A comparison framework for different batch acquisition functions. This is the code used to evaluate the performance of the new batch acquisition function, Optimistic Expected Improvement, against the state of the art in the paper Distributionally Ambiguous Optimization Techniques in Batch Bayesian Optimization by Nikitas Rontsis, Michael A. Osborne, Paul J. Goulart.
Please use the master branch for OEI as this one is outdated and intended only for the other algorithms.
This package was written in Python 3.6 and uses the packages listed in installation.txt.
The methods folder contains most of the source code. It implements the following acquisition functions (see the paper for references and more detailed descriptions)
- Optimistic expected improvement (
oei.py) (our novel algorithm) - Multipoint expected improvement (
qei.py) - Multipoint expected improvement, Constant Liar Heuristic (
qei_cl.py) - Batch Lower Confidence Bound (
blcb.py) - Local Penalization of Expected Improvement (
lp_ei.py)
Each of these is based on the parent class BO (bo.py) that implements a simple parametrizable Bayesian Optimization setup.
The out folder is where the output of each run is saved, while the results folder is where the final figures are produced. The test_functions folder defines synthetic & real world functions that are used as benchmarks for the algorithms.
Invoke as the following example:
python run.py --algorithm=OEI --function=hart6 --batch_size=10 --initial_size=10 --iterations=20 --noise=1e-6