Joost VandeVondele

@MJZ1977 I think it is great somebody is looking at the implementation of SPSA. I'm still puzzled why our tuning attempts have such a low success rate (@linrock recent experience)....

@ppigazzini concerning Nelder-Mead, I did work on interfacing cutechess games to the nevergrad suite of optimizers : https://github.com/vondele/nevergrad4sf and picked TBPSA which seems to be the recommended optimizer for noisy...

while the TBPSA might also work for global optimization (that's always hard), I don't think we're typically stuck in local minima. At least, I have never seen evidence of that....

that might not be such a good test... this could be so far off that e.g. the local gradient is zero, so progress won't be made. Maybe it could be...

I've documented this limitation here https://github.com/glinscott/fishtest/wiki/Running-the-worker:-overview#install-the-worker

yes, and the SF code needs to use the right types for filenames in all places etc... it is definitely not completely trivial to get right.

I think it is not worth trying to do it from the worker. Things in SF it self might break (like it searches for the net in a couple of...

yes, I agree, so well tested patches for SF would be welcome there :-) For whatever reason the Lc0 people said it is really non-trivial, so I take their word...

OK, if that fixes it, I'm of course fine with it.

it will first use the embedded net, after that it will search the directory of the binary.... that would probably go wrong.