Skip to contentFeatures
■ Active learning features
- Bayesian optimization (BO) with Gaussian process regression (GPR)
- choice of N-dimensional GP kernels (periodic/non-periodic)
- Inverse gamma priors on hyperparameters
- (heteroscedastic noise)
- variety of acquisition functions
- batch acquisition strategies
- convergence-based stopping criteria
- support for symmetry considerations
- acquisition cost to control sampling
- gradient-assisted GPR and BO
- multi-task BO models for multi-fidelity studies
■ Postprocessing features
- output and restart files
- acquisition data analysis
- global minima tracking with variance
- hyperparameter convergence tracking
- GPR model visualization
- multi-objective and Pareto front analysis
- local minima search
- minimum energy path (MEP) detection
■ Practical features
- interface choices: Python API and command line (CLI)
- BO restart functionality for queue-based HPC computing
- versatile objective function: integrate with simulations or experiments
- out-of-the-box postprocessing graphics