def get_model_from_layers(layers,
input_shape=None,
input_dtype=None,
name=None,
input_ragged=None,
input_sparse=None):
"""Builds a model from a sequence of layers.
Args:
layers: The layers used to build the network.
input_shape: Shape tuple of the input or 'TensorShape' instance.
input_dtype: Datatype of the input.
name: Name for the model.
input_ragged: Boolean, whether the input data is a ragged tensor.
input_sparse: Boolean, whether the input data is a sparse tensor.
Returns:
A Keras model.
"""
model_type = get_model_type()
if model_type == 'subclass':
inputs = None
if input_ragged or input_sparse:
inputs = keras.Input(shape=input_shape,
dtype=input_dtype,
ragged=input_ragged,
sparse=input_sparse)
return _SubclassModel(layers, name=name, input_tensor=inputs)
if model_type == 'subclass_custom_build':
layer_generating_func = lambda: layers
return _SubclassModelCustomBuild(layer_generating_func, name=name)
if model_type == 'sequential':
model = keras.models.Sequential(name=name)
if input_shape:
model.add(
keras.layers.InputLayer(input_shape=input_shape,
dtype=input_dtype,
ragged=input_ragged,
sparse=input_sparse))
for layer in layers:
model.add(layer)
return model
if model_type == 'functional':
if not input_shape:
raise ValueError(
'Cannot create a functional model from layers with no '
'input shape.')
inputs = keras.Input(shape=input_shape,
dtype=input_dtype,
ragged=input_ragged,
sparse=input_sparse)
outputs = inputs
for layer in layers:
outputs = layer(outputs)
return keras.Model(inputs, outputs, name=name)
raise ValueError('Unknown model type {}'.format(model_type))
def _single_op_with_attrs():
inputs = keras.Input(shape=(10, ))
x = math_ops.reduce_mean(inputs, axis=1, keepdims=True)
outputs = keras.layers.Dense(10)(x)
return inputs, outputs
def test_minimal_rnn_cell_layer(self):
class MinimalRNNCell(keras.layers.Layer):
def __init__(self, units, **kwargs):
self.units = units
self.state_size = units
super(MinimalRNNCell, self).__init__(**kwargs)
def build(self, input_shape):
self.kernel = self.add_weight(shape=(input_shape[-1],
self.units),
initializer='uniform',
name='kernel')
self.recurrent_kernel = self.add_weight(
shape=(self.units, self.units),
initializer='uniform',
name='recurrent_kernel')
self.built = True
def call(self, inputs, states):
prev_output = states[0]
h = keras.backend.dot(inputs, self.kernel)
output = h + keras.backend.dot(prev_output,
self.recurrent_kernel)
return output, [output]
def get_config(self):
config = {'units': self.units}
base_config = super(MinimalRNNCell, self).get_config()
return dict(list(base_config.items()) + list(config.items()))
with self.cached_session():
# Test basic case.
x = keras.Input((None, 5))
cell = MinimalRNNCell(32)
layer = keras.layers.RNN(cell)
y = layer(x)
model = keras.models.Model(x, y)
model.compile(optimizer='rmsprop', loss='mse')
model.train_on_batch(np.zeros((6, 5, 5)), np.zeros((6, 32)))
# Test basic case serialization.
x_np = np.random.random((6, 5, 5))
y_np = model.predict(x_np)
weights = model.get_weights()
config = layer.get_config()
with keras.utils.CustomObjectScope(
{'MinimalRNNCell': MinimalRNNCell}):
layer = keras.layers.RNN.from_config(config)
y = layer(x)
model = keras.models.Model(x, y)
model.set_weights(weights)
y_np_2 = model.predict(x_np)
self.assertAllClose(y_np, y_np_2, atol=1e-4)
# Test stacking.
cells = [MinimalRNNCell(8), MinimalRNNCell(12), MinimalRNNCell(32)]
layer = keras.layers.RNN(cells)
y = layer(x)
model = keras.models.Model(x, y)
model.compile(optimizer='rmsprop', loss='mse')
model.train_on_batch(np.zeros((6, 5, 5)), np.zeros((6, 32)))
