mathurinm

With @qb3 we had to write one file to do this. It may be a nice enough solution to commit it? where should it go?

I don't know if it's worth adding one parameter. Do you envision usecases apart from NNLS ? Otherwise this can be handled directly in `simulated.py` in the NNLS benchmark (use...

I cannot reproduce. `test_cli` pass now, with a few warnings: ``` /home/mathurin/miniconda3/lib/python3.8/site-packages/click/core.py:2905: DeprecationWarning: 'autocompletion' is renamed to 'shell_complete'. The old name is deprecated and will be removed in Click 8.1....

I also change the name of the objective (from "Lasso Regression" to the shorter "Lasso") so I'm not super in favor of using it as an id.

Yes I agree. - precision 1e-6 here means: solver did not go below "optimal value found across solvers + 1e-6". 1. this is not super clear from the message, but...

Apparently there exists a Python wrapper for the Fortran code of GLMNET: https://github.com/civisanalytics/python-glmnet It's conda installable. I will give it a try today.

@qb3 encountered the problem a while ago, removing the `outputs` folder fixed the issue. It just worked for @PABannier too. Apparently the problem stemmed from upgrading benchopt

My (limited and personal) experience is that when I pull the repo, the `__version__` is updated (since usually it is hardcoded in the `__init__.py`), without me having to do anything...

I can't see an obvious clean solution. I also realised that two different branches with the same number of commits since they diverged from `master` will have the same version...

It's clearly not critical, but I thought it was somehow a convention of python packages to expose this attribute directly. numpy, scipy, sklearn, matplotlib, pandas all do it I believe.