Example for multi objective BO

[A-ep93]

Hello all,
I am new to PyTorch and Botorch and I want to use multi-objective BO. I was wondering whether there is a simple example for multiobjective BO so that I can start from. I saw the tutorials on the website, but they use qEHVI and qParEGO. I think these are for parallelization and make those tutorials complicated. Even if calculations are serial, it is fine with me. Are there any other simpler examples of multi BO that I can use?

[saitcakmak]

Hi @A-ep93. Unfortunately, I don't know of a simpler BoTorch example for MOBO that I can point to (Ax may simplify things compared to BoTorch). The only part that makes those notebooks parallel is where they set BATCH_SIZE = 4. If you replace this with BATCH_SIZE = 1, then the acquisition functions (particularly the output of optimize_acqf) would recommend one point at a time.

Looking a bit closer into the notebook, optimize_qparego_and_get_observation constructs an acq_func_list, which you should not need if you're using BATCH_SIZE = 1, in which case this function would look very similar to optimize_qehvi_and_get_observation. Otherwise, the main BO loop, the part after "Perform Bayesian Optimization loop with ..." looks a bit long, though that is only because it is running three algorithms at once. The variables corresponding to each algorithm is conveniently labeled with _random, _parego, _nehvi; so running only a single one of them should be as simple as deleting the lines that involve the others :).

[A-ep93]

Thank you very much for your reply @saitcakmak.