GPBayesTools-HIC

Gaussian Process Bayesian Toolkit with Monte Carlo Sampler Integration for Heavy Ion Collisions

This toolkit implements a wrapper for Gaussian Process (GP) emulators and Monte Carlo (MC) samplers used in high-energy heavy-ion simulations.

The following wrappers for GP emulators are currently included:

• Scikit Learn GP emulator wrapper
• PCGP and PCSK wrapper for the GPs implemented in the surmise package of the BAND Collaboration

The following wrappers for MC sampling are included:

• MCMC wrapper for the emcee package
• PTLMC from the surmise package (Parallel Tempering Langevin Monte Carlo)
• pocoMC Preconditioned Monte Carlo method for accelerated Bayesian inference

We recommend to use the pocoMC sampler.

Latin Hypercube Sampling

There is also a script to generate Latin Hypercube Design parameter files. An example how to use it is given in the examples directory in the generate_LHD_Bayes.py script. This requires a file specifying the parameter ranges, see for example examples/modelDesign_example.txt.

Posterior Cluster Sampling

The posterior_cluster_sampling.py script in the examples directory can be used to sample parameter clusters from the posterior chain file after a Bayesian inference run and propagate model uncertainties to the observables. The final cluster_centers.txt file contains the sampled parameter clusters as separate columns.

Requirements

Check the requirements.txt file for the dependencies of this code.

❗ The jupyter notebooks are just meant as examples for how to use the emulators and samplers and analyze the output. Paths and data files need the proper input formats.