Assessing the convergence of runs

[lpittet92]

Hello,

First of all, thank you for giving us the opportunity to run BGC so easily.

I am having some trouble understanding my results. I attached the log likelihood and hybrid index graphs that I obtained from two runs (100000 burn-in / 200000 MCMC iterations). I did it only as a test and want to add more runs and probably also more MCMC iterations.

Could you help me understand why my two runs give quite different log likelihood and hybrid index. It looks to me like the model found convergence inside each run but is not coherent between runs.

All the best,

Loïc
ABH_hi_convergence.pdf
ABH_LnL_convergence.pdf

[tkchafin]

I would suggest a longer run with a larger burn-in, and increase the thinning parameter (2nd MCMC in particular looks to have a lot of autocorrelation)

[lpittet92]

That was fast !!

Thank you very much, I will give it a try and let you know how it goes :)

[lpittet92]

Hello again,

I have been trying a lot of things since the last message... Now my runs converge. When I included all the populations of the hybrid zone, BGC identified only one population as hybrid (hybrid index of 0.5), and all the other populations as P1 (average 0.98 HI). So I ran it again with only the hybrid pop and I get the result attached.
FWH_FINAL_all_hiXphi_alphaAndBeta-2.pdf

I am not sure if everything ran normally. Do you have any idea what is happening here ?

Thanks