Not able to reproduce results.

[prateek-malhotra]

In the latest lolopy version (1.2.0), I fixed random_seed but still, results are not reproducible (I have already fixed numpy random seed). Can you please fix it or tell me the reason for this?

[bfolie]

Hi Prateek. Lolo training is not entirely reproducible because the base learners are trained in parallel and we don't use splittable random numbers. This is a known deficiency, but I realize we don't have an issue for it, so I opened one: #259.

I also did a sweep to make sure random number generators were being used everywhere, and I found a bug that is corrected in #258. This bug would only have affected you if you were considering a subset of features at each split (the default for regression is to consider all features for each split, in which case this bug would not have affected you).

The parallelization issue prevents full reproducibility, but it should be an extremely small effect. If it's causing your predictions to vary significantly relative to the error bars, then please say so.