Update SVC tests for CCCL update#6569
Merged
rapids-bot[bot] merged 8 commits intorapidsai:branch-25.06from Apr 25, 2025
Merged
Conversation
viclafargue
force-pushed
the
fix-svc-test_svm_numeric_arraytype-test
branch
from
15ee169 to
f4dc0ac
Compare
divyegala
approved these changes
Apr 24, 2025
Contributor
|
I'm merging this to unblock CCCL updates to 2.8.x. Thanks @viclafargue and @divyegala for all your help! We can re-enable these tests after rapidsai/rapids-cmake#793 is merged. |
Contributor
|
/merge |
Ofek-Haim
pushed a commit
to Ofek-Haim/cuml
that referenced
this pull request
May 13, 2025
This test was failing after the introduction of CCCL 2.8 in rapidsai#6421. The cause was [a non-deterministic minimum search in the SmoBlockSolve kernel](https://github.com/rapidsai/cuml/blob/b01e2d1d4cf358dbc0a435a663c546fa9cf3e3cb/cpp/src/svm/smoblocksolve.cuh#L206) with the `cub::BlockReduce` feature of the CCCL library. During the first iteration of the first launch of the kernel, the values being searched are set to -1.0 making the result of the search undefined/non-deterministic. The minimum search in this kernel produces a different result on this first iteration when libcuml is compiled with CCCL 2.8 rather than version 2.7 (probably because of underlying algorithms updates) making the training of SVC produce different results.. Since the value of the `gamma` hyper-parameter does not seem to be critical to this pytest test, I modified its value to stabilize the test. Also updated the expected values in the SVC gtests. Authors: - Victor Lafargue (https://github.com/viclafargue) - Divye Gala (https://github.com/divyegala) Approvers: - Divye Gala (https://github.com/divyegala) URL: rapidsai#6569
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This test was failing after the introduction of CCCL 2.8 in #6421. The cause was a non-deterministic minimum search in the SmoBlockSolve kernel with the
cub::BlockReducefeature of the CCCL library. During the first iteration of the first launch of the kernel, the values being searched are set to -1.0 making the result of the search undefined/non-deterministic. The minimum search in this kernel produces a different result on this first iteration when libcuml is compiled with CCCL 2.8 rather than version 2.7 (probably because of underlying algorithms updates) making the training of SVC produce different results..xfailed the pytests and commented out the gtests for now.