def downsample(inputs,
pool_type: str = 'max',
sorted: bool = True,
stage: int = 1): # noqa: A002
layers_pool = []
if pool_type == 'max':
layers_pool.append(
L.MaxPooling1D(pool_size=3,
strides=2,
padding='same',
name=f'pool_{stage}'))
elif pool_type == 'k_max':
k = int(inputs.shape[1].value / 2)
layers_pool.append(
KMaxPoolingLayer(k=k, sorted=sorted, name=f'pool_{stage}'))
elif pool_type == 'conv':
layers_pool.append(
L.Conv1D(filters=inputs.shape[-1].value,
kernel_size=3,
strides=2,
padding='same',
name=f'pool_{stage}'))
layers_pool.append(L.BatchNormalization())
elif pool_type is None:
layers_pool = []
else:
raise ValueError(f'unsupported pooling type `{pool_type}`!')
tensor_out = inputs
for layer in layers_pool:
tensor_out = layer(tensor_out)
return tensor_out
def build_model_arc(self):
output_dim = len(self.pre_processor.label2idx)
config = self.hyper_parameters
embed_model = self.embedding.embed_model
layers_rcnn_seq = [
L.SpatialDropout1D(**config['spatial_dropout']),
L.Bidirectional(L.GRU(**config['rnn_0'])),
L.Conv1D(**config['conv_0'])
]
layers_sensor = [
L.GlobalMaxPooling1D(),
AttentionWeightedAverageLayer(),
L.GlobalAveragePooling1D()
]
layer_concat = L.Concatenate(**config['concat'])
layers_full_connect = [
L.Dropout(**config['dropout']),
L.Dense(**config['dense']),
L.Dense(output_dim, **config['activation_layer'])
]
tensor = embed_model.output
for layer in layers_rcnn_seq:
tensor = layer(tensor)
tensors_sensor = [layer(tensor) for layer in layers_sensor]
tensor_output = layer_concat(tensors_sensor)
for layer in layers_full_connect:
tensor_output = layer(tensor_output)
self.tf_model = tf.keras.Model(embed_model.inputs, tensor_output)
def build_model_arc(self):
output_dim = len(self.pre_processor.label2idx)
config = self.hyper_parameters
embed_model = self.embedding.embed_model
layer_embed_dropout = L.SpatialDropout1D(**config['spatial_dropout'])
layers_conv = [L.Conv1D(**config[f'conv_{i}']) for i in range(4)]
layers_sensor = [KMaxPoolingLayer(**config['maxpool_i4']), L.Flatten()]
layer_concat = L.Concatenate(**config['merged_tensor'])
layers_seq = []
layers_seq.append(L.Dropout(**config['dropout']))
layers_seq.append(L.Dense(**config['dense']))
layers_seq.append(L.Dense(output_dim, **config['activation_layer']))
embed_tensor = layer_embed_dropout(embed_model.output)
tensors_conv = [layer_conv(embed_tensor) for layer_conv in layers_conv]
tensors_sensor = []
for tensor_conv in tensors_conv:
tensor_sensor = tensor_conv
for layer_sensor in layers_sensor:
tensor_sensor = layer_sensor(tensor_sensor)
tensors_sensor.append(tensor_sensor)
tensor = layer_concat(tensors_sensor)
# tensor = L.concatenate(tensors_sensor, **config['merged_tensor'])
for layer in layers_seq:
tensor = layer(tensor)
self.tf_model = tf.keras.Model(embed_model.inputs, tensor)
def build_model_arc(self):
output_dim = len(self.pre_processor.label2idx)
config = self.hyper_parameters
embed_model = self.embedding.embed_model
layers_seq = []
layers_seq.append(L.Conv1D(**config['conv_layer']))
layers_seq.append(L.MaxPooling1D(**config['max_pool_layer']))
layers_seq.append(L.GRU(**config['gru_layer']))
layers_seq.append(L.Dense(output_dim, **config['activation_layer']))
tensor = embed_model.output
for layer in layers_seq:
tensor = layer(tensor)
self.tf_model = tf.keras.Model(embed_model.inputs, tensor)
def build_model_arc(self):
output_dim = len(self.pre_processor.label2idx)
config = self.hyper_parameters
embed_model = self.embedding.embed_model
layers_region = [
L.Conv1D(**config['region_embedding']),
L.BatchNormalization(),
L.PReLU(),
L.Dropout(**config['region_dropout'])
]
layers_main = [
L.GlobalMaxPooling1D(),
L.Dense(**config['dense']),
L.BatchNormalization(),
L.PReLU(),
L.Dropout(**config['dropout']),
L.Dense(output_dim, **config['activation'])
]
tensor_out = embed_model.output
# build region tensors
for layer in layers_region:
tensor_out = layer(tensor_out)
# build the base pyramid layer
tensor_out = conv_block(tensor_out, **config['conv_block'])
# build the above pyramid layers while `steps > 2`
seq_len = tensor_out.shape[1].value
if seq_len is None:
raise ValueError(
'`sequence_length` should be explicitly assigned, but it is `None`.'
)
for i in range(floor(log2(seq_len)) - 2):
tensor_out = resnet_block(tensor_out,
stage=i + 1,
**config['resnet_block'])
for layer in layers_main:
tensor_out = layer(tensor_out)
self.tf_model = tf.keras.Model(embed_model.inputs, tensor_out)
def build_model_arc(self):
output_dim = len(self.pre_processor.label2idx)
config = self.hyper_parameters
embed_model = self.embedding.embed_model
layer_embed_dropout = L.SpatialDropout1D(**config['spatial_dropout'])
layers_conv = [L.Conv1D(**config[f'conv_{i}']) for i in range(4)]
layers_sensor = []
layers_sensor.append(L.GlobalMaxPooling1D())
layers_sensor.append(AttentionWeightedAverageLayer())
layers_sensor.append(L.GlobalAveragePooling1D())
layer_view = L.Concatenate(**config['v_col3'])
layer_allviews = L.Concatenate(**config['merged_tensor'])
layers_seq = []
layers_seq.append(L.Dropout(**config['dropout']))
layers_seq.append(L.Dense(**config['dense']))
layers_seq.append(L.Dense(output_dim, **config['activation_layer']))
embed_tensor = layer_embed_dropout(embed_model.output)
tensors_conv = [layer_conv(embed_tensor) for layer_conv in layers_conv]
tensors_matrix_sensor = []
for tensor_conv in tensors_conv:
tensor_sensors = []
tensor_sensors = [
layer_sensor(tensor_conv) for layer_sensor in layers_sensor
]
# tensor_sensors.append(L.GlobalMaxPooling1D()(tensor_conv))
# tensor_sensors.append(AttentionWeightedAverageLayer()(tensor_conv))
# tensor_sensors.append(L.GlobalAveragePooling1D()(tensor_conv))
tensors_matrix_sensor.append(tensor_sensors)
tensors_views = [
layer_view(list(tensors))
for tensors in zip(*tensors_matrix_sensor)
]
tensor = layer_allviews(tensors_views)
# tensors_v_cols = [L.concatenate(tensors, **config['v_col3']) for tensors
# in zip(*tensors_matrix_sensor)]
# tensor = L.concatenate(tensors_v_cols, **config['merged_tensor'])
for layer in layers_seq:
tensor = layer(tensor)
self.tf_model = tf.keras.Model(embed_model.inputs, tensor)
def build_model_arc(self):
"""
build model architectural
"""
output_dim = len(self.pre_processor.label2idx)
config = self.hyper_parameters
embed_model = self.embedding.embed_model
layer_conv = L.Conv1D(**config['layer_conv'], name='layer_conv')
layer_lstm = L.LSTM(**config['layer_lstm'], name='layer_lstm')
layer_dropout = L.Dropout(**config['layer_dropout'],
name='layer_dropout')
layer_time_distributed = L.TimeDistributed(
L.Dense(output_dim, **config['layer_time_distributed']),
name='layer_time_distributed')
layer_activation = L.Activation(**config['layer_activation'])
tensor = layer_conv(embed_model.output)
tensor = layer_lstm(tensor)
tensor = layer_dropout(tensor)
tensor = layer_time_distributed(tensor)
output_tensor = layer_activation(tensor)
self.tf_model = keras.Model(embed_model.inputs, output_tensor)
def conv_block(inputs,
filters: int,
kernel_size: int = 3,
activation: str = 'linear',
shortcut: bool = True):
layers_conv_unit = []
layers_conv_unit.append(L.BatchNormalization())
layers_conv_unit.append(L.PReLU())
layers_conv_unit.append(
L.Conv1D(filters=filters,
kernel_size=kernel_size,
strides=1,
padding='same',
activation=activation))
layers_conv_block = layers_conv_unit * 2
tensor_out = inputs
for layer in layers_conv_block:
tensor_out = layer(tensor_out)
if shortcut:
tensor_out = L.Add()([inputs, tensor_out])
return tensor_out