A PROC HPNEURAL model is trained with MLPARCH(20 10) and the VALIDATION= option. Training loss decreases monotonically but validation loss begins rising at epoch 15. What does PROC HPNEURAL do when the VALIDATION= option is specified?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Here's the deal — b is correct because PROC HPNEURAL with a VALIDATION= dataset monitors validation error and implements early stopping, saving the model state at the epoch with the lowest validation error. The final model uses these weights, not the last-iteration weights.
Full explanation below image
Full Explanation
B is correct because PROC HPNEURAL with a VALIDATION= dataset monitors validation error and implements early stopping, saving the model state at the epoch with the lowest validation error. The final model uses these weights, not the last-iteration weights. A is wrong; PROC HPNEURAL does not blindly run to MAXITER when VALIDATION= is provided. C is wrong; PROC HPNEURAL does not restart automatically with a new seed. D is wrong; learning rate decay is not an automatic response to validation loss increase in PROC HPNEURAL.