experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Simplify Keras unit tests by removing unnecessary session scopes and introducing a utility function for repeated code.

Browse Source 
PiperOrigin-RevId: 209523944
This commit is contained in:
Francois Chollet 2018-08-20 18:48:13 -07:00 committed by TensorFlower Gardener
parent c5f27df3a5
commit d29759fa53
4 changed files with 701 additions and 760 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

101
tensorflow/python/keras/callbacks_test.py
View File

@ -235,11 +235,8 @@ class KerasCallbacksTest(test.TestCase):
num_classes=NUM_CLASSES) num_classes=NUM_CLASSES)
y_test = keras.utils.to_categorical(y_test) y_test = keras.utils.to_categorical(y_test)
y_train = keras.utils.to_categorical(y_train) y_train = keras.utils.to_categorical(y_train)
model = keras.models.Sequential() model = testing_utils.get_small_sequential_mlp(
model.add( num_hidden=NUM_HIDDEN, num_classes=NUM_CLASSES, input_dim=INPUT_DIM)
keras.layers.Dense(
NUM_HIDDEN, input_dim=INPUT_DIM, activation='relu'))
model.add(keras.layers.Dense(NUM_CLASSES, activation='softmax'))
model.compile( model.compile(
loss='categorical_crossentropy', loss='categorical_crossentropy',
optimizer='rmsprop', optimizer='rmsprop',
@ -298,9 +295,8 @@ class KerasCallbacksTest(test.TestCase):
test_samples=50, test_samples=50,
input_shape=(1,), input_shape=(1,),
num_classes=NUM_CLASSES) num_classes=NUM_CLASSES)
model = keras.models.Sequential((keras.layers.Dense( model = testing_utils.get_small_sequential_mlp(
1, input_dim=1, activation='relu'), keras.layers.Dense( num_hidden=1, num_classes=1, input_dim=1)
1, activation='sigmoid'),))
model.compile( model.compile(
optimizer='sgd', loss='binary_crossentropy', metrics=['accuracy']) optimizer='sgd', loss='binary_crossentropy', metrics=['accuracy'])
@ -334,11 +330,8 @@ class KerasCallbacksTest(test.TestCase):
num_classes=NUM_CLASSES) num_classes=NUM_CLASSES)
y_test = keras.utils.to_categorical(y_test) y_test = keras.utils.to_categorical(y_test)
y_train = keras.utils.to_categorical(y_train) y_train = keras.utils.to_categorical(y_train)
model = keras.models.Sequential() model = testing_utils.get_small_sequential_mlp(
model.add( num_hidden=NUM_HIDDEN, num_classes=NUM_CLASSES, input_dim=INPUT_DIM)
keras.layers.Dense(
NUM_HIDDEN, input_dim=INPUT_DIM, activation='relu'))
model.add(keras.layers.Dense(NUM_CLASSES, activation='softmax'))
model.compile( model.compile(
loss='categorical_crossentropy', loss='categorical_crossentropy',
optimizer='sgd', optimizer='sgd',
@ -388,12 +381,8 @@ class KerasCallbacksTest(test.TestCase):
def make_model(): def make_model():
random_seed.set_random_seed(1234) random_seed.set_random_seed(1234)
np.random.seed(1337) np.random.seed(1337)
model = keras.models.Sequential() model = testing_utils.get_small_sequential_mlp(
model.add( num_hidden=NUM_HIDDEN, num_classes=NUM_CLASSES, input_dim=INPUT_DIM)
keras.layers.Dense(
NUM_HIDDEN, input_dim=INPUT_DIM, activation='relu'))
model.add(keras.layers.Dense(NUM_CLASSES, activation='softmax'))
model.compile( model.compile(
loss='categorical_crossentropy', loss='categorical_crossentropy',
optimizer=keras.optimizers.SGD(lr=0.1), optimizer=keras.optimizers.SGD(lr=0.1),
@ -498,12 +487,8 @@ class KerasCallbacksTest(test.TestCase):
def make_model(): def make_model():
np.random.seed(1337) np.random.seed(1337)
model = keras.models.Sequential() model = testing_utils.get_small_sequential_mlp(
model.add( num_hidden=NUM_HIDDEN, num_classes=NUM_CLASSES, input_dim=INPUT_DIM)
keras.layers.Dense(
