Skip to content
master
Switch branches/tags
Code

SSVMjulia

Cutting-plane Structured SVM in Julia, solving the Primal QP of the 1-slack formulation

This implementation has the following advantages over SVM-struct-matlab:

  • Constraint caching (reuse full inference results)
  • Parallelization of the inference step (one process per example)
  • Additional positivity constraints on arbitrary coordinates of w

The two first points are very beneficial in settings were the inference task takes most of the time.

See SSVMjulia-tutorial.ipynb for a tutorial in IJulia notebook.

Reference

Joachims, T., Finley, T., & Yu, C. N. J. (2009). Cutting-plane training of structural SVMs. Machine Learning, 77(1), 27-59.

About

Structural SVMs in Julia

Resources

License

Releases

No releases published

Packages

No packages published