pyPESTO
pyPESTO copied to clipboard
Adaptive Stopping and Continuation of Sampling
Feature description
- Include option to stop sampling if convergence is already achieved.
- Enable the option to continue sampling where the procedure stopped (e.g. give the sampler an sampling result and continue the sampling at the last point in the result, for the adaptive schemes we would also need access to the covariance matrix etc.).
Motivation/Application Often it is not known which sample size is requiered until convergence of an MCMC run. We could check every x-samples if we converged already, and continue or stop sampling depending on that instead of creating a fixed number of samples. Also when we already created samples, but not enough, we might want to continue sampling at the last sampled point.
Actually, the dynesty sampler already allows to do both