NUM_HIDDEN, input_dim=INPUT_DIM, activation='relu'))
model.add(keras.layers.Dense(NUM_CLASSES, activation='softmax'))
model.compile( model.compile(
loss='categorical_crossentropy', loss='categorical_crossentropy',
optimizer=keras.optimizers.SGD(lr=0.1), optimizer=keras.optimizers.SGD(lr=0.1),
@ -985,9 +970,8 @@ class KerasCallbacksTest(test.TestCase):
yield x, y yield x, y
with self.test_session(): with self.test_session():
model = keras.models.Sequential() model = testing_utils.get_small_sequential_mlp(
model.add(keras.layers.Dense(10, input_dim=100, activation='relu')) num_hidden=10, num_classes=10, input_dim=100)
model.add(keras.layers.Dense(10, activation='softmax'))
model.compile( model.compile(
loss='categorical_crossentropy', loss='categorical_crossentropy',
optimizer='sgd', optimizer='sgd',
@ -1083,11 +1067,8 @@ class KerasCallbacksTest(test.TestCase):
y_test = keras.utils.to_categorical(y_test) y_test = keras.utils.to_categorical(y_test)
y_train = keras.utils.to_categorical(y_train) y_train = keras.utils.to_categorical(y_train)
model = keras.models.Sequential() model = testing_utils.get_small_sequential_mlp(
model.add( num_hidden=NUM_HIDDEN, num_classes=NUM_CLASSES, input_dim=INPUT_DIM)
keras.layers.Dense(
NUM_HIDDEN, input_dim=INPUT_DIM, activation='relu'))
model.add(keras.layers.Dense(NUM_CLASSES, activation='softmax'))
model.compile( model.compile(
loss='binary_crossentropy', optimizer='sgd', metrics=['accuracy']) loss='binary_crossentropy', optimizer='sgd', metrics=['accuracy'])
@ -1179,40 +1160,36 @@ class KerasCallbacksTest(test.TestCase):
@test_util.run_in_graph_and_eager_modes @test_util.run_in_graph_and_eager_modes
def test_Tensorboard_eager(self): def test_Tensorboard_eager(self):
with self.test_session(): temp_dir = tempfile.mkdtemp(dir=self.get_temp_dir())
temp_dir = tempfile.mkdtemp(dir=self.get_temp_dir()) self.addCleanup(shutil.rmtree, temp_dir, ignore_errors=True)
self.addCleanup(shutil.rmtree, temp_dir, ignore_errors=True)
(x_train, y_train), (x_test, y_test) = testing_utils.get_test_data( (x_train, y_train), (x_test, y_test) = testing_utils.get_test_data(
train_samples=TRAIN_SAMPLES, train_samples=TRAIN_SAMPLES,
test_samples=TEST_SAMPLES, test_samples=TEST_SAMPLES,
input_shape=(INPUT_DIM,), input_shape=(INPUT_DIM,),
num_classes=NUM_CLASSES) num_classes=NUM_CLASSES)
y_test = keras.utils.to_categorical(y_test) y_test = keras.utils.to_categorical(y_test)
y_train = keras.utils.to_categorical(y_train) y_train = keras.utils.to_categorical(y_train)
model = keras.models.Sequential() model = testing_utils.get_small_sequential_mlp(
model.add( num_hidden=NUM_HIDDEN, num_classes=NUM_CLASSES, input_dim=INPUT_DIM)
keras.layers.Dense( model.compile(
NUM_HIDDEN, input_dim=INPUT_DIM, activation='relu')) loss='binary_crossentropy',
model.add(keras.layers.Dense(NUM_CLASSES, activation='softmax')) optimizer=adam.AdamOptimizer(0.01),
model.compile( metrics=['accuracy'])
loss='binary_crossentropy',
optimizer=adam.AdamOptimizer(0.01),
metrics=['accuracy'])
cbks = [keras.callbacks.TensorBoard(log_dir=temp_dir)] cbks = [keras.callbacks.TensorBoard(log_dir=temp_dir)]
model.fit( model.fit(
x_train, x_train,
y_train, y_train,
batch_size=BATCH_SIZE, batch_size=BATCH_SIZE,
validation_data=(x_test, y_test), validation_data=(x_test, y_test),
callbacks=cbks, callbacks=cbks,
epochs=2, epochs=2,
verbose=0) verbose=0)
self.assertTrue(os.path.exists(temp_dir)) self.assertTrue(os.path.exists(temp_dir))
def test_RemoteMonitorWithJsonPayload(self): def test_RemoteMonitorWithJsonPayload(self):
if requests is None: if requests is None:

