Posted by P. P. on June 28, 19103 at 03:10:45:
In Reply to: stop criterium posted by Bart Hamers on June 06, 19102 at 18:31:02:
The problem is the validation error is not decreasing monotonically before it reaches the valley.
: I have a question on the stopping criterium of adaboost. I assume that to many rounds in your boosting algorithm lead to overfitting. Therefore, is it a good idea to monitor your performance on an extra validation set? (Like this is typically done in neural networks). So, you can stop the boosting once the error on the validation set increases.