0
$\begingroup$

I have a NN with about 750 inputs (sparsely encoded text, mostly). The output is 1 cell - yes or no, this is of type X.

I can guess all day long on the size and depth of my hidden layers. I hate guessing; it's lame and worse, inefficient.

Will someone offer me guidelines on how to determine appropriate sizes and numbers of hidden layers?

Improve this question
$\endgroup$
1
  • 2
    $\begingroup$ if you really have no prior to encode in the architecture, then usually you start with 2/3 layers of 128 neurons and Adam optimizer, but i'd suggest you to first analyze the data with PCA to understand how uniform it is, and possibly use the first N principal component and some ML model on such subspace $\endgroup$
    – Alberto
    Commented Jul 9 at 20:49

0

Reset to default

You must log in to answer this question.

Browse other questions tagged .