search

isl-org / LMRS

Source code for ICLR 2020 paper: "Learning to Guide Random Search"
MIT License
39 stars 5 forks source link

Testing on LSGO / BBOB / others ? #2

Open teytaud opened 3 years ago

teytaud commented 3 years ago

Your work is super interesting! Would it be possible to test it on standard benchmarks in black-box optimization ?

For example BBOB https://coco.gforge.inria.fr/ or LSGO (large-scale global optimization) ?

We have all of them including in our benchmark suite in Nevergrad (https://github.com/facebookresearch/nevergrad).

If your black-box optimization code can be extracted and applied to a generic black-box function that can be applicable to a wide range of problems way beyond linear control. I'm just not sure if there is a strong reason for which applying your code to classical benchmarks ?

If your code is packaged in PyPi and there is an example of how to optimize lambda x: np.norm(x) with it, I can do the rest by myself.

ozansener commented 3 years ago

Thanks for the note and your comments.

We are planning to test it on standard benchmarks using Nevergrad. We did not yet have time to try.

I will look into this next month and will either try it directly or prepare a package so you can test it.

teytaud commented 3 years ago

Thanks for the note and your comments.

We are planning to test it on standard benchmarks using Nevergrad. We did not yet have time to try.

I will look into this next month and will either try it directly or prepare a package so you can test it.

I would be super happy to play with your code. Looks like a candidate for really beating existing algorithms on generic black-box benchmarks.

Good luck!