Sampler stops accepting new parameter values in rare cases. #15
Description
A couple times, the RAM sampler stopped accepting new values into the chain. The acceptance rate quickly dropped down from ~23% and the entire calibration then finished really quickly (since it wasn't accepting anything). It returned a final proposal covariance matrix of all NaNs (see attached image below). This has only happened a couple times and I can't replicate it or figure out what's causing it to do this. I also attached a trace plot, roughly around 2.5 million samples it just stops accepting anything.