# Test stacked RNN serialization.
x_np = np.random.random((6, 5, 5))
y_np = model.predict(x_np)
weights = model.get_weights()
config = layer.get_config()
with keras.utils.CustomObjectScope(
{'MinimalRNNCell': MinimalRNNCell}):
layer = keras.layers.RNN.from_config(config)
y = layer(x)
model = keras.models.Model(x, y)
model.set_weights(weights)
y_np_2 = model.predict(x_np)
self.assertAllClose(y_np, y_np_2, atol=1e-4)
def _multiple_ops_at_end():
inputs = keras.Input(shape=(10, ))
x = keras.layers.Dense(10)(inputs)
x = gen_nn_ops.relu(x)
outputs = gen_nn_ops.relu(x)
return keras.Model(inputs, outputs)
def _single_op_in_middle():
inputs = keras.Input(shape=(10, ))
x = keras.layers.Dense(10)(inputs)
x = gen_nn_ops.relu(x, name='hey')
outputs = keras.layers.Dense(10)(x)
return inputs, outputs
def _add_n():
inputs = keras.Input(shape=(10, ))
outputs = math_ops.add_n([inputs, inputs, inputs])
return keras.Model(inputs, outputs)
def _inner_layer():
inputs = keras.Input(shape=(10, ))
outputs = LayerWithLayer()(inputs)
return keras.Model(inputs, outputs)
def test_updates_and_losses_for_nested_models_in_subclassed_model(self):
# Case 1: deferred-build sequential nested in subclass.
class TestModel1(keras.Model):
def __init__(self):
super(TestModel1, self).__init__()
self.fc = keras.layers.Dense(10,
input_shape=(784, ),
activity_regularizer='l1')
self.bn = keras.Sequential(
[keras.layers.BatchNormalization(axis=1)])
def call(self, x):
return self.bn(self.fc(x))
with ops.get_default_graph().as_default(), self.cached_session():
model = TestModel1()
x = array_ops.ones(shape=[100, 784], dtype='float32')
model(x)
self.assertEqual(len(model.get_updates_for(x)), 2)
self.assertEqual(len(model.get_losses_for(x)), 1)
# Case 2: placeholder-sequential nested in subclass.
class TestModel2(keras.Model):
def __init__(self):
super(TestModel2, self).__init__()
self.fc = keras.layers.Dense(10,
input_shape=(784, ),
activity_regularizer='l1')
self.bn = keras.Sequential([
keras.layers.BatchNormalization(axis=1, input_shape=(10, ))
])
def call(self, x):
return self.bn(self.fc(x))
with ops.get_default_graph().as_default(), self.cached_session():
model = TestModel2()
x = array_ops.ones(shape=[100, 784], dtype='float32')
model(x)
self.assertEqual(len(model.get_updates_for(x)), 2)
self.assertEqual(len(model.get_losses_for(x)), 1)
# Case 3: functional-API model nested in subclass.
with ops.get_default_graph().as_default():
inputs = keras.Input((10, ))
outputs = keras.layers.BatchNormalization(axis=1)(inputs)
bn = keras.Model(inputs, outputs)
class TestModel3(keras.Model):
def __init__(self):
super(TestModel3, self).__init__()
self.fc = keras.layers.Dense(10,
input_shape=(784, ),
activity_regularizer='l1')
self.bn = bn
def call(self, x):
return self.bn(self.fc(x))
with self.cached_session():
model = TestModel3()
x = array_ops.ones(shape=[100, 784], dtype='float32')
model(x)
self.assertEqual(len(model.get_updates_for(x)), 2)
self.assertEqual(len(model.get_losses_for(x)), 1)
def test_eager_switch_case_input(self):
with context.eager_mode():
task = keras.Input(shape=(), dtype=dtypes.int32)
control_flow_ops.switch_case(
task[0],
[lambda: constant_op.constant(1.0) for _ in range(10)])
def test_rnn_with_time_major(self):
batch = 10
time_step = 5
embedding_dim = 4
units = 3
# Test basic case.
x = keras.Input((time_step, embedding_dim))
time_major_x = keras.layers.Lambda(
lambda t: array_ops.transpose(t, [1, 0, 2]))(x)
layer = keras.layers.SimpleRNN(
units, time_major=True, return_sequences=True)
self.assertEqual(
layer.compute_output_shape((time_step, None,
embedding_dim)).as_list(),
[time_step, None, units])
y = layer(time_major_x)
self.assertEqual(layer.output_shape, (time_step, None, units))
y = keras.layers.Lambda(lambda t: array_ops.transpose(t, [1, 0, 2]))(y)
model = keras.models.Model(x, y)
model.compile(
optimizer='rmsprop',
loss='mse',
run_eagerly=testing_utils.should_run_eagerly())
model.train_on_batch(
np.zeros((batch, time_step, embedding_dim)),
np.zeros((batch, time_step, units)))
# Test stacking.
x = keras.Input((time_step, embedding_dim))
time_major_x = keras.layers.Lambda(
lambda t: array_ops.transpose(t, [1, 0, 2]))(x)
cell_units = [10, 8, 6]
cells = [keras.layers.SimpleRNNCell(cell_units[i]) for i in range(3)]
layer = keras.layers.RNN(cells, time_major=True, return_sequences=True)
y = layer(time_major_x)
self.assertEqual(layer.output_shape, (time_step, None, cell_units[-1]))
y = keras.layers.Lambda(lambda t: array_ops.transpose(t, [1, 0, 2]))(y)
model = keras.models.Model(x, y)
model.compile(
optimizer='rmsprop',
loss='mse',
run_eagerly=testing_utils.should_run_eagerly())
model.train_on_batch(
np.zeros((batch, time_step, embedding_dim)),
np.zeros((batch, time_step, cell_units[-1])))
