search

pints-team / pints

Probabilistic Inference on Noisy Time Series
http://pints.readthedocs.io
Other
226 stars 33 forks source link

Implement "stupid-mcmc" (also known as "prefetching") #26

Closed MichaelClerx closed 6 years ago

MichaelClerx commented 7 years ago

See:

  1. Brockwell (2006) Parallel Markov chain Monte Carlo simulation by pre-fetching
  2. Strid (2009) Efficient parallelisation of Metropolis-Hastings algorithms using a pre-fetching approach
MichaelClerx commented 6 years ago

I'm inclined to drop this issue, as @ben18785 has repeatedly pointed out that you shouldn't just do one chain (in parallel), you should do multiple and then compare them to see if they're acceptable!

We can already do multiple in parallel, so unless you have lots and lots of cores (not a trend in home computing at the moment) doing individual methods in parallel seems less desirable? @mirams ?

mirams commented 6 years ago

Yeah, certainly low enough priority to close for now. Have to keep scope sensible!