39
tensorflow/python/keras/engine/sequential_test.py
View File

@ -25,22 +25,12 @@ from tensorflow.python import keras
from tensorflow.python.data.ops import dataset_ops from tensorflow.python.data.ops import dataset_ops
from tensorflow.python.eager import function from tensorflow.python.eager import function
from tensorflow.python.framework import test_util as tf_test_util from tensorflow.python.framework import test_util as tf_test_util
from tensorflow.python.keras import testing_utils
from tensorflow.python.ops import array_ops from tensorflow.python.ops import array_ops
from tensorflow.python.platform import test from tensorflow.python.platform import test
from tensorflow.python.training import rmsprop from tensorflow.python.training import rmsprop
def _get_small_mlp(num_hidden, num_classes, input_dim=None):
model = keras.models.Sequential()
if input_dim:
model.add(keras.layers.Dense(num_hidden, activation='relu',
input_dim=input_dim))
else:
model.add(keras.layers.Dense(num_hidden, activation='relu'))
model.add(keras.layers.Dense(num_classes, activation='softmax'))
return model
class TestSequential(test.TestCase, parameterized.TestCase): class TestSequential(test.TestCase, parameterized.TestCase):
"""Most Sequential model API tests are covered in `training_test.py`. """Most Sequential model API tests are covered in `training_test.py`.
""" """
@ -63,7 +53,8 @@ class TestSequential(test.TestCase, parameterized.TestCase):
batch_size = 5 batch_size = 5
num_classes = 2 num_classes = 2
model = _get_small_mlp(num_hidden, num_classes, input_dim) model = testing_utils.get_small_sequential_mlp(
num_hidden, num_classes, input_dim)
model.compile(loss='mse', optimizer=rmsprop.RMSPropOptimizer(1e-3)) model.compile(loss='mse', optimizer=rmsprop.RMSPropOptimizer(1e-3))
x = np.random.random((batch_size, input_dim)) x = np.random.random((batch_size, input_dim))
y = np.random.random((batch_size, num_classes)) y = np.random.random((batch_size, num_classes))
@ -94,7 +85,7 @@ class TestSequential(test.TestCase, parameterized.TestCase):
batch_size = 5 batch_size = 5
num_classes = 2 num_classes = 2
model = _get_small_mlp(num_hidden, num_classes) model = testing_utils.get_small_sequential_mlp(num_hidden, num_classes)
model.compile( model.compile(
loss='mse', loss='mse',
optimizer=rmsprop.RMSPropOptimizer(1e-3), optimizer=rmsprop.RMSPropOptimizer(1e-3),
@ -118,7 +109,7 @@ class TestSequential(test.TestCase, parameterized.TestCase):
num_samples = 50 num_samples = 50
steps_per_epoch = 10 steps_per_epoch = 10
model = _get_small_mlp(num_hidden, num_classes) model = testing_utils.get_small_sequential_mlp(num_hidden, num_classes)
model.compile( model.compile(
loss='mse', loss='mse',
optimizer=rmsprop.RMSPropOptimizer(1e-3), optimizer=rmsprop.RMSPropOptimizer(1e-3),
@ -145,9 +136,9 @@ class TestSequential(test.TestCase, parameterized.TestCase):
def get_model(): def get_model():
if deferred: if deferred:
model = _get_small_mlp(10, 4) model = testing_utils.get_small_sequential_mlp(10, 4)
else: else:
model = _get_small_mlp(10, 4, input_dim=3) model = testing_utils.get_small_sequential_mlp(10, 4, input_dim=3)
model.compile( model.compile(
optimizer=rmsprop.RMSPropOptimizer(1e-3), optimizer=rmsprop.RMSPropOptimizer(1e-3),
loss='categorical_crossentropy', loss='categorical_crossentropy',
@ -262,7 +253,7 @@ class TestSequential(test.TestCase, parameterized.TestCase):
batch_size = 5 batch_size = 5
num_classes = 2 num_classes = 2
model = _get_small_mlp(num_hidden, num_classes) model = testing_utils.get_small_sequential_mlp(num_hidden, num_classes)
model.compile( model.compile(
loss='mse', loss='mse',
