search

real-itu / benchmarking_evolution_and_bo

A collection of simple examples in which we benchmark evolutionary algorithms plus Bayesian Optimization
0 stars 0 forks source link

parametrise dimensionality #8

Closed enajx closed 1 year ago

miguelgondu commented 1 year ago

What do you mean by this exactly, @enajx?

enajx commented 1 year ago

Many of the standard test functions can have arbitrary dimension but right now their dimension is hard coded in their definition. I suggest we have the dimension as a parameter. I'm not sure if all the standard artificial landscape functions have a natural n-dimensional extension though.

For reference https://al-roomi.org/benchmarks/unconstrained/n-dimensions and https://www.sfu.ca/~ssurjano/optimization.html

Does this make sense?

miguelgondu commented 1 year ago

I see, but the new signature for the objective functions is Callable[[torch.Tensor], torch.Tensor], right? In other words, we can already start testing the optimization on RL tasks by passing the weights.

Some have a natural extension (like the Rastrigin), but I don't really know about the rest. Will check