question archive Imagine we are running stochastic gradient descent, and we determine the stopping criteria based on the validation set
Imagine we are running stochastic gradient descent, and we determine the stopping criteria based on the validation set
Subject:MathPrice: Bought3
Imagine we are running stochastic gradient descent, and we determine the stopping criteria based on the validation set. Suppose the stopping criteria is if the validation set error increases (i.e. stop SGD as soon as the validation set error starts increasing). Is this a good stopping criteria? If not, how would you fix it?
