Use L-BFGS-B Fortran library for native logistic regression benchmark #8
Conversation
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
native/Makefile
Outdated
| @@ -26,10 +27,19 @@ all: $(addprefix bin/,$(BENCHMARKS)) | |||
| bin: | |||
| mkdir -p bin | |||
|
|
|||
| bin/log_reg_lbfgs: log_reg_lbfgs_bench.cpp $(FOBJ) | bin | |||
| $(CXX) $^ $(CXXINCLUDE) $(CXXFLAGS) $(LDFLAGS) -lmkl_rt -lm -lifcore \ | |||
There was a problem hiding this comment.
Linking against mkl_rt, are we making sure that MKL is using TBB as the underlying threading layer?
MKL's default is to use OpenMP, as since DAAL's default is to use TBB we end up incurring the cost of runtime of both.
One can either set the threading layer in the benchmark itself, by calling mkl_set_threading_layer, or use explicit dynamic linking, see MKL Linking Advisor.
There was a problem hiding this comment.
I added a call to mkl_set_threading_layer in native/log_reg_lbfgs_bench.cpp.
bibikar
added a commit
that referenced
this pull request
Jun 27, 2019
This PR adds logistic regression replicating the results of sklearn.linear_model.LogisticRegression but implemented in daal4py. It supports solvers lbfgs and newton-cg. test_fit uses the canonical form of logistic regression for the binary case (which in scikit-learn is multi_class='ovr') and the multinomial (softmax of exponentiated scores, multi_class='multinomial') for the multi-class case. test_predict supports any combination of n_classes and multi_class, but we use the same multi_class used in test_fit. While we cannot directly use daal4py's logistic regression because it isn't as easy as native DAAL to pass in a custom solver (see #8), we use daal4py's logistic regression objective functions and math primitives to compute logistic regression.
razdoburdin
pushed a commit
to razdoburdin/scikit-learn_bench
that referenced
this pull request
Jun 13, 2023
optional build mode without distributed mode
razdoburdin
pushed a commit
to razdoburdin/scikit-learn_bench
that referenced
this pull request
Jun 13, 2023
* adding cycling notebook example * Making adjustments to the Jupyter notebook * adding in main landing page changes * Adding in examples page * Fixing examples, formatting, and wording * fixing removed items and typos in example code
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.
This PR swaps out the native logistic regression benchmark's solver for the same one used in SciPy. L-BFGS-B is wrapped in a minimal DAAL
optimization_solverclass and directly set as the solver for DAAL's logistic regression. This brings native performance for this benchmark to match our optimized scikit-learn performance for a sample binary classification problem with 100k samples and 1k features.Because we must now directly call BLAS and LAPACK functions instead of using DAAL's internal BLAS and LAPACK implementations, we add a dependency on MKL as well.