Update GPUTreeshap #6163
Update GPUTreeshap #6163
Conversation
Codecov Report
@@ Coverage Diff @@
## master #6163 +/- ##
==========================================
+ Coverage 78.83% 78.91% +0.08%
==========================================
Files 12 12
Lines 3090 3102 +12
==========================================
+ Hits 2436 2448 +12
Misses 654 654
Continue to review full report at Codecov.
|
| unsigned ntree_limit, | ||
| std::vector<bst_float>* tree_weights, | ||
| bool approximate) override { | ||
| if (approximate) { |
There was a problem hiding this comment.
We need dispatching in gbm get predictor.
There was a problem hiding this comment.
There might be a benefit to explicitly failing, maybe the user is expecting the algorithm to use GPU and it is silently switching to CPU.
| import unittest | ||
| import pytest | ||
|
|
||
| try: |
There was a problem hiding this comment.
Can we move this into testing.py?
There was a problem hiding this comment.
I prefer it to stay here, I don't think it will be reused.
RAMitchell
force-pushed
the
treeshap
branch
from
386b375
to
4d3fdd3
Compare
|
Can we close #5875 by this PR? |
Enable feature interactions
Reduce python test time
Add integration test with shap library