BUG fix BatchedLevelAlgo DtClsTest & DtRegTest failing tests #3690
Conversation
This file contains 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
* updated to new method for quantiles computation
* deleted unused tempmem
venkywonka
added
bug
teju85
approved these changes
Apr 1, 2021
Codecov Report
@@ Coverage Diff @@
## branch-0.19 #3690 +/- ##
===============================================
+ Coverage 80.70% 82.48% +1.77%
===============================================
Files 227 227
Lines 17615 17541 -74
===============================================
+ Hits 14217 14469 +252
+ Misses 3398 3072 -326
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
JohnZed
approved these changes
Apr 1, 2021
|
@gpucibot merge |
rapids-bot bot
pushed a commit
that referenced
this pull request
Apr 4, 2021
…ogram calculation and occupancy optimization (#3616) * This PR introduces: * A faster way to calculate the histograms containing splits in the `ML::DecisionTree::computeSplitClassificationKernel` . These histograms are used for node-splitting in decision trees for the task of classification. * A change in the default `gridDim.x` in the launch configuration of the above kernel from `4` to based on occupancy calculator and other dimension gridDims, thus improving the occupancy to theoretical limits * Earlier too many atomic adds to shared memory limited the kernel times, which has been avoided by blockwide sum-scans to obtain the same histogram using fewer atomic writes to shared memory. * The resulting kernel time speedups are significant (upto 30x for some nodes) * `computeSplitRegressionKernel` has different share-memory write patterns that deserves it's own PR for optimization 😬 * Tests will pass once #3690 is merged Authors: - Venkat (https://github.com/venkywonka) Approvers: - AJ Schmidt (https://github.com/ajschmidt8) - Philip Hyunsu Cho (https://github.com/hcho3) - Thejaswi. N. S (https://github.com/teju85) - John Zedlewski (https://github.com/JohnZed) URL: #3616
8 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
BatchedLevelAlgo/DtClsTestFandBatchedLevelAlgo/DtRegTestFwherein the quantiles parameter passed togrow_treefunction was uninitialized garbage memory as opposed to what should have been quantiles computed for each column.preprocess_quantiles) with new, more accurate one (computeQuantiles)tempmemin the setup phase of the test fixture.BatchedLevelAlgo/DtRegTestFtests as reported in issue [BUG] CUDA 11.2 libcuml++ C++ test failures EDIT: Updated with 11.2 update 2 #3406BatchedLevelAlgo/DtClsTestFtests in PR ENH Decision Tree new backendcomputeSplitClassificationKernelhistogram calculation and occupancy optimization #3616cc @teju85 @vinaydes @JohnZed @hcho3