search

university-of-newcastle-research / pmwg

Repository for code for the Samplers Team at UoN
3 stars 3 forks source link

Test for unique samples should check over all adapt stage particles, not just current iterations #45

Closed gjcooper closed 3 years ago

gjcooper commented 3 years ago

Is your feature request related to a problem? Please describe. When starting a second adapt stage, the first n_unique iterations are currently run before testing that you have enough unique samples. Instead it should notice the previously collected adapt stage particles and factor those into deciding to run the test.

Describe the solution you'd like Check for previous adapt stage samples, and current adapt stage samples (but not "future", ie unfilled adapt samples), combine them and check the criteria on these.

Describe alternatives you've considered Leave as is and always run at least n_unique iterations in the new stage.

gjcooper commented 3 years ago

If we are running a second (or third) adaptation stage we have already committed to a presumably long-running operation, dropping out early because of a previous run seems short-sighted, and adds complexity for a relatively small gain.