When I have N samples of 2 different features (different scale, different meaning etc.), but whose values are scalar numbers of the same format.

Does it make sense to flatten the data into a 1D array of length [N*2] and feed it in this 1D shape into the neural network. Or must I keep them separate (as columns or otherwise)? I will standardize each feature before combining them into the single array.

(I want to do this for convenience reasons i.e. not deal with multidimensional arrays).

The data is input for a classification problem.

  • $\begingroup$ Can you be more specific about your classification problem? Are you supposed to classify each sample separately? Also, note that not wanting to deal with multi-dimensional arrays for convenience is unwise and, in my opinion, not a good excuse. You might be needing to improve your grasp of ND arrays in your language of choice for this (separately, of course). $\endgroup$ – E_net4 the flagger Aug 10 '17 at 9:15

I believe it does, for example, if you flatten both features into 1-D array and then concatenate them, they should still preserve most of the information. Because total number of feature vector still remains the same.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.