Benchmark repository for MCP

Benchopt is a package to simplify and make more transparent and reproducible the comparisons of optimization algorithms. Regression with the Minimax Concave Penalty (MCP) consists in solving the following program:

$$\min_w \frac{1}{2 n} \Vert y - Xw \Vert^2_2 \ + \ \sum_j \rho_{\lambda, \gamma}(w_j)$$

with the penalty

$$ \rho_{\lambda, \gamma} (t) = \begin{cases} \lambda \vert t \vert - \frac{t^2}{2\gamma} & , & \text{ if } \vert t \vert \ \leq \ \gamma \lambda \\ \frac{\lambda^2 \gamma}{2} & , & \text{ if } \vert t \vert \ > \ \gamma \lambda \end{cases}$$

where $n$ (or n_samples) stands for the number of samples, $p$ (or n_features) stands for the number of features and

$$y \in \mathbb{R}^n, \ X \in \mathbb{R}^{n \times p}$$

Install

This benchmark can be downloaded using the following commands:

$ pip install -U benchopt
$ git clone https://github.com/benchopt/benchmark_mcp
$ cd benchmark_mcp/

To demonstrate the use of benchopt on a simple benchmark configuration, one can run, from the benchmark_lasso folder:

$ benchopt install . -s cd -s pgd --env
$ benchopt run . --config example_config.yml --env

Alternatively, one can use the command line interface to select which objectives, datasets and solvers are used:

$ benchopt run -s cd -s pgd -d simulated --max-runs 10 --n-repetitions 5

Use benchopt run -h for more details about these options, or visit https://benchopt.github.io/api.html.

Troubleshooting

If you run into some errors when running the examples present in this Readme, try installing the development version of benchopt:

$ pip install -U git+https://github.com/benchopt/benchopt

If issues persist, you can also try running the benchmark in local mode with the -l option, e.g.:

$ benchopt run . -l -s cd -d simulated --max-runs 10 --n-repetitions 10

Use benchopt run -h for more details about these options, or visit https://benchopt.github.io/cli.html.