# Test masking.
x = keras.Input((time_step, embedding_dim))
time_major = keras.layers.Lambda(
lambda t: array_ops.transpose(t, [1, 0, 2]))(x)
mask = keras.layers.Masking()(time_major)
rnn = keras.layers.SimpleRNN(
units, time_major=True, return_sequences=True)(mask)
y = keras.layers.Lambda(lambda t: array_ops.transpose(t, [1, 0, 2]))(rnn)
model = keras.models.Model(x, y)
model.compile(
optimizer='rmsprop',
loss='mse',
run_eagerly=testing_utils.should_run_eagerly())
model.train_on_batch(
np.zeros((batch, time_step, embedding_dim)),
np.zeros((batch, time_step, units)))
# Test layer output
x = keras.Input((time_step, embedding_dim))
rnn_1 = keras.layers.SimpleRNN(units, return_sequences=True)
y = rnn_1(x)
model = keras.models.Model(x, y)
model.compile(
optimizer='rmsprop',
loss='mse',
run_eagerly=testing_utils.should_run_eagerly())
model.train_on_batch(
np.zeros((batch, time_step, embedding_dim)),
np.zeros((batch, time_step, units)))
x_np = np.random.random((batch, time_step, embedding_dim))
y_np_1 = model.predict(x_np)
time_major = keras.layers.Lambda(
lambda t: array_ops.transpose(t, [1, 0, 2]))(x)
rnn_2 = keras.layers.SimpleRNN(
units, time_major=True, return_sequences=True)
y_2 = rnn_2(time_major)
y_2 = keras.layers.Lambda(
lambda t: array_ops.transpose(t, [1, 0, 2]))(y_2)
model_2 = keras.models.Model(x, y_2)
rnn_2.set_weights(rnn_1.get_weights())
y_np_2 = model_2.predict(x_np)
self.assertAllClose(y_np_1, y_np_2, atol=1e-4)
def test_nest_input_output_with_init_state(self):
batch = 10
t = 5
i1, i2, i3 = 3, 4, 5
o1, o2, o3 = 2, 3, 4
cell = NestedCell(o1, o2, o3)
rnn = keras.layers.RNN(cell, return_sequences=True, return_state=True)
input_1 = keras.Input((t, i1))
input_2 = keras.Input((t, i2, i3))
init_s1 = keras.Input((o1,))
init_s2 = keras.Input((o2, o3))
output1, output2, s1, s2 = rnn((input_1, input_2),
initial_state=(init_s1, init_s2))
self.assertEqual(output1.shape.as_list(), [None, t, o1])
self.assertEqual(output2.shape.as_list(), [None, t, o2, o3])
self.assertEqual(s1.shape.as_list(), [None, o1])
self.assertEqual(s2.shape.as_list(), [None, o2, o3])
model = keras.models.Model([input_1, input_2, init_s1, init_s2],
[output1, output2])
model.compile(
optimizer='rmsprop',
loss='mse',
run_eagerly=testing_utils.should_run_eagerly())
model.train_on_batch(
[np.zeros((batch, t, i1)),
np.zeros((batch, t, i2, i3)),
np.zeros((batch, o1)),
np.zeros((batch, o2, o3))],
[np.zeros((batch, t, o1)),
np.zeros((batch, t, o2, o3))])
self.assertEqual(model.output_shape, [(None, t, o1), (None, t, o2, o3)])
cell = NestedCell(o1, o2, o3, use_tuple=True)
rnn = keras.layers.RNN(cell, return_sequences=True, return_state=True)
input_1 = keras.Input((t, i1))
input_2 = keras.Input((t, i2, i3))
init_s1 = keras.Input((o1,))
init_s2 = keras.Input((o2, o3))
init_state = NestedState(s1=init_s1, s2=init_s2)
output1, output2, s1, s2 = rnn(NestedInput(t1=input_1, t2=input_2),
initial_state=init_state)
self.assertEqual(output1.shape.as_list(), [None, t, o1])
self.assertEqual(output2.shape.as_list(), [None, t, o2, o3])
self.assertEqual(s1.shape.as_list(), [None, o1])
self.assertEqual(s2.shape.as_list(), [None, o2, o3])
model = keras.models.Model([input_1, input_2, init_s1, init_s2],
[output1, output2])
model.compile(
optimizer='rmsprop',
loss='mse',
run_eagerly=testing_utils.should_run_eagerly())
model.train_on_batch(
[np.zeros((batch, t, i1)),
