Neural network processing regression problems better model and loss function have?

CodePudding user response:

Feeling to make an issue of the loss function (1) to define a new optimization criteria, after all your data gap is too big, if use the mean square error (mse) small data fitting effect will be poor 2. Add some regular items in the loss function, but your regular item to design is very clever, or owe fitting (3) can do it for all y log processing and fitting for the model structure itself, I think is to deepen a road network