I have built a 3 layer neural network to perform a binary mapping (2016 inputs, 288 outputs.) I am getting decent results with mean square error and stochastic gradient decent. My question is: Is there a more appropriate loss function for regression when the output is binary?
$\begingroup$
$\endgroup$
-
$\begingroup$ Hmm, for binary output it is more convenient to have sigmoid (softmax is not appropriate in this case, I guess) output and cross-entropy loss function. $\endgroup$ – Dmytro Prylipko Jan 11 '19 at 20:41
-
$\begingroup$ What is meant by binary? Are you saying two outputs always or what? Share the output format as well. $\endgroup$ – vipin bansal Feb 6 at 10:36
$\begingroup$
$\endgroup$
Yes, use binary cross-entropy loss. In case you are using Keras, this has been already implemented as a standard loss function for binary outputs.
