97031ed8f6afaf26a21dd98bb0bfdb4cacc65ca2,keras/layers/wrappers.py,Bidirectional,compute_output_shape,#Bidirectional#Any#,277

Before Change

    def compute_output_shape(self, input_shape):
        if self.merge_mode in ["sum", "ave", "mul"]:
            return self.forward_layer.compute_output_shape(input_shape)
        elif self.merge_mode == "concat":
            shape = list(self.forward_layer.compute_output_shape(input_shape))
            shape[-1] *= 2
            return tuple(shape)
        elif self.merge_mode is None:
            return [self.forward_layer.compute_output_shape(input_shape)] * 2

    def call(self, inputs, training=None, mask=None):

After Change

            output_shape = output_shape[0]

        if self.merge_mode == "concat":
            output_shape = list(output_shape)
            output_shape[-1] *= 2
            output_shape = tuple(output_shape)
        elif self.merge_mode is None:
            output_shape = [output_shape] * 2

        if self.return_state:
            if self.merge_mode is None:
                return output_shape + state_shape * 2
            return [output_shape] + state_shape * 2
        return output_shape

    def call(self, inputs, training=None, mask=None, initial_state=None):
        kwargs = {}
        if has_arg(self.layer.call, "training"):

Instances