Sampler occasionally gets static in pymc3.1, but not in pymc3.0

Dear Junpeng,

in R’s “MCMCgllm” toolbox there is one argument “degree of belief parameter (nu)” in some prior configurations.
Taken from one of the tutorials in this series, I had remembered that (at least in analytical Bayesian calculations) the posterior is kind of a “mixture” (plain average or with mixing ratios) of prior and data. A higher amount of data pulls the posterior more towards it, but this could also be achieved by weighting the prior down (low “degree of belief” in my prior).

For example, I could just duplicate my data set (count each observation twice) and should get narrower posteriors. Right?

Falk