null model error message

[Andrea]

Hi - 

I am running into an error message that I haven't seen before. I have created my capthist file and everything checks out ok. I've looked at the summary information, plotted the mask, traps, tracks, etc. All looks good. However, when I run a null model:

m0 <- secr.fit(CHBM, model=list(D~1,g0~1,sigma~1), mask = maskBM, trace = FALSE)

I get the following warning message:

Warning message:

In dpois(data$nc, N * meanpdot, log = TRUE) : NaNs produced

When I look at the model output, the information is incorrect. The number of occasions is listed as 1 (there should be 140). Also, detector type is wrong - it is listed as "count" but should be "proximity."

Interestingly, when I add a covariate to either g0 or sigma (e.g. bk or T), things run correctly and there are no error messages. Also, in those cases the model output is correct (i.e. right number of occasions, correct detector type, etc.).

Any suggestions as to what might be going on with the null model?

Thanks

Andrea

[Andrea]

As a follow up, I tried running the model in an older version of secr (2.10.4) and I do not get any error messages. 

[Murray Efford]

Firstly, these are not error messages. They are warnings.

I expect you are using secr 4.0.2 and should read about the changes between 3.2 and 4.0. By default, compliant proximity data are compressed to a single count occasion before analysis with a binomial model. You can suppress the new behaviour by setting details = list(fastproximity = FALSE). I could have saved these headaches by making it default to the old behaviour, but I was dazzled by the speed of the new behaviour. Note also that bugs in 4.0 have been fixed in 4.1 (on CRAN now but Windows binary lagging; find here if you're in a hurry).

The dpois warning is almost certainly innocuous - I expect it comes from a single likelihood evaluation early in maximization and has no effect on results.

Murray

[Andrea]

Thanks Murray. I'm running 4.1. I had skimmed the changes but obviously didn't read the document close enough - your suggestion fixed the problem and I no longer receive a warning. Thanks

Andrea