Excellent paper on Comparing JAGS, NIMBLE, and Stan

Publicity
21

I came across this thread from 16 months ago on these terribly designed MCMC comparisons posted on arXiv. I am a NIMBLE developer and just want to point anyone interested to two blog posts we made about these comparisons: one on their accelerated failure time examples and one on their linear model examples. The authors not only made poor use of NIMBLE (I see there are related points about Stan), but the comparisons they made were apples vs oranges in the extreme because they gave different models to different tools. By running some cases, we saw they underestimated NIMBLE’s performance by a lot, even by 45-fold in the worst case. There were still plenty of caveats and unknowns, but we stopped there. I hope this is helpful for future readers who find this thread.

7 Likes
22

FYI, as of 1 May 2024, this link
https://nature.berkeley.edu/~pdevalpine/MCMC_comparisons/nimble_MCMC_comparisons.html

Edit If forgot to add the link doesn’t work

23

Link doesn’t work.

1 Like
24

Thank you. I forgot to add “this link doesn’t work”.

I’ve edited my original post.