Be still my heart... could this be... the missing gradient-free optim library for Python? Particle swarm, evolutionary, AND Bayesian all in a single package? A no-nonsense, simple API with strong enough defaults that you can basically just call .search() with a function (I’m looking at you, nevergrad)??
I’ll have to test it of course but this looks truly life changing. A few clarifications, if the author happens to be watching this thread:
- looks like the search space is “enumerated”, I.e. I define the grid spacing for each independent variable in search_space and it puts together an outer product? In other words, this will never be truly continuous?
- by the same token, constraints are applied by virtue of specifying the endpoints of each grid in the search_space? Is it possible for me to apply more complex constraints by generating my own enumerated list of search_space points and feeding that directly?
Seriously well done. Like I said above, this looks to me like an immediate staple library.
I’ll have to test it of course but this looks truly life changing. A few clarifications, if the author happens to be watching this thread:
- looks like the search space is “enumerated”, I.e. I define the grid spacing for each independent variable in search_space and it puts together an outer product? In other words, this will never be truly continuous?
- by the same token, constraints are applied by virtue of specifying the endpoints of each grid in the search_space? Is it possible for me to apply more complex constraints by generating my own enumerated list of search_space points and feeding that directly?
Seriously well done. Like I said above, this looks to me like an immediate staple library.