pseudo-residuals for zero-inflated gamma 2

[adeline95]

Dear authors, thank you very much for this interesting package that enables to include random effects on observation parameters.
I am using hmmTMB version 1.0.2 on R 4.3.2 and working with activity data. I fitted a 2-states model with one response variable with distribution "zero-inflated gamma 2" and random effect on the mean of the response variable, but when trying to compute the pseudo-residuals, I got this error:
pr <- hmm1$pseudores() Computing CDFs... done Error in hmm1$pseudores() : object 'hmm' not found

To better understand the problem, I tried to run the vignette example (energy prices) with these parameters:

dists <- list(Price = "zigamma2") par0_2s <- list(Price = list(mean = c(3, 6), sd = c(1, 1), z=c(0.5,0.1)))

But I got the same error. Therefore I assume it could be something linking to the "zigamma2" distribution.

I hope you may be able to help me solving this problem.

Thanks for your support and best regards,

Adeline

[TheoMichelot]

Hi Adeline,

Thanks for your message, this was caused by a small bug in HMM$pseudores(). I fixed the bug in commit 64aabb4.

However, the reason this bug came up specifically in your case is that pseudo-residuals are not implemented for the zero-inflated gamma distribution. If you install the latest version of the package from Github, you will now get a warning message informing you that pseudo-residuals are not available for "zigamma2". This is because it isn't straightforward to deal with observations that are exactly zero when computing pseudo-residuals.

You could consider other approaches to model checking, e.g., using simulations in HMM$check().

Best wishes,
Théo

[adeline95]

Hi Theo,
Thank you very much for your quick answer and for the heads up regarding the pseudo-residual for zero-inflated gamma distribution.
I'll explore the option of using simulations in HMM$check().
Thanks again for your help!
Best,
Adeline