0
$\begingroup$

Is there a rule of thumb for selecting for a neural network or an autoencoder:

(a) Number of hidden neurons

(b) Number of hidden layers

(c) In general, to begin applying a machine learning algorithm is there a statistical method to select the number of features or those features which are more relevant? Intuitively, if the entropy of a feature is high then the information content of that feature is high. So we should select that feature. However, I have no idea how to calculate entropy of continuous valued single feature. Therefore, there must be other ways to determine which feature is more relevant out of a pool of several features?

$\endgroup$

marked as duplicate by Sycorax, Michael Chernick, kjetil b halvorsen, Peter Flom Apr 12 '18 at 12:26

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.

0
$\begingroup$

Lots of good answers in comments link but in practice for me works:

A) 3 or 4 times your features input

B) keep adding layers until you don't see any improvement

C)I do a correlation matrix to have an idea.

$\endgroup$

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