John Zedlewski comments

Results 6 comments of


                                            John Zedlewski

[FEA] Add support for computing feature_importances in RF

Definitely agreed. Not sure we'll have enough bandwidth to get this for 0.19 (given work going to new backend) but should be highly prioritized high after that.

[REVIEW] remove no_bad_cuml_array_check, no longer needed in 0.17

Hi @jpurviance - looks great! Did you get a chance to merge with the latest changes or open a new PR?

TSNE: Isolate condition that freezes on certain data & providing warning/assertion

Thanks for filing! It seems you're using CUDA 11.2. CUDA 11.2 is not tested or supported at this time. Can you try again with CUDA 11 and see if you...

TSNE: Isolate condition that freezes on certain data & providing warning/assertion

@divyegala can you take a look and try to repro?

[DOC] More guidance about parallel usage for many small problems on single GPU

Thanks for the detailed issue, @frankier! A couple of things here: * Performance with multiple streams is not well-optimized yet. It's on our radar and something we're hoping to improve...

[FEA] Simplify process to train cuml KMeans on GPU and save the model and later load on a CPU machine for inference

We will consider this a feature request for simplification of this process in a future release (and documenting better). Thank you for filing!