I was reviewing the answer to the question about the "Gradient of Hinge Loss" here: and noticed that one of the answers proposed returning of averaged weights in function:

def grad_descent(): 
return np.sum(ws,1)/np.size(ws,1)

I m curious if there is any special reason for averaging weights? Why not to find best weights by using the validating set on each epoch?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.