Shruti Nath
Shruti Nath
Thanks @myenugula, looks good to me as well, will wait for the changelog modification
From an initial pass LGTM, all tests pass for me as well. Will wait for @jeremiedbb to review but otherwise looks good to go!
I adapted the stochastic sample weighting test to check the performance when sample_weight.sum()
> > I adapted the stochastic sample weighting test to check the performance when sample_weight.sum() > Is it absolute `max_samples=1` (int), which would mean fit each estimator with only one...
Note: in the test_logistic_regression_sample_weight sag and saga are left out as they systematically fail
This hopefully helps sample weight handling in the non-deterministic case (i.e., when subsample is specified). Nevertheless, in the deterministic case when subsample is at the default 200_000 value we get...
I performed more tests and added a "use_weights_in_resampling" so we can easily check both before making a decision. I also realised that it is sensitive to the subsampling portion, before...
> So, basically, for small number of bins there is no statistical difference between the methods. > > But I would say that there is no statistical test that will...
Agreed as well thanks @jeremiedbb, I suppose if this is the case we need to check the impact on HistGradietBoostingRegressor since it also uses resampling (and if so we may...
Yes of course... I meant that if we incorporate the changes i.e., add weighted resampling into the resample function then this may cause HistGradientBoostingRegressor to break so not sure if...