I run a multilayer perceptron model in Weka. I used 10-fold cross validation to validate the model. In order to prevent the model from overfitting, I put "30" in validation set size. Which meant, I used 30% of my training data for validation. Is this the correct way to apply early stopping procedure in WEKA?
Dear Jaya,
I'm not an expert in the weak framework, but, I think your early stopping process is not correct. Early stopping consists of stopping the training when some criterion was been achieved, for instance, accuracy, loss, etc...
Dear artur, in validating the error, WEKA used the validation threshold parameter, which I set equal to 20. From what I understand, WEKA will stop the learning if the number of consequetive error that occur is more than 20.
Hi,
The answer is Yes. In WEKA, the other parameter relevant for early stopping is -E, which determines the number of consecutive increases in error that are tolerated before early stopping is performed.
HTH.
Dr. Samer Sarsam
- Badges
- Science topic
- Similar topics
- Correction