Closed jgattiker closed 3 years ago
Before evaluating acceptance of a new mcmc candidate for each parameter, both the SepiaModel and SepiaSharedThetaModels objects check that aCorr is not False. This check was not included in the mcmc code for SepiaHierarchicalThetaModels. Adding this check has fixed the issue. The other two model types check aCorr when they check that the parameter is in bounds, so that is where I added the check.
fixed in commit: 04cface
Hierarchical calibration always shows a warning on the first acceptance test, "divide by zero encountered in log". Need to track this down, it's probably just initializing some parameter differently.