Basic Bayesian Optimization Methods with Bo-Torch

This repository contains an archetype to build algorithms by using BO-Torch. In this moment this repository uses the Bo-Torch, GPyTorch and PyTorch as main ones to develop the forthcoming methods. Nevertheless, this library just includes a directory archetype and a main function to test the built algorithm with the BBOB problems called from IOH-Experimenter interface (see: https://iohprofiler.github.io/IOHexperimenter/)

Libraries and dependencies

The implementation is in Python 3.10.12 and all the libraries used are listed in requirements.txt.

Structure

• main.py -> An archetype file, which is an example on how to call an instance of one of the 24 BBOB problems by using IOH interface and use the Vanilla-BO Algorithm stored in the repository.
• myexperiment -> An example of the generated files by a logger to assess the performance of your algorithm on a problem (namely from BBOB).
• /Algorithms -> A folder which stores the Algorithms to be developed. This repository contains an AbstractAlgorithm class, which works as a basic set up of the Algorithm. As this works with IOH, then there are some bypasses whenever the optimization comes from the BBOB/RealSingleObjective instances from IOH in order to ease the definition of the properties of the Optimizer such as the dimensionality and the bounds. To call an instance of a new algorithm, you must include your new algorithm within the __init__.py file in the same level.
• /Algorithms/Bayesian_Optimization -> In this folder you may save all the new Bayesian Based Optimizers. There's an AbstractBayesianOptimizer class to define an archetypical one. When you make a new BO algorithm, consider building upon this class as a parent class as this might ease your implementation. Additionally, generate a new folder per new variant of the algorithm you might create.

Execution from source

Dependencies to run from source

Running this code from source requires Python 3.10.14, and the libraries given in requirements.txt (Warning: preferably use a virtual environment for this specific project, to avoid breaking the dependencies of your other projects). In Ubuntu, installing the dependencies can be done using the following command:

pip install -r requirements.txt

Then, in order to have a glance on how to use it with any problem, see the main.py file to call the optimizer.