[SYSTEMDS-3951] MM Chain Optimization w/ Basic Average Estimator#2525
[SYSTEMDS-3951] MM Chain Optimization w/ Basic Average Estimator#2525ywcb00 wants to merge 8 commits into
Conversation
|
Thank you very much @Baunsgaard. This was exactly what I was searching for and I changed the test validation accordingly. :) Does anybody know why the builtin.part2 java test resulted in a timeout and is this behavior related to the current PR? Test output states: If this is not related to the PR, I would like to merge the changes soon. All the best, |
|
Baunsgaard
left a comment
There was a problem hiding this comment.
few Nits, otherwise LGTM
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #2525 +/- ##
============================================
+ Coverage 71.54% 71.61% +0.06%
- Complexity 50240 50270 +30
============================================
Files 1623 1623
Lines 194317 194314 -3
Branches 37965 37965
============================================
+ Hits 139032 139149 +117
+ Misses 44366 44264 -102
+ Partials 10919 10901 -18 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
840e84d to
76d9d52
Compare
76d9d52 to
bad0d56
Compare
…): delete content of method to get the input matrices since it does not work and replace it with a not implemented exception for now chore(main/hops/rewrite/ProgramRewriteStatus.java): remove unused methods and member variables to make clear that the program rewrite status does not hold a variable map
… leaf with its metadata only change predicate method to check whether it is a leaf accordingly
… check if the mm chain was rewritten with respect to sparsity by checking for the respective keyword in the trace log chore(test/AutomatedTestBase.java): add check if buffer is null when checking if a buffer contains a certain string
…stimator by the basic average estimator for matrix multiplication chain rewrites NOTE: Missing information about the matrix histograms of input data caused certain limitations for the implementation with the mnc estimator.
… capture and validate the log output to remove the necessity of having a console print in the code
…tighter checks for correct rewrites based on sparsity estimates
…apply proper codestyle refactor(test/functions/rewrite/RewriteMatrixMultChainOptSparseTest.java): apply proper codestyle
…): apply codestyle
ada0809 to
6aa1556
Compare
Hi,
this PR changes the estimator for sparse mm chain rewrites from MNC to the metadata average estimator. Integration of MNC for these rewrites is missing the information about the underlying matrix histograms.
This PR also fixes the corresponding test, which previously only validated if there was any mm chain optimization.
Thank you and all the best,
David