Hey Sagar, I updated the code of the Flatten Layer, I made a small mistake.

I wrote :

self.output = input_data.flatten().reshape((-1,1))

Instead of :

self.output = input_data.flatten().reshape((1,-1))

I corrected it anyway, so if you look at the gist it should be the right version now.

Regarding the MNIST the dataset. Here is a sample working code. I didn’t use a MaxPool Layer, it worked fine without it.

I used the same ConvLayer that is in the article. The link for the mathematics behind this layer is also in the article.

Feel free to ask any question ;)