Reductive

Training of optimized product quantizers

Training of optimized product quantizers requires a LAPACK implementation. For this reason, training of the OPQ and GaussianOPQ quantizers is feature-gated by the opq-train feature. This feature must be enabled if you want to use OPQ or GaussianOPQ:

~~~toml [dependencies] reductive = { version = "0.7", features = ["opq-train"] } ~~~

This also requires that a crate that links a LAPACK library is added as a dependency, e.g. accelerate-src, intel-mkl-src, openblas-src, or netlib-src.

Running tests

Linux

You can run all tests on Linux, including tests for optimized product quantizers, using the intel-mkl-test feature:

~~~shell $ cargo test --features intel-mkl-test ~~~

macOS

All tests can be run on macOS with the accelerate-test feature:

~~~shell $ cargo test --features accelerate-test ~~~

Multi-threaded OpenBLAS

reductive uses Rayon to parallelize quantizer training. However, multi-threaded OpenBLAS is known to conflict with application threading. Is you use OpenBLAS, ensure that threading is disabled, for instance by setting the number of threads to 1:

~~~shell $ export OPENBLASNUMTHREADS=1 ~~~