ICB-DCM
Multi-level sampling
Allow to use surrogate models in non-final iterations, and only the full, expensive, model in the last iteration. This will make things more efficient. What to do? One way that would suffice would be to make the model simulation time-aware (pass t to .sample). There might also be some more handy alternatives.
This reference may be of interest: "Multifidelity Approximate Bayesian Computation with Sequential Monte Carlo Parameter Sampling" https://arxiv.org/abs/2001.06256
Hi @ljschumacher . Thanks, this is exactly one of the ways we had in mind :smile:. Are you interested in using such a method?
Yes, I would be. We have a use case where a faster deterministic model may help speed up inference for the full stochastic one
On 11 Aug 2020, at 12:43, Yannik Schälte [email protected] wrote:
Hi @ljschumacher https://github.com/ljschumacher . Thanks, this is exactly one of the ways we had in mind 😄. Are you interested in using such a method?
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/ICB-DCM/pyABC/issues/172#issuecomment-671896120, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAZ75UFEFFFKDCKRE3D7L63SAEVGBANCNFSM4I2YECUQ.