np.zeros((batch, t, i2, i3)),
np.zeros((batch, o1)),
np.zeros((batch, o2, o3))],
[np.zeros((batch, t, o1)),
np.zeros((batch, t, o2, o3))])
self.assertEqual(model.output_shape, [(None, t, o1), (None, t, o2, o3)])
with open(LOGDIR, 'rb') as f:
lda = pickle.load(f)
else:
lda = fast_lda_topics(train,
n_components=sc.n_labels,
max_iter=1000,
total_samples=total_samples)
with open(LOGDIR, 'wb') as f:
pickle.dump(lda, f)
topics_text = get_topics_string(lda,
vocabulary=sc.vocabulary,
n_topics=sc.n_labels,
n_words=20)
### LDA with amortized inference
else:
layers = [keras.Input(shape=shape)]
if lognorm:
layers.append(keras.layers.Lambda(lambda x: tf.math.log1p(x)))
layers += [
keras.layers.Dense(300, activation='relu'),
keras.layers.Dense(300, activation='relu'),
keras.layers.Dense(300, activation='relu'),
]
encoder = keras.Sequential(
layers,
name="Encoder",
)
lda = AmortizedLDA(n_words=shape[-1],
n_topics=sc.n_labels,
lda_posterior=posterior,
word_distribution=distribution,
def test_merge_add_dynamic_shape(self): i1 = keras.Input(batch_shape=(4, None), dtype='float32') i2 = keras.Input(batch_shape=(4, 5), dtype='float32') layer = keras.layers.Add() o = layer([i1, i2]) self.assertListEqual(o.shape.as_list(), [4, 5])
def test_clone_functional_model(self, share_weights):
if share_weights:
clone_fn = functools.partial(keras.models._clone_functional_model,
layer_fn=models.share_weights)
else:
clone_fn = keras.models.clone_model
val_a = np.random.random((10, 4))
val_b = np.random.random((10, 4))
val_out = np.random.random((10, 4))
input_a = keras.Input(shape=(4, ))
input_b = keras.Input(shape=(4, ))
dense_1 = keras.layers.Dense(4, )
dense_2 = keras.layers.Dense(4, )
x_a = dense_1(input_a)
x_a = keras.layers.Dropout(0.5)(x_a)
x_a = keras.layers.BatchNormalization()(x_a)
x_b = dense_1(input_b)
x_a = dense_2(x_a)
outputs = keras.layers.add([x_a, x_b])
model = keras.models.Model([input_a, input_b], outputs)
# With placeholder creation
new_model = clone_fn(model)
self.assertEqual(len(new_model.get_updates_for(new_model.inputs)), 2)
new_model.compile(testing_utils.get_v2_optimizer('rmsprop'),
'mse',
run_eagerly=testing_utils.should_run_eagerly(),
experimental_run_tf_function=testing_utils.
should_run_tf_function())
new_model.train_on_batch([val_a, val_b], val_out)
# On top of new tensors
input_a = keras.Input(shape=(4, ), name='a')
input_b = keras.Input(shape=(4, ), name='b')
new_model = keras.models.clone_model(model,
input_tensors=[input_a, input_b])
self.assertEqual(len(new_model.get_updates_for(new_model.inputs)), 2)
new_model.compile(testing_utils.get_v2_optimizer('rmsprop'),
'mse',
run_eagerly=testing_utils.should_run_eagerly(),
experimental_run_tf_function=testing_utils.
should_run_tf_function())
new_model.train_on_batch([val_a, val_b], val_out)
# On top of new, non-Keras tensors
if not context.executing_eagerly():
# TODO(b/121277734):Skip Eager contexts, as Input() layers raise an error
# saying they should not be used with EagerTensors
input_a = keras.backend.variable(val_a)
input_b = keras.backend.variable(val_b)
new_model = clone_fn(model, input_tensors=[input_a, input_b])
self.assertEqual(len(new_model.get_updates_for(new_model.inputs)),
2)
new_model.compile(testing_utils.get_v2_optimizer('rmsprop'),
'mse',
run_eagerly=testing_utils.should_run_eagerly(),
experimental_run_tf_function=testing_utils.
