Closed AaronDJohnson closed 1 year ago
hypermodel.setup_sampler duplicates shared parameters between models when creating groups.
For example, if two models being concatenated share their first parameter, then the index 0 will appear twice in the groups.
If the models are very different, this could lead to longer burn-in times and poor convergence as these groups will not appear as often to be sampled.
Thanks to Jonathan Nay for finding this bug.
Fixed with #202
hypermodel.setup_sampler duplicates shared parameters between models when creating groups.
For example, if two models being concatenated share their first parameter, then the index 0 will appear twice in the groups.
If the models are very different, this could lead to longer burn-in times and poor convergence as these groups will not appear as often to be sampled.
Thanks to Jonathan Nay for finding this bug.