Consider extending the GridSearchCV class in order to include warm start (when available) and preprocessing?
Closed this issue · 2 comments
matteobarbieri commented
This could solve the 'RangeScaler' issue and the fact that no warm-start strategy is implemented.
fdtomasi commented
I found a main issue of sklearn
scikit-learn/scikit-learn#1674
in which they discuss the possibility to add a warm-start strategy to GridSearchCV.
It doesn't seem to be clear on how structure a GridSearchCV with warm-start strategy, mainly because of how GridSearchCV is implemented, its parallel jobs and so on.
fdtomasi commented
See also this scikit-learn/scikit-learn#1626
It is not clear on how to implement it.
I close this since this is not an issue of PALLADIO, which is independent on the estimator used.