I want to remove the following warning.
There were 43 divergent transitions after warmup. Increasing adapt_delta above 0.9999 may help. See
I want to includ my Stan code, but my statistical model is very complicated and thus I cannot exhibit here. The above warning is always occur and its number of divergent transitions is very near to the number of iterations iter
in the rstan::stan()
. How to remove this warning ?
I think that in the model block, the descriptions for model is not unique since , e.g., by the additive property of Poisson distributions:
X \sim Poisson(\lambda) , Y \sim Poisson(\mu) \text{ implies } X+Y \sim Poisson(\lambda + \mu)
yields non-unique representations for my statistical model. I am not sure, but I think using such changing of representations are only methods to avoid the warnings. Or are there any other methods to eliminate this warning ? I am not sure.
My attempt is e.g., the following idea:
I have tried the following:
Suppose that if random variables X_1,X_2,... is distributed by Poisson(\sum_{i \geq 1} \lambda_i), Poisson(\sum_{i \geq 2} \lambda_i) where \lambda_1,\lambda_2,... are parameters of a model…
There are two representations in model block in stan file.
(1)
one way is the sequence such as:
X_1 \sim Poisson(\sum_{i \geq 1} \lambda_i)
X_2 \sim Poisson(\sum_{i \geq 2} \lambda_i)
(2) …
And the other way is a sequence
X_1-X_2\sim Poisson( \lambda_1)
X_2-X_3\sim Poisson( \lambda_2)
…
This change of expression cannot remove the above warning.
I am not sure these treatment is appropriate or its efficacy.
The follwoing web page says that
" If the divergent transitions cannot be eliminated by increasing the adapt_delta
parameter, we have to find a different way to write the model that is logically equivalent but simplifies the geometry of the posterior distribution."
I am not sure, e.g., which writing way[(1) and (2) ] is more simpler. Or both is wrong ?