rank2plan 0.2.0

No project description provided

rank2plan

Implementation of constraint generation and column generation for solving large L1-RankSVMs with hinge loss (with pair-specific gaps) and sample weights. This is based on the work by Dedieu et al (2022) on solving large L1-SVMs with hinge loss. See documents/theory.pdf for how we extend their work. The "2plan" part of the package name comes from the tool being used to learn heuristics for planning.

Installation

Install with

pip install rank2plan

The PyPI release only supports 3.8 <= python <= 3.10 for now.

Examples

See under tests for examples.

Todo

• Support for sparse matrices and vectors
• Bayesian optimisation for tuning the C value.
• We log pretty aggressively, probably should add a verbosity control

References

• A. Dedieu, R. Mazumder, and H. Wang. Solving L1-regularized SVMs and Related Linear Programs: Revisiting the Effectiveness of Column and Constraint Generation. J. Mach. Learn. Res., 23:164:1–164:41, 2022. [URL].