So far the tests with acceptance_fraction and acor haven't been 100% successful at making an automated metric for convergence. Instead, test how much chain properties (e.g. sample mean, sample variance, 68% width) change with nSteps or nWalkers. Either do several runs with increasing length, or just do very long ones and compare the full chain to shortened ones. Note that convergence appears to depend on galaxy properties, with edge-on galaxies taking longer than face on ones.
So far the tests with acceptance_fraction and acor haven't been 100% successful at making an automated metric for convergence. Instead, test how much chain properties (e.g. sample mean, sample variance, 68% width) change with nSteps or nWalkers. Either do several runs with increasing length, or just do very long ones and compare the full chain to shortened ones. Note that convergence appears to depend on galaxy properties, with edge-on galaxies taking longer than face on ones.