bayesian - How to test the convergence in bugs model? -


i want explain convergence in bugs model command plot(). example of output in follow figureenter image description here

i don't sure can read output well, :)

unfortunately, not if can confirm convergence figure showing (edit: there @ least information, see below). left hand side of figure caterpillar plot, shows 95% intervals of distribution each parameter.

assessing convergence more nuanced process, there multiple ways decide if model has converged. want determine model has appropriately explored parameter space each parameter (through trace plots, traceplot function in coda library), between , within chain variance (the gelman-rubin diagnostic, gelman.diag in coda library), , auto-correlation in chains (autocorr.plot in coda). there variety of other measures others have suggested assess if model has converged, , looking through rest of coda package illustrate this.

i highly suggest go through winbugs tutorial in user manual (link pdf), has section addresses checking model convergence. want ensure traceplots well-mixed (look tutorial see means), gelman rubin diagnostic < 1.10 each parameter (general rule), , chains not correlated (this reduce effective sample size in chains).

good luck, , read bit on subject, benefit if interested in bayesian inference!

edit

as @jacobsocolar pointed out, , missed, plots available in question @ least have information indicates model did converge. did not see r-hat plot on right side of left plot. these values should less 1.1 each parameter if model did indeed converge. eyeballing above plot hint model converged, far easier see if there vertical line @ 1.1 mark on plot, there not.


Comments

Popular posts from this blog

Hatching array of circles in AutoCAD using c# -

ios - UITEXTFIELD InputView Uipicker not working in swift -

Python Pig Latin Translator -