What exactly do transform(X) method? about concrete-autoencoders

[mfbalin]

K is the number of features you want to select, and so after you transform, what you get is the input with only the selected features.

[FarshidShekari]

this is my code:

selector.fit(X_tr, Y_tr, X_va, Y_va)
X_tr = selector.transform(X_tr)
print(X_tr.shape)

My trainning data has 20000 sample and 250 features, but after transform I get X_tr with shape (k,250) and not (20000,k), what's the problem?

[mfbalin]

Can you confirm that X_tr has the shape (20000, 250)?

[mfbalin]

Actually, I looked at the code and there is a bug, you can use this instead:
X_tr[:, selector.get_support(indices=true)]

[FarshidShekari]

Can you confirm that X_tr has the shape (20000, 250)?

Yes

[FarshidShekari]

Actually, I looked at the code and there is a bug, you can use this instead:
X_tr[:, selector.get_support(indices=true)]

I want to transform my data to new feature space because I want to feed it to my DNN.

[nasimulhasan]

@FarshidShekari could you find a solution to transform your data into a new feature space?

[dcorr039]

Hello I have a question about figure 1 from your paper. How is reconstruction supposed to be done? I only understand up to
part c where k=20 can be used to mask the original image. How is it you can reconstruct image as in part d.