should_run_tf_function())
new_model.train_on_batch(None, val_out)
def test_TensorBoard_multi_input_output(self):
np.random.seed(1337)
tmpdir = self.get_temp_dir()
self.addCleanup(shutil.rmtree, tmpdir, ignore_errors=True)
with ops.Graph().as_default(), self.cached_session():
filepath = os.path.join(tmpdir, 'logs')
(x_train, y_train), (x_test, y_test) = testing_utils.get_test_data(
train_samples=TRAIN_SAMPLES,
test_samples=TEST_SAMPLES,
input_shape=(INPUT_DIM, ),
num_classes=NUM_CLASSES)
y_test = np_utils.to_categorical(y_test)
y_train = np_utils.to_categorical(y_train)
def data_generator(train):
if train:
max_batch_index = len(x_train) // BATCH_SIZE
else:
max_batch_index = len(x_test) // BATCH_SIZE
i = 0
while 1:
if train:
# simulate multi-input/output models
yield ([x_train[i * BATCH_SIZE:(i + 1) * BATCH_SIZE]] *
2,
[y_train[i * BATCH_SIZE:(i + 1) * BATCH_SIZE]] *
2)
else:
yield ([x_test[i * BATCH_SIZE:(i + 1) * BATCH_SIZE]] *
2,
[y_test[i * BATCH_SIZE:(i + 1) * BATCH_SIZE]] *
2)
i += 1
i %= max_batch_index
inp1 = keras.Input((INPUT_DIM, ))
inp2 = keras.Input((INPUT_DIM, ))
inp = keras.layers.add([inp1, inp2])
hidden = keras.layers.Dense(2, activation='relu')(inp)
hidden = keras.layers.Dropout(0.1)(hidden)
output1 = keras.layers.Dense(NUM_CLASSES,
activation='softmax')(hidden)
output2 = keras.layers.Dense(NUM_CLASSES,
activation='softmax')(hidden)
model = keras.models.Model([inp1, inp2], [output1, output2])
model.compile(loss='categorical_crossentropy',
optimizer='sgd',
metrics=['accuracy'])
# we must generate new callbacks for each test, as they aren't stateless
def callbacks_factory(histogram_freq):
return [
callbacks_v1.TensorBoard(log_dir=filepath,
histogram_freq=histogram_freq,
write_images=True,
write_grads=True,
batch_size=5)
]
# fit without validation data
model.fit([x_train] * 2, [y_train] * 2,
batch_size=BATCH_SIZE,
callbacks=callbacks_factory(histogram_freq=0),
epochs=3)
# fit with validation data and accuracy
model.fit([x_train] * 2, [y_train] * 2,
batch_size=BATCH_SIZE,
validation_data=([x_test] * 2, [y_test] * 2),
callbacks=callbacks_factory(histogram_freq=1),
epochs=2)
# fit generator without validation data
model.fit_generator(data_generator(True),
len(x_train),
epochs=2,
callbacks=callbacks_factory(histogram_freq=0))
# fit generator with validation data and accuracy
model.fit_generator(data_generator(True),
len(x_train),
epochs=2,
validation_data=([x_test] * 2, [y_test] * 2),
callbacks=callbacks_factory(histogram_freq=1))
assert os.path.isdir(filepath)
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)
def _op_with_tensor_list():
inputs = keras.Input(shape=(10, ))
x = array_ops.concat([inputs, inputs], axis=1)
outputs = keras.layers.Dense(10)(x)
return keras.Model(inputs, outputs)
def test_cudnnrnn_bidirectional(self):
if test.is_gpu_available(cuda_only=True):
with self.session(use_gpu=True):
rnn = keras.layers.CuDNNGRU
samples = 2
dim = 2
timesteps = 2
output_dim = 2
mode = 'concat'
x = np.random.random((samples, timesteps, dim))
target_dim = 2 * output_dim if mode == 'concat' else output_dim
y = np.random.random((samples, target_dim))
# test with Sequential model
model = keras.Sequential()
model.add(
keras.layers.Bidirectional(rnn(output_dim),
merge_mode=mode,
input_shape=(None, dim)))
model.compile(loss='mse',
optimizer=RMSPropOptimizer(learning_rate=0.001))
model.fit(x, y, epochs=1, batch_size=1)
# test config
model.get_config()
model = keras.models.model_from_json(model.to_json())
model.summary()
# test stacked bidirectional layers
model = keras.Sequential()
model.add(
keras.layers.Bidirectional(rnn(output_dim,
return_sequences=True),
merge_mode=mode,
input_shape=(None, dim)))
model.add(
keras.layers.Bidirectional(rnn(output_dim),
merge_mode=mode))
model.compile(loss='mse',
optimizer=RMSPropOptimizer(learning_rate=0.001))
model.fit(x, y, epochs=1, batch_size=1)
# test with functional API
inputs = keras.Input((timesteps, dim))
outputs = keras.layers.Bidirectional(rnn(output_dim),
merge_mode=mode)(inputs)
model = keras.Model(inputs, outputs)
model.compile(loss='mse',
optimizer=RMSPropOptimizer(learning_rate=0.001))
model.fit(x, y, epochs=1, batch_size=1)
# Bidirectional and stateful
inputs = keras.Input(batch_shape=(1, timesteps, dim))
outputs = keras.layers.Bidirectional(rnn(output_dim,
stateful=True),
merge_mode=mode)(inputs)
