def test_merge_4d(self):
layers_to_be_tested = [Add, Maximum, Multiply, Average, Concatenate]
N, C, H, W = 2, 2, 1, 3
x1 = create_tensor(N, C, H, W)
x2 = create_tensor(N, C, H, W)
for layer in layers_to_be_tested:
input1 = Input(shape=(H, W, C))
input2 = Input(shape=(H, W, C))
output = layer()([input1, input2])
model = Model(inputs=[input1, input2], outputs=output)
model.compile(optimizer='adagrad', loss='mse')
self._test_one_to_one_operator_core_channels_last(model, [x1, x2])
def test_recursive_model(self):
N, C, D = 2, 3, 3
x = create_tensor(N, C)
sub_input1 = Input(shape=(C, ))
sub_mapped1 = Dense(D)(sub_input1)
sub_model1 = Model(inputs=sub_input1, outputs=sub_mapped1)
sub_input2 = Input(shape=(C, ))
sub_mapped2 = Dense(D)(sub_input2)
sub_model2 = Model(inputs=sub_input2, outputs=sub_mapped2)
input1 = Input(shape=(D, ))
input2 = Input(shape=(D, ))
mapped1_2 = sub_model1(input1)
mapped2_2 = sub_model2(input2)
sub_sum = Add()([mapped1_2, mapped2_2])
keras_model = Model(inputs=[input1, input2], output=sub_sum)
try:
coreml_model = coremltools.converters.keras.convert(keras_model)
except (AttributeError, ImportError) as e:
warnings.warn(
"Unable to test due to an error in coremltools '{0}'".format(
e))
return
onnx_model = onnxmltools.convert_coreml(coreml_model)
x = [x, 2 * x]
y_reference = keras_model.predict(x)
y_produced = evaluate_deep_model(onnx_model, x).reshape(N, D)
def test_sequential_model_with_multiple_operators(self):
N, C, H, W = 2, 3, 5, 5
x = create_tensor(N, C, H, W)
model = Sequential()
model.add(
Conv2D(2,
kernel_size=(1, 2),
strides=(1, 1),
padding='valid',
input_shape=(H, W, C),
data_format='channels_last'))
model.add(Activation('relu'))
model.add(
Conv2D(2,
kernel_size=(1, 2),
strides=(1, 1),
padding='valid',
input_shape=(H, W, C),
data_format='channels_last'))
model.add(Activation('relu'))
model.add(
MaxPooling2D((2, 2), strides=(2, 2), data_format='channels_last'))
model.compile(optimizer='adagrad', loss='mse')
self._test_one_to_one_operator_core_channels_last(model, x)
def test_recursive_and_shared_model(self):
N, C, D = 2, 3, 3
x = create_tensor(N, C)
sub_input1 = Input(shape=(C, ))
sub_mapped1 = Dense(D)(sub_input1)
sub_output1 = Activation('sigmoid')(sub_mapped1)
sub_model1 = Model(inputs=sub_input1, outputs=sub_output1)
sub_input2 = Input(shape=(C, ))
sub_mapped2 = sub_model1(sub_input2)
sub_output2 = Activation('tanh')(sub_mapped2)
sub_model2 = Model(inputs=sub_input2, outputs=sub_output2)
input1 = Input(shape=(D, ))
input2 = Input(shape=(D, ))
mapped1_1 = Activation('tanh')(input1)
mapped2_1 = Activation('sigmoid')(input2)
mapped1_2 = sub_model1(mapped1_1)
mapped1_3 = sub_model1(mapped1_2)
mapped2_2 = sub_model2(mapped2_1)
sub_sum = Add()([mapped1_3, mapped2_2])
model = Model(inputs=[input1, input2], outputs=sub_sum)
