h2o4gpu
h2o4gpu copied to clipboard
Kmeans||
This is still WIP, for early review.
TODOs:
- Make the algorithm work as expected.
- Add proper tests.
- Make the algorithm work with k-means.
- Factor out operations.
- Fix and rebase build-system.
- Merge tSVD matrix with KmMatrix.
Above and tuning the algorithm described in #642
close #642
All tests for KmMatrix should pass.
I will at least get CMake part of the build system get merged. I will work on it as soon as I can see the jenkin report.
@RAMitchell Ready for another review. \O/ This time I re-based on a build-system branch that contains only CMake file changes, hopefully better than the last time.
The example about using thrust to generate random number is not implemented yet. The PR is getting way~~ too heavy now. I might need the build log to get the build system right. If that is not gonna be available soon, then maybe we or someone else can handle the build later. The code builds on Fedora 27 with CUDA9.2, if there's error on your local system, please send the complete log to me.
We can switch to doing some work on xgboost after you have dealt with the review issues until the continuous integration problem gets resolved.
@trivialfis @RAMitchell just had a quick glance, changes look really good! Will do a proper review over the weekend.
@RAMitchell by CI issues you mean @trivialfis not being able to see the logs? Would a log emailed when the build fails be sufficient? I remember adding that a long time ago but maybe someone changed it and need to add it back in.
@mdymczyk Thanks for spending extra time on this. I haven't receive any mail from the CI yet.