* Added stacked what where autoencoder.
SWWAE uses residual blocks. Trains fast. Creates very good reconstructions.
* Added newline at end for PEP8
* Went through PEP8 errors and corrected all (except for the imports which following the numpy seed, but this should be ok). Also, for the pool_size of 2, we halved the number of features maps and the number of epochs, and it still trains a net that can very nicely reconstruct the input.
* Added spaces arround - and + when they are used as binary operators (more PEP8).
* In decoder, the index of the features and pool size and wheres are all equal to nlayers-1-i, so set ind variable to this value and passed it to them.
* With ind variable in decoder, don't need two lines for the upsampling layer.
* Added title to plot, got rid of ticks on plot.
* PEP8 for * binary operator. Corrected some grammar issues in the docstring.