def build(self, hp, inputs=None):
input_node = nest.flatten(inputs)[0]
output_node = input_node
block_type = self.block_type or hp.Choice(
'block_type', ['resnet', 'xception', 'vanilla'], default='resnet')
normalize = self.normalize
if normalize is None:
normalize = hp.Choice('normalize', [True, False], default=True)
augment = self.augment
if augment is None:
augment = hp.Choice('augment', [True, False], default=True)
if normalize:
output_node = preprocessor.Normalization()(output_node)
if augment:
output_node = preprocessor.ImageAugmentation(
seed=self.seed)(output_node)
sub_block_name = self.name + '_' + block_type
if block_type == 'resnet':
output_node = block.ResNetBlock(name=sub_block_name)(output_node)
elif block_type == 'xception':
output_node = block.XceptionBlock(name=sub_block_name)(output_node)
elif block_type == 'vanilla':
output_node = block.ConvBlock(name=sub_block_name)(output_node)
return output_node
def test_conv_block():
input_shape = (32, 32, 3)
block = block_module.ConvBlock()
hp = kerastuner.HyperParameters()
block.build(hp, ak.Input(shape=input_shape).build())
assert name_in_hps('kernel_size', hp)
assert name_in_hps('num_blocks', hp)
assert name_in_hps('separable', hp)
def build(self, hp, inputs=None):
input_node = nest.flatten(inputs)[0]
output_node = input_node
block_type = self.block_type or hp.Choice(
'block_type', ['resnet', 'xception', 'vanilla'], default='resnet')
if block_type == 'resnet':
output_node = block.ResNetBlock()(output_node)
elif block_type == 'xception':
output_node = block.XceptionBlock()(output_node)
elif block_type == 'vanilla':
output_node = block.ConvBlock().build(output_node)
return output_node
def assemble(self, input_node):
output_node = input_node
ratio = self.sw_ratio()
if not isinstance(input_node, node.TextNode):
raise ValueError('The input_node should be a TextNode.')
if ratio < 1500:
output_node = processor.TextToNgramVector()(output_node)
output_node = block.DenseBlock()(output_node)
else:
output_node = processor.TextToIntSequence()(output_node)
output_node = block.EmbeddingBlock(
pretrained=(ratio < 15000))(output_node)
output_node = block.ConvBlock(separable=True)(output_node)
return output_node
def build(self, hp, inputs=None):
input_node = nest.flatten(inputs)[0]
output_node = input_node
vectorizer = self.vectorizer or hp.Choice(
'vectorizer', ['sequence', 'ngram'], default='sequence')
if not isinstance(input_node, node.TextNode):
raise ValueError('The input_node should be a TextNode.')
if vectorizer == 'ngram':
output_node = preprocessor.TextToNgramVector()(output_node)
output_node = block.DenseBlock()(output_node)
else:
output_node = preprocessor.TextToIntSequence()(output_node)
output_node = block.EmbeddingBlock(
pretraining=self.pretraining)(output_node)
output_node = block.ConvBlock(separable=True)(output_node)
return output_node
def test_type_error_for_call():
block = block_module.ConvBlock()
with pytest.raises(TypeError) as info:
block(block)
assert 'Expect the inputs to layer' in str(info.value)