optimizer=rmsprop.RMSPropOptimizer(1e-3), optimizer=rmsprop.RMSPropOptimizer(1e-3),
@ -284,21 +275,21 @@ class TestSequential(test.TestCase, parameterized.TestCase):
@tf_test_util.run_in_graph_and_eager_modes @tf_test_util.run_in_graph_and_eager_modes
def test_sequential_shape_inference_deferred(self): def test_sequential_shape_inference_deferred(self):
model = _get_small_mlp(4, 5) model = testing_utils.get_small_sequential_mlp(4, 5)
output_shape = model.compute_output_shape((None, 7)) output_shape = model.compute_output_shape((None, 7))
self.assertEqual(tuple(output_shape.as_list()), (None, 5)) self.assertEqual(tuple(output_shape.as_list()), (None, 5))
@tf_test_util.run_in_graph_and_eager_modes @tf_test_util.run_in_graph_and_eager_modes
def test_sequential_build_deferred(self): def test_sequential_build_deferred(self):
model = _get_small_mlp(4, 5) model = testing_utils.get_small_sequential_mlp(4, 5)
model.build((None, 10)) model.build((None, 10))
self.assertTrue(model.built) self.assertTrue(model.built)
self.assertEqual(len(model.weights), 4) self.assertEqual(len(model.weights), 4)
# Test with nested model # Test with nested model
model = _get_small_mlp(4, 3) model = testing_utils.get_small_sequential_mlp(4, 3)
inner_model = _get_small_mlp(4, 5) inner_model = testing_utils.get_small_sequential_mlp(4, 5)
model.add(inner_model) model.add(inner_model)
model.build((None, 10)) model.build((None, 10))
@ -308,8 +299,8 @@ class TestSequential(test.TestCase, parameterized.TestCase):
@tf_test_util.run_in_graph_and_eager_modes @tf_test_util.run_in_graph_and_eager_modes
def test_sequential_nesting(self): def test_sequential_nesting(self):
model = _get_small_mlp(4, 3) model = testing_utils.get_small_sequential_mlp(4, 3)
inner_model = _get_small_mlp(4, 5) inner_model = testing_utils.get_small_sequential_mlp(4, 5)
model.add(inner_model) model.add(inner_model)
model.compile(loss='mse', optimizer=rmsprop.RMSPropOptimizer(1e-3)) model.compile(loss='mse', optimizer=rmsprop.RMSPropOptimizer(1e-3))
@ -353,7 +344,7 @@ class TestSequentialEagerIntegration(test.TestCase):
@tf_test_util.run_in_graph_and_eager_modes @tf_test_util.run_in_graph_and_eager_modes
def test_build_before_fit(self): def test_build_before_fit(self):
# Fix for b/112433577 # Fix for b/112433577
model = _get_small_mlp(4, 5) model = testing_utils.get_small_sequential_mlp(4, 5)
model.compile(loss='mse', optimizer=rmsprop.RMSPropOptimizer(1e-3)) model.compile(loss='mse', optimizer=rmsprop.RMSPropOptimizer(1e-3))
model.build((None, 6)) model.build((None, 6))

1302
tensorflow/python/keras/engine/training_test.py
View File

File diff suppressed because it is too large Load Diff

19
tensorflow/python/keras/testing_utils.py
View File

@ -184,3 +184,22 @@ def layer_test(layer_cls, kwargs=None, input_shape=None, input_dtype=None,
# for further checks in the caller function # for further checks in the caller function
return actual_output return actual_output
def get_small_sequential_mlp(num_hidden, num_classes, input_dim=None):
model = keras.models.Sequential()
if input_dim:
model.add(keras.layers.Dense(num_hidden, activation='relu',
input_dim=input_dim))
else:
model.add(keras.layers.Dense(num_hidden, activation='relu'))
activation = 'sigmoid' if num_classes == 1 else 'softmax'
model.add(keras.layers.Dense(num_classes, activation=activation))
return model
def get_small_functional_mlp(num_hidden, num_classes, input_dim):
inputs = keras.Input(shape=(input_dim,))
outputs = keras.layers.Dense(num_hidden, activation='relu')(inputs)
activation = 'sigmoid' if num_classes == 1 else 'softmax'
outputs = keras.layers.Dense(num_classes, activation=activation)(outputs)
return keras.Model(inputs, outputs)