Include sampler to PyMC3?

naesseth / nestedsmc

Implementation of various algorithms in the Nested Sequential Monte Carlo family of methods.

MIT License

14 stars 1 forks source link

Include sampler to PyMC3? #1

Open twiecki opened 8 years ago

twiecki commented 8 years ago

Just writing to test the waters of whether you think this would make a good addition to PyMC3: https://github.com/pymc-devs/pymc3. Creating a new sampler is pretty straight-forward and would allow it to run on a large variety of models. Specifically, nested samplers were requested here: https://github.com/pymc-devs/pymc3/issues/719

naesseth commented 8 years ago

I made a short comment high-lighting some differences between nested SMC (inherently a particle filter) and the nested sampling algorithm as first presented by Skilling. Maybe sequential Monte Carlo algorithms are not what you are after? Could of course be potentially interesting to implement particle MCMC algorithms: http://www.stats.ox.ac.uk/~doucet/andrieu_doucet_holenstein_PMCMC.pdf

These are a way of using SMC samplers/particle filters to construct invariant MC kernels.