In the SciKit documentation of the MLP classifier, there is the early_stopping flag which allows to stop the learning if there is not any improvement in several iterations. However, it does not seem specified if the best weights found are restored or the final weights fo the model are those obtained at the last iteration. If not, is possible to restore the best weights found?
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
To restore the best weights you would need a way to monitor a metric of your choosing and keep track of the best weights up to a given point. This has nothing to do with early stopping, you could return the best weights regardless of how many iterations the classifier has performed. Unfortunately, as far as I know, this functionality is not supported by scikit-learn.
In early stopping you assume that the best weights are those that the point you stopped your training. Early stopping isn't as much a convergence speed technique as it is a regularization, one.
