sentence_cnn_encoder.py 文件源码

def bidirectional_rnn(self) -> Tuple[Tuple[tf.Tensor, tf.Tensor],
                                         Tuple[tf.Tensor, tf.Tensor]]:
        # BiRNN Network
        fw_cell, bw_cell = self.rnn_cells()  # type: RNNCellTuple
        seq_lens = tf.ceil(tf.divide(
            self.input_sequence.lengths,
            self.segment_size))
        seq_lens = tf.cast(seq_lens, tf.int32)
        return tf.nn.bidirectional_dynamic_rnn(
            fw_cell, bw_cell, self.highway_layer,
            sequence_length=seq_lens,
            dtype=tf.float32)