Skip to content
Approximation path and optimal selection of regularization hyperparameter for some machine learning problems.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Safe Grid Search with Optimal Complexity.

This package implements a safe grid search strategy for hyperparameter tunning. We use a coordinate descent algorithm to solve the optimization algorithms (though the choice of coordinate descent is "arbitrary" here: indeed any other optimization solver can be used as soon as it can output a duality gap). The approximation path algorithms and their complexity are based only on the duality gap computation to quantify the optimality of a vector at a given regularization parameter.

See the paper for more details.

Installation & Requirements

We first need to install the optimization solvers safegridoptim and the compilation proceed as follows:

$ pip install -e .

This package has the following requirements:

We recommend to install or update anaconda (at least version 0.16.1).


  • Figure 1 is generated by (time<1s.)
  • Figure 2 is generated by (time<1s.)
  • Figure 3 (a) is generated by (time<3s. - make_regression(100, 500))
  • Figure 3 (b) is generated by (time<60s. - make_classification(100, 500))
  • Figure 4 (a) and (b) are generated by (time<2s. for a; time<2mn. for b;)
  • Figure 5 (appendix) is generated by
You can’t perform that action at this time.