SdA.py implements a stacked de-noising autoencoder in the tutorial style. I think DBN/SdA are a bit different from the rest, so if anybody has the time to read through it and give some feedback I would appreciate it; One choice that I made was to fix the number of layers and not use any sort of looping over the layers [ I thought it makes things clearer but also longer .. comments ? should I use a for loop ? ]; Another thing was that I tried to stick to the same names for variables as in the paper on SdA but these names are not always obvious (for example the hidden layer is named y) [ should I change this and name the variable hidden ?]

Author: pascanur