model = keras.Model(inputs, outputs)
model.compile(loss='mse',
optimizer=RMSPropOptimizer(learning_rate=0.001))
model.fit(x, y, epochs=1, batch_size=1)
def _layer_with_tensor_arg():
inputs = keras.Input(shape=(10, ))
x = inputs * 2
outputs = MyAdd()(inputs, x)
return keras.Model(inputs, outputs)
def test_output_shape(self): input_data = keras.Input(shape=(4,), dtype=dtypes.string) layer = get_layer_class()() int_data = layer(input_data) self.assertAllEqual(int_data.shape[1:], input_data.shape[1:])
def _single_identity_op_at_end():
inputs = keras.Input(shape=(10, ))
x = keras.layers.Dense(10)(inputs)
outputs = array_ops.identity(x)
return keras.Model(inputs, outputs)
def test_weight_preprocessing(self):
input_dim = 3
output_dim = 3
size = 2
cases = [
[
(keras.layers.Bidirectional(keras.layers.SimpleRNN(2))),
[np.random.random((2, 1)),
np.random.random((2, 1))],
(None, 3, 2),
],
[
(keras.layers.TimeDistributed(keras.layers.Dense(1))),
[np.random.random((2, 1)),
np.random.random((1, ))],
(None, 3, 2),
],
[
(keras.layers.Conv1D(output_dim, size, use_bias=False)),
[np.random.random((output_dim, input_dim, size, 1))],
(None, 4, input_dim),
],
[
(keras.layers.Conv2D(output_dim,
size,
use_bias=False,
data_format='channels_first')),
[np.random.random((output_dim, input_dim, size, size))],
(None, input_dim, 4, 4),
],
[
(keras.layers.Conv2DTranspose(output_dim,
size,
use_bias=False,
data_format='channels_first')),
[np.random.random((output_dim, input_dim, size, size))],
(None, input_dim, 4, 4),
],
[
(keras.layers.Conv2DTranspose(output_dim,
size,
use_bias=False,
data_format='channels_last')),
[np.random.random((size, size, input_dim, output_dim))],
(None, 4, 4, input_dim),
],
[
(keras.layers.Conv3D(output_dim,
size,
use_bias=False,
data_format='channels_first')),
[np.random.random((output_dim, input_dim, size, size, size))],
(None, input_dim, 4, 4, 4),
],
[
(keras.layers.GRU(output_dim)),
[
np.random.random((input_dim, output_dim)),
np.random.random((output_dim, output_dim)),
np.random.random((output_dim, )),
np.random.random((input_dim, output_dim)),
np.random.random((output_dim, output_dim)),
np.random.random((output_dim, )),
np.random.random((input_dim, output_dim)),
np.random.random((output_dim, output_dim)),
np.random.random((output_dim, ))
],
(None, 4, input_dim),
],
[
(keras.layers.LSTM(output_dim)),
[
np.random.random((input_dim, output_dim)),
np.random.random((output_dim, output_dim)),
np.random.random((output_dim, )),
np.random.random((input_dim, output_dim)),
np.random.random((output_dim, output_dim)),
np.random.random((output_dim, )),
np.random.random((input_dim, output_dim)),
np.random.random((output_dim, output_dim)),
np.random.random((output_dim, )),
np.random.random((input_dim, output_dim)),
np.random.random((output_dim, output_dim)),
np.random.random((output_dim, ))
],
(None, 4, input_dim),
],
]
for layer, weights, input_shape in cases:
layer.build(input_shape)
_ = hdf5_format.preprocess_weights_for_loading(
layer, weights, original_keras_version='1')
model = keras.models.Sequential([keras.layers.Dense(2, input_dim=2)])
_ = hdf5_format.preprocess_weights_for_loading(
model, model.weights, original_keras_version='1')
x = keras.Input((2, ))
y = keras.layers.Dense(2)(x)
model = keras.models.Model(x, y)
_ = hdf5_format.preprocess_weights_for_loading(
model, model.weights, original_keras_version='1')
def _single_op_at_end():
inputs = keras.Input(shape=(10, ))
x = keras.layers.Dense(10)(inputs)
outputs = gen_nn_ops.relu(x, name='hey')
return inputs, outputs
def test_output_shape(self): input_data = keras.Input(shape=(4,), dtype=dtypes.int64) layer = integer_lookup.IntegerLookup(max_tokens=2, num_oov_indices=1) int_data = layer(input_data) self.assertAllEqual(int_data.shape[1:], input_data.shape[1:])
def _single_standalone_branch():
inputs = keras.Input(shape=(10, ))
x = keras.layers.Dense(10)(inputs)
outputs = x * 2
return inputs, outputs
def get_functional_model():
inputs = keras.Input(shape=(4, ))
x = keras.layers.Dense(4, activation='relu')(inputs)
x = keras.layers.BatchNormalization()(x)
outputs = keras.layers.Dense(2)(x)
return keras.Model(inputs, outputs)
def test_invalid_forward_pass(self):
inputs = keras.Input((3,))
with self.assertRaisesRegexp(ValueError, 'You did something wrong!'):
_ = InvalidLayer()(inputs)
def test_Bidirectional_merged_value(self):
