This is the code for the following paper:
"Collaborative Bayesian Optimization with Fair Regret". Rachael Hwee Ling Sim, Yehong Zhang, Bryan Kian Hsiang Low and Patrick Jaillet. In Proceedings of the 38th International Conference on Machine Learning (ICML), 2021.
python >= 3.7
pytorch >= 1.6.0
gpytorch
botorch
matplotlib
tenorflow 2
pickle
joblib
Hartmann-6d function
python run_hartmann.py --c1 0.08 --c2 5 --noise 0.01
Hyperparameter tuning of LR with mobile sensor dataset
python run_hyp_tuning.py --c1 0.01 --c2 10 --exp lr
Hyperparameter tuning of CNN with FEMNIST
python run_hyp_tuning.py --c1 0.001 --c2 1 --exp cnn
Mobility demand hotspot discovery on traffic dataset.
python run_traffic.py --c1 0.1 --c2 1 --nAgent 8
The implementation of our Fair BO algorithm is mainly inside the fairbo folder.
acquisitions.pycontains the acquisition function.functions.pycontains the objective functions for querying.models.pycontains the underlying GP model used in BO.testmodels.pycontains the models of LR and CNN for hyperparameters tuning.utils.pycontains functions for the optimization step and analysing the results e.g. cumulative regret, variance.
Other experiment settings e.g. the choice of weights, c1, c2, number of initial observations and time horizon are set within run_x.py.
The results are saved in a ./results/ folder and can be plotted by running result_plots.ipynb.