linmix fails with a large sample

[neobar]

I have been trying to fit a linear model to a sample with a size of ~1000, but the lm.run_mcmc() takes a very long time and seems never to end. If I reduce the sample size slightly, say 950, the sampling will finish very quickly.

Is there a way to make linemix work with a large sample (>=1000)?

[jmeyers314]

Hi @neobar,

I just tried modifying the test.py that comes in the tests directory to use n=2000, and things appeared to run. Do you have a minimal failing example perhaps that I could look at? Otherwise, this may be pretty hard to diagnose...

Also, you can use the maxiter keyword to run_mcmc() to force the computation to terminate after a given number of steps. You should definitely plot the resulting chains in that case to see if they look like they've converged or not. (Actually, I'd recommend plotting the chains in general to look for any weirdness).

[neobar]

I have tried maxiter=100, but it does not work either.
It feels like a deadlock to me.

I have put a link to the data that I have tried to fit.