Implementing backprop of conv1d with stride_tricks

Question

I\'m learning how to use numpy\'s stride_tricks module. So I\'m using it to implement 1d convolution, and since I\'m at it, I\'m trying to implement backpropaga