Filippo Vicentini

Normally if you run `pytest` with `-n0` xdist is completely disabled. Is this not the case? Make sure it's among the first arguments, so something like `pytest -n0 whatever`.

For future reference: it might be possible to parse cmdline arguments to pytest, detect that we want to run benchmarks, and add -n0, disabling `xdist`, manually. https://stackoverflow.com/questions/66407583/how-to-disable-pytest-xdist-only-when-pytest-is-called-with-filters Not for now,...

Yes that is already great! SR needs indeed benchmarking. If you open a PR we can already start discussing about any detail that might arise. Don't worry about the license...

For R_hat the expert is @femtobit , but in general you want it to be the closest to 1 as possible, while a deviation of \pm0.01 from it means that...

Hey @nikosavola , you still up for this? do you need some guidance?

Hey @nikosavola how are you with that? UnitaryHack is ending soon. We can tag a PR as accepted even before it's merged if it's mostly there, but we have to...

It is surely useful but I'd like to discuss how to ask the logger this before you make a PR, so that we can agree on it.

And it could take something like `BestStateLog(metric="Energy", k=3)` as well to specify which metric and how many 'best' iterations to keep ?

My only take is that to make it truly useful one would need to log both the parameters in an mpack file (or archive) as well as the corresponding iteration...

> So either we should either set the variance of the weights to be 1/(alpha x n_sites^2) instead of lecun_normal() or we downscale by sqrt(alpha x n_sites) in the sum...