rnn = keras.layers.LSTM
samples = 2
dim = 5
timesteps = 3
units = 3
x = [np.random.rand(samples, timesteps, dim)]
with self.cached_session():
for merge_mode in ['sum', 'mul', 'ave', 'concat', None]:
if merge_mode == 'sum':
merge_func = lambda y, y_rev: y + y_rev
elif merge_mode == 'mul':
merge_func = lambda y, y_rev: y * y_rev
elif merge_mode == 'ave':
merge_func = lambda y, y_rev: (y + y_rev) / 2
elif merge_mode == 'concat':
merge_func = lambda y, y_rev: np.concatenate(
(y, y_rev), axis=-1)
else:
merge_func = lambda y, y_rev: [y, y_rev]
# basic case
inputs = keras.Input((timesteps, dim))
layer = keras.layers.Bidirectional(rnn(units,
return_sequences=True),
merge_mode=merge_mode)
f_merged = keras.backend.function([inputs],
_to_list(layer(inputs)))
f_forward = keras.backend.function(
[inputs], [layer.forward_layer(inputs)])
f_backward = keras.backend.function(
[inputs],
[keras.backend.reverse(layer.backward_layer(inputs), 1)])
y_merged = f_merged(x)
y_expected = _to_list(
merge_func(f_forward(x)[0],
f_backward(x)[0]))
assert len(y_merged) == len(y_expected)
for x1, x2 in zip(y_merged, y_expected):
self.assertAllClose(x1, x2, atol=1e-5)
# test return_state
inputs = keras.Input((timesteps, dim))
layer = keras.layers.Bidirectional(rnn(units,
return_state=True),
merge_mode=merge_mode)
f_merged = keras.backend.function([inputs], layer(inputs))
f_forward = keras.backend.function([inputs],
layer.forward_layer(inputs))
f_backward = keras.backend.function(
[inputs], layer.backward_layer(inputs))
n_states = len(layer.layer.states)
y_merged = f_merged(x)
y_forward = f_forward(x)
y_backward = f_backward(x)
y_expected = _to_list(merge_func(y_forward[0], y_backward[0]))
assert len(y_merged) == len(y_expected) + n_states * 2
for x1, x2 in zip(y_merged, y_expected):
self.assertAllClose(x1, x2, atol=1e-5)
y_merged = y_merged[-n_states * 2:]
y_forward = y_forward[-n_states:]
y_backward = y_backward[-n_states:]
for state_birnn, state_inner in zip(y_merged,
y_forward + y_backward):
self.assertAllClose(state_birnn, state_inner, atol=1e-5)
def test_rnn_cell_with_constants_layer(self):
class RNNCellWithConstants(keras.layers.Layer):
def __init__(self, units, **kwargs):
self.units = units
self.state_size = units
super(RNNCellWithConstants, self).__init__(**kwargs)
def build(self, input_shape):
if not isinstance(input_shape, list):
raise TypeError('expects constants shape')
[input_shape, constant_shape] = input_shape
# will (and should) raise if more than one constant passed
self.input_kernel = self.add_weight(shape=(input_shape[-1],
self.units),
initializer='uniform',
name='kernel')
self.recurrent_kernel = self.add_weight(
shape=(self.units, self.units),
initializer='uniform',
name='recurrent_kernel')
self.constant_kernel = self.add_weight(
shape=(constant_shape[-1], self.units),
initializer='uniform',
name='constant_kernel')
self.built = True
def call(self, inputs, states, constants):
[prev_output] = states
[constant] = constants
h_input = keras.backend.dot(inputs, self.input_kernel)
h_state = keras.backend.dot(prev_output, self.recurrent_kernel)
h_const = keras.backend.dot(constant, self.constant_kernel)
output = h_input + h_state + h_const
return output, [output]
def get_config(self):
config = {'units': self.units}
base_config = super(RNNCellWithConstants, self).get_config()
return dict(list(base_config.items()) + list(config.items()))
with self.cached_session():
# Test basic case.
x = keras.Input((None, 5))
c = keras.Input((3, ))
cell = RNNCellWithConstants(32)
layer = keras.layers.RNN(cell)
y = layer(x, constants=c)
model = keras.models.Model([x, c], y)
model.compile(optimizer='rmsprop', loss='mse')
model.train_on_batch([np.zeros(
(6, 5, 5)), np.zeros((6, 3))], np.zeros((6, 32)))
with self.cached_session():
# Test basic case serialization.
x_np = np.random.random((6, 5, 5))
c_np = np.random.random((6, 3))
y_np = model.predict([x_np, c_np])
weights = model.get_weights()
config = layer.get_config()
custom_objects = {'RNNCellWithConstants': RNNCellWithConstants}
with keras.utils.CustomObjectScope(custom_objects):
layer = keras.layers.RNN.from_config(config.copy())
y = layer(x, constants=c)
model = keras.models.Model([x, c], y)
model.set_weights(weights)
y_np_2 = model.predict([x_np, c_np])
self.assertAllClose(y_np, y_np_2, atol=1e-4)
with self.cached_session():
