99fd2f42d951c0e1e46c5c556d9c2f87a0bda733,deeplift/blobs.py,Pool2D,_get_mxts_increments_for_inputs,#Pool2D#,730

Before Change

    def _get_mxts_increments_for_inputs(self):
        input_act_vars = self._get_input_activation_vars() 
        return B.pool2d_grad(
                pool_out=self.get_mxts(),
                pool_in=input_act_vars,
                pool_size=self.pool_size,
                strides=self.strides,
                border_mode=self.border_mode,
                ignore_border=self.ignore_border,
                pool_mode=self.pool_mode
            )


class Flatten(SingleInputMixin, OneDimOutputMixin, Node):
    

After Change

    def _get_mxts_increments_for_inputs(self):
        input_act_vars = self._get_input_activation_vars() 

        out_grads = self.get_mxts()
        if (self.pool_mode == B.PoolMode.max):
            //For maxpooling, an addiitonal scale factor may be needed
            //in case all the inputs don"t have the same reference.
            //multiply by diff-from-default of output here,
            //and divide by diff-from-default of output later
            out_grads = out_grads*self._get_diff_from_default_vars()

        to_return = B.pool2d_grad(
                        out_grad=out_grads,
                        pool_in=input_act_vars,
                        pool_size=self.pool_size,
                        strides=self.strides,
                        border_mode=self.border_mode,
                        ignore_border=self.ignore_border,
                        pool_mode=self.pool_mode)

        if (self.pool_mode == B.PoolMode.max):
            //rescale back down according to diff-from-default of inputs
            pseudocounted_inp_diff_default = pseudocount_near_zero(to_return)
            to_return = to_return/pseudocounted_inp_diff_default 

        return to_return


class Flatten(SingleInputMixin, OneDimOutputMixin, Node):
    

Instances