Overcoming obstacles in running program

[mortonjt]

Hello, thank you for making this program available.

I was able to get your adversarial autoencoder working, but it did require quite a bit of modification do so, namely

1. Modifying BaseData.getitem to np.int64 -- sample_x, sample_y -> sample_x, sample_y.astype(np.int64)
2. Forcing the batch ids to be 1-based indexed - there were some really weird indexing errors popping up in _forward_autoencode when 0-based indexed batch ids (which is naturally outputted from sklearn.preprocessing.LabelEncoder).
3. Removing preprocessing steps -- they weren't relevant for my data, since I already preprocessed them.

(1) looks like a bug to me - and having example datasets on the github repo to test NormAE would have been tremendously useful, since it would also help users validate their install. (2) could benefit from better documentation. (3) is probably un-necessary given how many different preprocessing procedures there are.

[luyiyun]

Thanks for your advises. I will soon fix these problems.

[S-KD]

@luyiyun, I agree with @mortonjt please provide example datasets from your publication to test NormAE.