Open lucagiovagnoli opened 4 years ago
Hi @komiya-atsushi did you have a chance to look at this?
I've been working on getting xgboost-predictor-java integrated into the MLeap project (here's the PR: https://github.com/combust/mleap/pull/645).
It'd be great to apply the Yelp's tolerance bugfix for issue #21 (parity lost on complex models) to your project. Our alternative is to adopt the Yelp's fork but I'd like to avoid duplicating work and contribute the fix upstream.
Hi @komiya-atsushi,
This is in relation to this tolerance bug https://github.com/komiya-atsushi/xgboost-predictor-java/issues/21 that affects correctness of results on complex models. We have since fixed it in our fork at Yelp and achieved complete parity of results with xgboost4j.
We're now wondering if you would be open to PRs so that we can contribute this upstream. Just asking before we do the work and open the PR :)