# test flat list inputs.
with keras.utils.CustomObjectScope(custom_objects):
layer = keras.layers.RNN.from_config(config.copy())
y = layer([x, c])
model = keras.models.Model([x, c], y)
model.set_weights(weights)
y_np_3 = model.predict([x_np, c_np])
self.assertAllClose(y_np, y_np_3, atol=1e-4)
with self.cached_session():
# Test stacking.
cells = [
keras.layers.recurrent.GRUCell(8),
RNNCellWithConstants(12),
RNNCellWithConstants(32)
]
layer = keras.layers.recurrent.RNN(cells)
y = layer(x, constants=c)
model = keras.models.Model([x, c], y)
model.compile(optimizer='rmsprop', loss='mse')
model.train_on_batch([np.zeros(
(6, 5, 5)), np.zeros((6, 3))], np.zeros((6, 32)))
with self.cached_session():
# Test GRUCell reset_after property.
x = keras.Input((None, 5))
c = keras.Input((3, ))
cells = [keras.layers.recurrent.GRUCell(32, reset_after=True)]
layer = keras.layers.recurrent.RNN(cells)
y = layer(x, constants=c)
model = keras.models.Model([x, c], y)
model.compile(optimizer='rmsprop', loss='mse')
model.train_on_batch([np.zeros(
(6, 5, 5)), np.zeros((6, 3))], np.zeros((6, 32)))
with self.cached_session():
# Test stacked RNN serialization
x_np = np.random.random((6, 5, 5))
c_np = np.random.random((6, 3))
y_np = model.predict([x_np, c_np])
weights = model.get_weights()
config = layer.get_config()
with keras.utils.CustomObjectScope(custom_objects):
layer = keras.layers.recurrent.RNN.from_config(config.copy())
y = layer(x, constants=c)
model = keras.models.Model([x, c], y)
model.set_weights(weights)
y_np_2 = model.predict([x_np, c_np])
self.assertAllClose(y_np, y_np_2, atol=1e-4)
def test_training_pandas(self):
try:
import pandas as pd # pylint: disable=g-import-not-at-top
except ImportError:
self.skipTest('Skipping test because pandas is not installed.')
input_a = keras.Input(shape=(3, ), name='input_a')
input_b = keras.Input(shape=(3, ), name='input_b')
input_c = keras.Input(shape=(1, ), name='input_b')
x = keras.layers.Dense(4, name='dense_1')(input_a)
y = keras.layers.Dense(3, name='dense_2')(input_b)
z = keras.layers.Dense(1, name='dense_3')(input_c)
model_1 = keras.Model(inputs=input_a, outputs=x)
model_2 = keras.Model(inputs=[input_a, input_b], outputs=[x, y])
model_3 = keras.Model(inputs=input_c, outputs=z)
model_1.compile(optimizer='rmsprop', loss='mse')
model_2.compile(optimizer='rmsprop', loss='mse')
input_a_np = np.random.random((10, 3))
input_b_np = np.random.random((10, 3))
input_a_df = pd.DataFrame(input_a_np)
input_b_df = pd.DataFrame(input_b_np)
output_a_df = pd.DataFrame(np.random.random((10, 4)))
output_b_df = pd.DataFrame(np.random.random((10, 3)))
model_1.fit(input_a_df, output_a_df)
model_2.fit([input_a_df, input_b_df], [output_a_df, output_b_df])
model_1.fit([input_a_df], [output_a_df])
model_1.fit({'input_a': input_a_df}, output_a_df)
model_2.fit({
'input_a': input_a_df,
'input_b': input_b_df
}, [output_a_df, output_b_df])
model_1.evaluate(input_a_df, output_a_df)
model_2.evaluate([input_a_df, input_b_df], [output_a_df, output_b_df])
model_1.evaluate([input_a_df], [output_a_df])
model_1.evaluate({'input_a': input_a_df}, output_a_df)
model_2.evaluate({
'input_a': input_a_df,
'input_b': input_b_df
}, [output_a_df, output_b_df])
# Verify predicting on pandas vs numpy returns the same result
predict_1_pandas = model_1.predict(input_a_df)
predict_2_pandas = model_2.predict([input_a_df, input_b_df])
predict_3_pandas = model_3.predict(input_a_df[0])
predict_1_numpy = model_1.predict(input_a_np)
predict_2_numpy = model_2.predict([input_a_np, input_b_np])
predict_3_numpy = model_3.predict(np.asarray(input_a_df[0]))
self.assertAllClose(predict_1_numpy, predict_1_pandas)
self.assertAllClose(predict_2_numpy, predict_2_pandas)
self.assertAllClose(predict_3_numpy, predict_3_pandas)
# Extra ways to pass in dataframes
model_1.predict([input_a_df])
model_1.predict({'input_a': input_a_df})
model_2.predict({'input_a': input_a_df, 'input_b': input_b_df})
