eggplantbren / DNest4

Diffusive Nested Sampling
MIT License
62 stars 22 forks source link

Post-processing with a different volume ratio #19

Open MichelleLochner opened 7 years ago

MichelleLochner commented 7 years ago

In Appendix A of the code paper, you mention you can change the level compression ratio with the "-c" option. My question is does, something corresponding need to change when post processing? Is there an explanation somewhere that I missed of the parameters in the post-processing function, as they have a big difference on the number of samples produced...

Thank you!

eggplantbren commented 7 years ago

You shouldn't need to do anything different with postprocess if you've used -c. Are you able to post the output plots? That might show what's happening.

MichelleLochner commented 7 years ago

Ok cool. The output plots look absolutely fine (posterior weights peaking nicely, levels being explored well etc.). I just wanted to check the output posterior samples are representative of reality. The only problem I face sometimes (in high dimensions) is that too few samples are produced, and running the chain longer doesn't actually seem to help (the effective sample size remains the same). There's a few options in post_process like the helpful "moreSamples" which seems to do the trick... But I just wanted to check if there was an implicit volume ratio assumption in post_process that I needed to change.

eggplantbren commented 7 years ago

The only problem I face sometimes (in high dimensions) is that too few samples are produced, and running the chain longer doesn't actually seem to help (the effective sample size remains the same).

This shouldn't be the case, but everything else you've said sounds good. I am puzzled.