Thomas Moreau

Ok I see what you mean now. I would proceed that way: - add an optional argument in `Benchmark` to not check that `objective.py` exists so you can still create...

there is two way to do this: - put the `**solver.parameters` in the `metadata` in the runner, that way is epxlicit but will make a lot of columns - use...

> @tomMoral why is the size a concern? For HTML pages, if we embed the csv in the page, yes. But this is second order problem as we can restrict...

We could generate images or HTML and include these rendered equations in the `README.rst`.

No I don't but I did not look a lot around. that is why it is simply a proposal 😅 I am also open to other suggestions. Le sam. 4...

That would be great. We will need to do a similar benchmark for `julia` and I guess if we also have an overhead, we will be able to correct it...

I don't understand the issue here. Indeed, as we are using `setuptools_scm` for versionning, it computes a version number from the git history when installing the package with `pip install...

Maybe a solution would be to output something different when benchopt is installed in editable mode? Either the commit id or `editable`.

This versionning system is necessary to be able to auto release to test pypi. But it is true that it is not necessary. Honestly, I don't really care about the...

This is linked to this issue in `setuptools_scm`: https://github.com/pypa/setuptools_scm/issues/518 Maybe an idea to have something less misleading, we could use `benchopt/utils/misc.py::get_benchopt_requirement` to check if `benchopt` is install in editable version...