search

theogf / ThermodynamicIntegration.jl

Thermodynamic Integration for Turing models and more
MIT License
13 stars 1 forks source link

Using already obtained posterior samples #19

Closed jonasmac16 closed 1 year ago

jonasmac16 commented 2 years ago

Hi, This package looks super interesting. I was wondering if it possible to use already obtained posterior samples from a Turing model to calculate the evidence rather than rerunning the model?

Best Jonas

theogf commented 2 years ago

Hi @jonasmac16, I don't think it would make a lot of sense to use obtained posterior samples as this would just be a minor part of the algorithm... In TI, you sample from the power posterior for different temperatures (power on the likelihood) and samples for the posterior would only correspond to samples from the final temperature at T=1.

theogf commented 1 year ago

Closing this as stale.