Dynamic Bayesian Optimization
This code replaces the existing Bayesian optimization (BayesianOptimization.m), which works with bayesopt.m, to enable dynamic BO
In preparation for using DBO code)
- DBO code is built based on the Matlab default BO code (bayesopt.m) in 2019b version of MATLAB.
- This code should be working by over-writing existing BayesianOptimization.m code located in (C:\Program Files\MATLAB\R2019b\toolbox\stats\bayesoptim) by the DBO code mentioned above.
- For over-writing, Matlab or explorer should be opened as administrator.
- Note: Save default BO code before over-writing.
- Note: This code is not fully validated with different versions of Matlab, especially over 2021.
- Note: this code only works with 1) not using parallel option, 2) Acquisition function of expected-improvement-plus
With in DBO code)
- Either BO or DBO can be switched by commenting line 4031 shown below
Comment: DBO
Uncomment: BO
- Standarize option (or other options in fitrgp) can be controlled by fixing line 3802 or 3803 shown below.
- For disabling periodic input, fix line 4036-4047
If it only include as shown below, code is not using periodic input.
- Initial guesses for hyperparameters
If you want to give better initial guesses for hyperparameters (for using APO), you can provide in 3769-3789.
- Time component for DBO code
Now, DBO code automatically gives iteration number as time component. For example, for observations 1-5, number 1-5 is automatically designated for each observation.
To revise this, change TimeTrace and TimeT component in DBO code.
引用格式
GilHwan (2026). Dynamic Bayesian Optimization (https://ww2.mathworks.cn/matlabcentral/fileexchange/183999-dynamic-bayesian-optimization), MATLAB Central File Exchange. 检索时间: .
