jfilter / sparse-svd-benchmark

Sparse Truncated SVD Benchmark (Python)

Sparse Truncated SVD Benchmark (Python)

Key findings:

• sparsesvd is really slow and thus was not considered for the graphs
• There is almost not differnce between the implementation of randomized SVD with Gensim SVD and Scikit-Learn SVD
• MKL is faster than OpenBlas for the randomized SVD (not clearly vissible in the graphs, sorry for this)
• Randomized SVD is faster as the problem gets more difficult

Setup

Done on i3 8100, Python 3.6, Ubuntu 18.04, average of 5 runs, with the recent version of OpenBlas and MKL as of 18th June 2019.

Check out the accompanying notebook.

Results

OpenBlas	MKL

Acknowledgements

Based on previous work:

• https://simplyml.com/benchmarking-the-singular-value-decomposition/
• https://github.com/charanpald/tyre-hug/blob/master/tyrehug/exp/svdbenchmark.py

License

MIT.

Sponsoring

This work was created as part of a project that was funded by the German Federal Ministry of Education and Research.