Multi-Objective Bayesian Optimisation

This is the repo that we will use for the multi-objective bayesian optimisation hackathon project! It is advised to used python 3.10.12 with this repo to ensure that the packages are downloaded without hassle.

---

Using the repo

1. Clone the repo by running:

   git clone https://github.com/A-Ijishakin/BayezOpt.git
   

2. Then create an environment:

   • conda:

   conda create --n <env> python=3.10.12
   

   • pyenv:

   python3 -m venv <env>
   

3. Activate it:

   • conda

   conda activate <env>
   

   • pyenv

   source <path to venv>/bin/activate 
   

4. Then run:

   pip install -r requirements.txt
   

5. You will now be able to use the multi_objective_bo.py script, which takes the following system arguments:

   • test_function: The type of test function we are using, this will be selected from those specified in optim_configs.py/
   • input_constraint: This specifies a type of input constraint, which should be defined in optim_configs.py.
   • output_constraint: The same as above but for output constraints/
   • noise_se: The standard error of the gaussian noise that will be added (if any).
   • w_dragonfly: Specifies whether we are running with or without dragonfly.

Ploting

There is some basic code for plotting the output of BO runs in plot.py. The first plot illustrates the log difference in hypervolume between the complete hypervolume and the hypervolume encompassed by candidates suggested by various acquisition functions. This comparison is presented as a function of time. The second plot is the mean pareto front of these candidates.