Roman Lutz

Thanks for submitting this issue @millerjohnp! I'm very much in favor of this project since we're not doing a good job at providing the information on why this dataset may...

@millerjohnp did you have any success with OpenML? We may have people willing to capture the most important information from the paper regarding the original Adult dataset and the new...

No worries! @kspieks and @britneyting will probably try to come up with a first version of a documentation through the user guide. This should be based on the findings in...

#614 does this.

I have no objection. But let's not use ROC as an acronym here. It will 100% get confused with the more commonly used receiver operating characteristic 🙂

Thanks @bramreinders97 ! This sounds like postprocessing with special treatment in this particular region. Many unfairness mitigation algorithms have this "advantaged"/"disadvantaged" or "deprived" terminology which is binary AFAICT and doesn't...

Thanks for the APIs! I'd imagine the theta would be set in the constructor or learned during `fit`, rather than provided in `predict`. Is there a `fit` method? I'm also...

I suppose `fit` could be a no-op, no harm done... I agree with @hildeweerts on ensembles being handled externally so that we can just assume we have a classifier. >...

Could we theoretically accept regressors and then threshold on the predictions? Because for ThresholdOptimizer we're doing that rather than expecting probabilities from predict_proba. I guess the same handling as we're...

That's not quite what I meant. Let me explain a bit better: Right now you're saying you'll accept values from 0 to 1 and then you'll put a threshold on...