[ENH] should we have a reset
method?
#1614
Labels
API design
API design & software architecture
implementing framework
Implementing or improving framework for learning tasks, e.g., base class functionality
Should
sktime
estimators have areset
method that sets the estimator's entire state to what it would be after initialization?The main reason is that this would be a clean way to reset the estimator at the beginning of
fit
whenever called for the second or any repeated time.This would prevent bugs such as in #1595 which can lead to unexpected and hard-to-debug behaviour in grid search etc, where
fit
might be called multiple times.Of course, a "clean" implementation would use
clone
, but not all contributors/extenders are aware of the convention.A more implementer friendly logic is the described behaviour, where
fit
resets the estimator when called a second time - this is "like" a call tosklearn
clone
, but it requires no proactivity and advanced knowledge ofsklearn
internals.FYI @mloning, @aiwalter, @ltsaprounis, @danbartl, @TonyBagnall, opinions?
The text was updated successfully, but these errors were encountered: