USER GUIDE¶
Installation¶
BayesValidRox provides functionalities for describing uncertain parameters, building surrogate models based on model outputs and evaluating them with Bayesian validation methods.
This package runs under Python 3.9 for versions <1.0.0 and 3.9+ from version 1.0.0 on. It can be installed with pip, best practice is to do so inside a virtual environment.
python3 -m venv bayes_env
cd bayes_env
source bin/activate
Here replace bayes_env
with your preferred name.
Then install the latest release of BayesValidRox inside the venv.
pip install bayesvalidrox
The current master can be installed by cloning the repository.
git clone https://git.iws.uni-stuttgart.de/inversemodeling/bayesvalidrox.git
cd bayesvalidrox
pip install .
Overview¶
This package is split into multiple aspects corresponding to its folder structure.
The folder surrogate_models
contains all the functions and classes that are necessary in order to create and train the surrogate model.
This includes
defining the input marginals
setting properties of the sampling in an experimental design
choosing the surrogate model and its properties
training the surrogate model on model evaluations in a straightforward manner or iteratively with active learning
The computational model is linked via a pylink
interface.
Multiple post-processing options are available, including the calculation of Sobol’ indices, checking the accuracy of the surrogate model and visualizations of the moments of the surrogate.
Bayesian inference can be performed with rejection sampling or MCMC, while taking into account the estimated uncertainty of the data that the (surrogate) model is compared to. If multiple (surrogate) models are given, they can be compared against each other with pairwise Bayes Factors, model weights or a justifiability analysis.
The next pages lead through the topics given in BayesValidRox and describe the available classes and give brief examples for their use.