Divergences in NUTS sampler

I am getting some divergences in my sampling chain. They are very few! For example I get 7 divergence when i run 4 chains with 10000 draw and 5000 tunes each.

They seem pretty random, and sometimes I get none. Should I be worried? The sampler seem converge pretty good, and my result agrees with experiments pretty good.

1 Like

Doesn’t seem like a big deal, there’s probably you can improve in the model to make them go away but a handful of divergences are usually not a strong indicator of a problem. You can also try reducing the stepsize.

1 Like

The little Aki’s on your shoulder says that any divergences could be nothing…or could be a sign of trouble. A few is probably fine, but a little investigation never hurts.