Hi, I am new to Stan. I created the following Beta-Binomial model for daily samples, which assumes the success rate of the Binomial distribution is sampled from a Beta distribution each day, and the mean of the Beta distribution is estimated using a logistic regression. I am trying to use that to e…

(Before people with more experience will reply) I would suggest to observe the traceplots including the warmup draws. Try to see if for any parameter there is one chain that explores incredibly large or small values that leads to that chain be less efficient. Sometimes applying more informative pri…

Update: I tried setting init = 1 in $sample() (using cmdstanr) and it looks like it completely solves the problem on my end! The slow chain is easily 20x faster now and keeping up with the other chains. My only concern is that I do not fully understand what init actually does. (I read the docs but …

[image] wlandau: My only concern is that I do not fully understand what init actually does. (I read the docs but have only an intuitive superficial grasp of what HMC is doing.) I’ll try to explain a bit more what this line in the CmdStanR docs actually means: A real number x>0 initializes r…

Thanks, that’s super helpful! I was originally taught that in order to be able to assess mixing, the starting values from one chain to another need to be more dispersed than the posterior. In grad school when I was custom-coding high-dimensional models from scratch, I always struggled with this. Bu…

Glad that was helpful! I might be misunderstanding, but it’s not clear to me how you could ever have inits that are “more dispersed than the posterior”. If you initialize somewhere with zero (or vanishingly small) posterior probability then the chain really won’t be able to move. But if you just me…

One chain always slower (x10) than other chains

Modeling

wlandau November 10, 2020, 1:57pm 3

Would a lower init_r be likely to help? Ref: One MCMC chain not moving.

Topic		Replies	Views
Initialization error Modeling	18	1885	August 8, 2020
One MCMC chain not moving Modeling cognitive-science	7	2759	November 11, 2020
Warning : Chains in stan are not mixed Modeling rstan , fitting-issues , specification	2	587	August 16, 2020
Inconsistent chain speed - does this give a clue about the problem? Algorithms optimization	10	4755	July 20, 2018
Variation in elapsed time of parallel chains and best practices for computing expectations from multiple chains Algorithms mcmc	18	3248	October 29, 2018

One chain always slower (x10) than other chains

Related topics