Improve sklearnex relative perf in sklearn_example.json#209
Improve sklearnex relative perf in sklearn_example.json#209ethanglaser wants to merge 2 commits intoIntelPython:mainfrom
Conversation
| "centers": 2, | ||
| "n_samples": 1000, | ||
| "n_samples": 5000, | ||
| "n_features": [16, 64] |
There was a problem hiding this comment.
Do we need the case with 16 features?
There was a problem hiding this comment.
I think this may have been originally added to demonstrate to users that configs can take lists instead of just individual values - so I'd be in favor of keeping it as is
|
@avolkov-intel with this diff: Seems like kd_tree does have better speedup for predict, but poor for fit (which matters for kd_tree) - for brute there is negligible compute happening in fit so that being red is not really a problem, therefore I am not sure kd_tree is better overall. LinReg predict is marginally better - I think it may be tough to get significant speedups here. And as for kmeans, I think part of the example config is to show off different features of the config, one of which is ignoring like it has currently - so we can add it or not. |
I guess let's keep kd_tree for KNN and ignore the rest of the suggestions |
Description
Before:
After:
Checklist:
Completeness and readability
Testing