Closed kosmitive closed 7 months ago
I would say that balancing class imbalance is the responsibility of the user, and therefore so is wrapping the model to do whatever makes sense for it. I don't think pydvl needs to provide any kind of support for this. I would agree to using imbalanced data in an example to illustrate the problem, thouh. Or do you think there is some non-trivial situation that we should handle?
Stale issue: awaiting OPs reply for 30 days
This issue was closed because it has been stalled for 7 days with no activity.
While it is clear for models which have a
class_weights
parameter in the constructor, some models only support settingsample_weights
infit
. We have two optionsfit
method.fit
is called, addsample_weights
to accept imbalanced datasets. Which method would you prefer?
I would tend for 2., because the interface
fit
is more compact and reusable.