def model(self):
x1 = L.Input(shape=(None, ))
x2 = L.Input(shape=(self.hidden_size, ))
x3 = L.Input(shape=(self.hidden_size, ))
return tf.keras.Model(inputs=[x1, x2, x3],
outputs=self.call(x1, x2, x3),
name='AttGRUDecoder')
def build_model_arc(self) -> None:
output_dim = self.label_processor.vocab_size
config = self.hyper_parameters
embed_model = self.embedding.embed_model
layer_stack = [
L.Bidirectional(L.GRU(**config['layer_bi_gru'])),
L.Dense(output_dim, **config['layer_output']),
self._activation_layer()
]
tensor = embed_model.output
for layer in layer_stack:
tensor = layer(tensor)
self.tf_model = keras.Model(embed_model.inputs, tensor)
def build_model_arc(self) -> None:
output_dim = self.label_processor.vocab_size
config = self.hyper_parameters
embed_model = self.embedding.embed_model
layer_stack = [
L.Bidirectional(L.LSTM(**config['layer_blstm']), name='layer_blstm'),
L.Dropout(**config['layer_dropout'], name='layer_dropout'),
L.Dense(output_dim, **config['layer_time_distributed']),
L.Activation(**config['layer_activation'])
]
tensor = embed_model.output
for layer in layer_stack:
tensor = layer(tensor)
self.tf_model = keras.Model(embed_model.inputs, tensor)
def build_model_arc(self) -> None:
output_dim = self.label_processor.vocab_size
config = self.hyper_parameters
embed_model = self.embedding.embed_model
# build model structure in sequent way
layer_stack = [
L.Conv1D(**config['conv1d_layer']),
L.GlobalMaxPooling1D(**config['max_pool_layer']),
L.Dense(**config['dense_layer']),
L.Dense(output_dim, **config['layer_output']),
self._activation_layer()
]
tensor = embed_model.output
for layer in layer_stack:
tensor = layer(tensor)
self.tf_model = keras.Model(embed_model.inputs, tensor)
def build_model_arc(self) -> None:
output_dim = self.label_processor.vocab_size
config = self.hyper_parameters
embed_model = self.embedding.embed_model
crf = KConditionalRandomField()
layer_stack = [
L.Bidirectional(L.GRU(**config['layer_bgru']), name='layer_bgru'),
L.Dropout(**config['layer_dropout'], name='layer_dropout'),
L.Dense(output_dim, **config['layer_time_distributed']), crf
]
tensor = embed_model.output
for layer in layer_stack:
tensor = layer(tensor)
self.tf_model = keras.Model(embed_model.inputs, tensor)
self.crf_layer = crf
def __init__(self,
embedding: ABCEmbedding,
vocab_size: int,
hidden_size: int = 1024):
super(AttGRUDecoder, self).__init__()
self.embedding = embedding
self.hidden_size = hidden_size
self.gru = tf.keras.layers.GRU(hidden_size,
return_sequences=True,
return_state=True,
recurrent_initializer='glorot_uniform')
self.fc = tf.keras.layers.Dense(vocab_size)
# 用于注意力
self.attention = L.BahdanauAttention(hidden_size)
def _activation_layer(self) -> L.Layer:
if self.multi_label:
return L.Activation('sigmoid')
else:
return L.Activation('softmax')
def model(self) -> tf.keras.Model:
x1 = L.Input(shape=(None, ))
x2 = L.Input(shape=(self.hidden_size, ))
return tf.keras.Model(inputs=[x1, x2],
outputs=self.call(x1, x2),
name='GRUEncoder')