Book above
What is implemented here is method (4), which uses attention.
def build_model(self):
""""""
encoder_input = layers.Input(shape=(self.input_seq_len,))
encoder_embeding = layers.Embedding(input_dim=len(self.en_word_id_dict),
output_dim=self.encode_embeding_len,
# mask_zero=True
)(encoder_input)
encoder_lstm, state_h, state_c = layers.LSTM(units=self.encode_lstm_hidden_len,
return_state=True, return_sequences=True)(encoder_embeding)
decoder_input = layers.Input(shape=(self.output_seq_len,), name="decoder_input")
decoder_attention_a = encoder_lstm
encoder_state_h = state_h
encoder_state_c = state_c
outputs = []
decoder_embeding = layers.Embedding(input_dim=len(self.ch_word_id_dict),
output_dim=self.decode_embeding_len,
# mask_zero=True
name="decoder_embeding"
)(decoder_input)
decoder_lstm_layer = layers.LSTM(self.decode_embeding_len,
return_state=True, name="decoder_lstm_layer")
decoder_embeding_slice_layer = layers.Lambda(lambda x: x[:, t, :], name="decoder_embeding_slice_layer")
decoder_embeding_slice_repeat_layer = layers.RepeatVector(1, name="decoder_embeding_slice_repeat_layer")
attention_layer = AttentionLayer_3()
concat_layer = layers.Concatenate(axis=-1, name="concat_layer")
for t in range(self.output_seq_len):
decoder_embeding_slice =decoder_embeding_slice_layer(decoder_embeding)
decoder_embeding_slice = decoder_embeding_slice_repeat_layer(decoder_embeding_slice)
decoder_lstm, h, c = decoder_lstm_layer(decoder_embeding_slice, initial_state=[encoder_state_h, encoder_state_c])
encoder_state_h = h
encoder_state_c = c
decoder_lstm_att = attention_layer([decoder_attention_a, decoder_lstm])
decoder_lstm = decoder_embeding_slice_repeat_layer(decoder_lstm)
decoder_lstm = concat_layer([decoder_lstm_att, decoder_lstm])
outputs.append(decoder_lstm)
outputs = layers.Concatenate(axis=1)(outputs)
decoder_out = layers.Dense(len(self.ch_word_id_dict), activation="softmax", name="decoder_out")(outputs)
model = Model([encoder_input, decoder_input], decoder_out)
model.compile(optimizer='rmsprop', loss='categorical_crossentropy')
model.summary()
return model
what is love? One point is also love!
