Reminder: Phase I EA logic had avoid_repeated_params as an argument controlling how exceptions were raised if EA could not select a set of parameters that had not been already tried.
We don't want to rerun identical parameter sets, obviously, but I think this is now handled by way of dask_searchcv base classes to avoid repeated parameters?
On the other hand our wrapping of those base classes may have stepped around the logic in the base classes for avoiding repeated parameters? Also consider the case that we may support multiple paired X, y input samples to a group of estimators in EaSearchCV or other multi-sample ideas where the uniqueness of a parameter set may be defined by a combination or parameter values and an identifier of a the subsample to which they have been fit.
Reminder: Phase I EA logic had
avoid_repeated_paramsas an argument controlling how exceptions were raised if EA could not select a set of parameters that had not been already tried.dask_searchcvbase classes to avoid repeated parameters?X, yinput samples to a group of estimators inEaSearchCVor other multi-sample ideas where the uniqueness of a parameter set may be defined by a combination or parameter values and an identifier of a the subsample to which they have been fit.