Closed mjo22 closed 6 months ago
I think probably not, to be honest. The problem with optimisation is how many kinds of it they are... ! So for example Bayesian optimisation wouldn't be in scope here either.
Thank you for the offer nonetheless! (I am starting to think we should find a place to collect "extra" JAX scientific libraries, if for example you were to implement such a thing yourself.)
This completely makes sense! I didn’t think it would fit to be honest, but wanted to ask to be sure. Regardless, I will probably use optimistix
-like ideas when writing an API!
And would be very happy to learn more what you’re thinking along these lines, this would be very helpful.
Hello! In my applications it is very common to optimize a function with an exhaustive grid search method. This is because our loss functions are sharped peaked and poorly behaved in a manageable subset of parameter space, so it is often best to do exhaustive search (perhaps in a clever way). I am planning on implementing this in JAX, and I am wondering if this is within the scope of
optimistix
. It seems to me that there are tools in the library that would be useful for this task.A rough outline of the implementation I am imagining is the following:
I'm not sure if this is within the scope of
optimistix
, and I would totally understand if it is not. If it were to be added to the library, I suppose it could be used as a method of ultra-last resort.