# coremltools can't convert this kind of model.
self._test_one_to_one_operator_keras(model, [x, 2 * x])
def test_merge_2d(self):
# Skip Concatenate for now because CoreML Concatenate needs 4-D input
layers_to_be_tested = [Add, Maximum, Multiply, Average, Dot]
N, C = 2, 3
x1 = create_tensor(N, C)
x2 = create_tensor(N, C)
for layer in layers_to_be_tested:
input1 = Input(shape=(C, ))
input2 = Input(shape=(C, ))
if layer == Dot:
result = layer(axes=-1)([input1, input2])
else:
result = layer()([input1, input2])
model = Model(inputs=[input1, input2], outputs=result)
model.compile(optimizer='adagrad', loss='mse')
self._test_one_to_one_operator_core(model, [x1, x2])
def test_reshape(self):
N, C, H, W = 2, 3, 1, 2
x = create_tensor(N, C, H, W)
keras_model = Sequential()
keras_model.add(Reshape((1, C * H * W, 1), input_shape=(H, W, C)))
keras_model.compile(optimizer='adagrad', loss='mse')
self._test_one_to_one_operator_core_channels_last(keras_model, x)
def test_convolution_transpose_2d(self):
N, C, H, W = 2, 2, 1, 1
x = create_tensor(N, C, H, W)
input = Input(shape=(H, W, C))
result = Conv2DTranspose(2, (2, 1), data_format='channels_last')(input)
model = Model(inputs=input, outputs=result)
model.compile(optimizer='adagrad', loss='mse')
self._test_one_to_one_operator_core_channels_last(model, x)
def test_upsample(self):
N, C, H, W = 2, 3, 1, 2
x = create_tensor(N, C, H, W)
input = Input(shape=(H, W, C))
result = UpSampling2D(input)
model = Model(inputs=input, outputs=result)
model.compile(optimizer='adagrad', loss='mse')
self._test_one_to_one_operator_core_channels_last(model, x)
def test_pooling_4d(self):
layers_to_be_tested = [MaxPooling2D, AveragePooling2D]
N, C, H, W = 1, 2, 4, 3
x = create_tensor(N, C, H, W)
for layer in layers_to_be_tested:
input = Input(shape=(H, W, C))
result = layer(2, data_format='channels_last')(input)
model = Model(inputs=input, outputs=result)
model.compile(optimizer='adagrad', loss='mse')
self._test_one_to_one_operator_core_channels_last(model, x)
def test_dense_with_dropout(self):
N, C, D = 2, 3, 2
x = create_tensor(N, C)
input = Input(shape=(C, ))
hidden = Dense(D, activation='relu')(input)
result = Dropout(0.2)(hidden)
keras_model = Model(inputs=input, outputs=result)
keras_model.compile(optimizer='sgd', loss='mse')
self._test_one_to_one_operator_core_keras(keras_model, x)
def test_batch_normalization(self):
N, C, H, W = 2, 2, 3, 4
x = create_tensor(N, C, H, W)
model = Sequential()
input = Input(shape=(H, W, C))
result = BatchNormalization(
beta_initializer='random_uniform',
gamma_initializer='random_uniform',
moving_mean_initializer='random_uniform',
moving_variance_initializer=RandomUniform(minval=0.1, maxval=0.5),
)(input)
model = Model(inputs=input, outputs=result)
model.compile(optimizer='adagrad', loss='mse')
self._test_one_to_one_operator_core_channels_last(model, x)
def test_conv_4d(self):
N, C, H, W = 1, 2, 4, 3
x = create_tensor(N, C, H, W)
input = Input(shape=(H, W, C))
result = Conv2D(2,
kernel_size=(1, 2),
strides=(1, 1),
padding='valid',
input_shape=(H, W, C),
data_format='channels_last')(input)
model = Model(inputs=input, outputs=result)
model.compile(optimizer='adagrad', loss='mse')
self._test_one_to_one_operator_core_channels_last(model, x)
def test_activation_2d(self):
activation_to_be_tested = [
'tanh', 'relu', 'sigmoid', 'softsign', 'elu', 'softplus', LeakyReLU
]
N, C = 2, 3
x = create_tensor(N, C)
for activation in activation_to_be_tested:
input = Input(shape=(C, ))
if isinstance(activation, str):
result = Activation(activation)(input)
else:
result = activation()(input)
model = Model(inputs=input, outputs=result)
model.compile(optimizer='adagrad', loss='mse')
self._test_one_to_one_operator_core(model, x)
def test_flatten(self):
N, C, H, W, D = 2, 3, 1, 2, 2
x = create_tensor(N, C, H, W)
keras_model = Sequential()
keras_model.add(Flatten(input_shape=(H, W, C)))
keras_model.add(Dense(D))
keras_model.compile(optimizer='adagrad', loss='mse')
try:
coreml_model = coremltools.converters.keras.convert(keras_model)
except ImportError:
warnings.warn("Issue in coremltools.")
return
onnx_model = onnxmltools.convert_coreml(coreml_model)
y_reference = keras_model.predict(np.transpose(x, [0, 2, 3, 1]))
y_produced = evaluate_deep_model(onnx_model, x).reshape(N, D)
self.assertTrue(np.allclose(y_reference, y_produced))
def test_dense(self):
N, C, D = 2, 3, 2
x = create_tensor(N, C)
input = Input(shape=(C, ))
result = Dense(D)(input)
keras_model = Model(inputs=input, outputs=result)
keras_model.compile(optimizer='adagrad', loss='mse')
try:
coreml_model = coremltools.converters.keras.convert(keras_model)
except (AttributeError, ImportError) as e:
warnings.warn(
"Unable to test due to an error in coremltools '{0}'.".format(
e))
return
onnx_model = onnxmltools.convert_coreml(coreml_model)
y_reference = keras_model.predict(x)
y_produced = evaluate_deep_model(onnx_model, x).reshape(N, D)
self.assertTrue(np.allclose(y_reference, y_produced))
