class OptimizerTest(test.TestCase, parameterized.TestCase):
|
||||
def testConstraint(self):
|
||||
constraint_01 = lambda x: clip_ops.clip_by_value(x, -0.1, 0.)
|
||||
constraint_0 = lambda x: clip_ops.clip_by_value(x, 0., 1.)
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
var0 = variables.Variable([1.0, 2.0],
|
||||
constraint=constraint_01)
|
||||
var1 = variables.Variable([3.0, 4.0],
|
||||
@ -278,14 +278,14 @@ class OptimizerTest(test.TestCase, parameterized.TestCase):
|
||||
|
||||
@combinations.generate(combinations.combine(mode=['graph', 'eager']))
|
||||
def testIterationWithoutMinimize(self):
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
sgd = gradient_descent.SGD(3.0)
|
||||
self.evaluate(sgd.iterations.initializer)
|
||||
self.assertEqual(0, self.evaluate(sgd.iterations))
|
||||
|
||||
@combinations.generate(combinations.combine(mode=['graph', 'eager']))
|
||||
def testConfig(self):
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
opt = gradient_descent.SGD(learning_rate=1.0)
|
||||
config = opt.get_config()
|
||||
opt2 = gradient_descent.SGD.from_config(config)
|
||||
@ -305,7 +305,7 @@ class OptimizerTest(test.TestCase, parameterized.TestCase):
|
||||
|
||||
@combinations.generate(combinations.combine(mode=['graph', 'eager']))
|
||||
def testConfigWithLearningRateDecay(self):
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
var0 = variables.Variable([[1.0], [2.0]], dtype=dtypes.float32)
|
||||
for decay_schedule in [
|
||||
learning_rate_schedule.InverseTimeDecay(
|
||||
@ -336,7 +336,7 @@ class OptimizerTest(test.TestCase, parameterized.TestCase):
|
||||
|
||||
@combinations.generate(combinations.combine(mode=['graph', 'eager']))
|
||||
def testGradClipValue(self):
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
var = variables.Variable([1.0, 2.0])
|
||||
loss = lambda: 3 * var
|
||||
opt = gradient_descent.SGD(learning_rate=1.0, clipvalue=1.0)
|
||||
@ -347,7 +347,7 @@ class OptimizerTest(test.TestCase, parameterized.TestCase):
|
||||
|
||||
@combinations.generate(combinations.combine(mode=['graph', 'eager']))
|
||||
def testGradClipNorm(self):
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
var = variables.Variable([1.0])
|
||||
loss = lambda: 3 * var
|
||||
opt = gradient_descent.SGD(learning_rate=1.0, clipnorm=1.0)
|
||||
@ -368,7 +368,7 @@ class OptimizerTest(test.TestCase, parameterized.TestCase):
|
||||
|
||||
@combinations.generate(combinations.combine(mode=['graph', 'eager']))
|
||||
def testWeights(self):
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
opt1 = adam.Adam(learning_rate=1.0)
|
||||
var1 = variables.Variable([1.0, 2.0], dtype=dtypes.float32)
|
||||
loss1 = lambda: 3 * var1
|
||||
@ -671,7 +671,7 @@ class OptimizersCompatibilityTest(keras_parameterized.TestCase):
|
||||
self.skipTest(
|
||||
'v1 optimizer does not run in eager mode')
|
||||
np.random.seed(1331)
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
train_samples = 20
|
||||
input_dim = 3
|
||||
num_classes = 2
|
||||
@ -757,7 +757,7 @@ class OptimizersCompatibilityTest(keras_parameterized.TestCase):
|
||||
self.skipTest(
|
||||
'v1 optimizer does not run in eager mode')
|
||||
np.random.seed(1331)
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
train_samples = 20
|
||||
input_dim = 3
|
||||
num_classes = 2
|
||||
@ -814,7 +814,7 @@ class OptimizersCompatibilityTest(keras_parameterized.TestCase):
|
||||
self.skipTest(
|
||||
'v1 optimizer does not run in eager mode')
|
||||
np.random.seed(1331)
|
||||
with test_util.use_gpu():
|
||||
with testing_utils.use_gpu():
|
||||
train_samples = 20
|
||||
input_dim = 3
|
||||
num_classes = 2
|
||||
|
||||
@ -31,6 +31,7 @@ from tensorflow.python.framework import dtypes
|
||||
from tensorflow.python.framework import ops
|
||||
from tensorflow.python.framework import test_util
|
||||
from tensorflow.python.keras import combinations
|
||||
from tensorflow.python.keras import testing_utils
|
||||
from tensorflow.python.keras.optimizer_v2 import learning_rate_schedule
|
||||
from tensorflow.python.keras.optimizer_v2 import rmsprop
|
||||
from tensorflow.python.ops import embedding_ops
|
||||
@ -104,7 +105,7 @@ class RMSpropOptimizerTest(test.TestCase):
|
||||
def testDense(self):
