Method prepare_input

caffe2/python/rnn_cell.py:213–224

If some operations in _apply method depend only on the input, not on recurrent states, they could be computed in advance. model: ModelHelper object new operators would be added to input_blob: either the whole input sequence with shape (sequence_length, batc

(self, model, input_blob)

        raise NotImplementedError('Abstract method')

    def prepare_input(self, model, input_blob):
        '''
        If some operations in _apply method depend only on the input,
        not on recurrent states, they could be computed in advance.

        model: ModelHelper object new operators would be added to

        input_blob: either the whole input sequence with shape
        (sequence_length, batch_size, input_dim) or a single input with shape
        (1, batch_size, input_dim).
        '''
        return input_blob

    def get_output_state_index(self):
        '''