This repository contains the source code for the paper "Bayesian Optimization over Hybrid Spaces" presented at Thirty-eighth ICML'21 conference.
-
By default, data is stored in
../EXPERIMENTS
. Directory can be changed inconfig.py
-
The command-line arguments are described below:
- n_eval : The number of evaluations
- objective : ['coco'] (example on how to create new objective can be seen in experiments/test_functions/mixed_integer.py)
- problem_id : applicable only for 'coco' and 'nn_ml_datasets' domain
- path : A path to the directory of the experiment to be continued. (Only required when you want to resume an experiment)
-
Example run
python main.py --objective coco --problem_id bbob-mixint_f001_i01_d10 --n_eval 180
-
There are 7 benchmarks used in the paper. For using synthetic benchmark, please see instructions in coco suite. For robot pushing benchmark, please see the original description in Ensemble-Bayesian-Optimization.
The discrete part of the code is built upon the source code provided by the COMBO authors. We thank them for their code and have added appropriate licenses.