|
||||
# TODO(tanzheny, omalleyt): Fix test in eager mode.
|
||||
for (dtype, learning_rate, rho, momentum, epsilon, centered) in _TESTPARAMS:
|
||||
with ops.get_default_graph().as_default(), test_util.use_gpu():
|
||||
with ops.get_default_graph().as_default(), testing_utils.use_gpu():
|
||||
# Initialize variables for numpy implementation.
|
||||
var0_np = np.array([1.0, 2.0], dtype=dtype.as_numpy_dtype)
|
||||
grads0_np = np.array([0.1, 0.2], dtype=dtype.as_numpy_dtype)
|
||||
@ -379,7 +380,7 @@ class RMSpropOptimizerTest(test.TestCase):
|
||||
def testSparse(self):
|
||||
# TODO(tanzheny, omalleyt): Fix test in eager mode.
|
||||
for (dtype, learning_rate, rho, momentum, epsilon, centered) in _TESTPARAMS:
|
||||
with ops.get_default_graph().as_default(), test_util.use_gpu():
|
||||
with ops.get_default_graph().as_default(), testing_utils.use_gpu():
|
||||
# Initialize variables for numpy implementation.
|
||||
var0_np = np.array([1.0, 2.0], dtype=dtype.as_numpy_dtype)
|
||||
grads0_np = np.array([0.1], dtype=dtype.as_numpy_dtype)
|
||||
|
||||
@ -18,6 +18,7 @@ from __future__ import absolute_import
|
||||
from __future__ import division
|
||||
from __future__ import print_function
|
||||
|
||||
import contextlib
|
||||
import functools
|
||||
import threading
|
||||
|
||||
@ -26,6 +27,7 @@ import numpy as np
|
||||
from tensorflow.python import tf2
|
||||
from tensorflow.python.eager import context
|
||||
from tensorflow.python.framework import dtypes
|
||||
from tensorflow.python.framework import ops
|
||||
from tensorflow.python.framework import tensor_shape
|
||||
from tensorflow.python.framework import tensor_spec
|
||||
from tensorflow.python.framework import test_util
|
||||
@ -917,3 +919,21 @@ def _set_v2_dtype_behavior(fn, enabled):
|
||||
base_layer_utils.V2_DTYPE_BEHAVIOR = v2_dtype_behavior
|
||||
|
||||
return tf_decorator.make_decorator(fn, wrapper)
|
||||
|
||||
|
||||
@contextlib.contextmanager
|
||||
def device(should_use_gpu):
|
||||
"""Uses gpu when requested and available."""
|
||||
if should_use_gpu and test_util.is_gpu_available():
|
||||
dev = '/device:GPU:0'
|
||||
else:
|
||||
dev = '/device:CPU:0'
|
||||
with ops.device(dev):
|
||||
yield
|
||||
|
||||
|
||||
@contextlib.contextmanager
|
||||
def use_gpu():
|
||||
"""Uses gpu when requested and available."""
|
||||
with device(should_use_gpu=True):
|
||||
yield
|
||||
|
||||
Loading…
Reference in New Issue
Block a user