mathurinm

BUG neurips results are not available on benchopt.github.io/results

1

Clicking "neurips 2022" takes me to https://benchopt.github.io/results/2206.13424.html and then I get a "Page not found"

FEAT generate config.yml and attaching it in the HTML

1

This would allow somebody to reproduce the figure by just downloading the config file and running benchopt.

- Where can we explain that parametrized solvers, datasets and objectives are instanciated with all possible combinations of their parameters? This is a crucial point for newcomers, because if they...

BUG parameter that are strings containing '/' break our parsing

3

Reproduce by adding a parameter to the Lasso benchmark ```python parameters = { 'fit_intercept': [True, False], 'reg': [.5, .1, .05], 'data': ["path/to/stuff"], } ``` ```python Traceback (most recent call last):...

``` rpy2) ➜ lasso git:(main) ✗ benchopt install -s celer -s cyanure -s glmnet -s lightning -s skglm Installing 'lasso' requirements Install in the current env 'rpy2'? [y/N]: y #...

as raised by @paquiteau

same as #556

E.g. here : https://benchopt.github.io/results/benchmark_logreg_l1_benchmark_logreg_l1_benchopt_run_preprint22.html

Yet is is required to implement one as it's an abstractmethod of BaseSolver. This could be misleading for a newcomer

We first print that the solver is available (`- 'SLOPE' already available`), then raise an error: ``` (rpy2) ➜ slope git:(install_rslope) ✗ benchopt install -s rslope Installing 'slope' requirements Install...