Percent deviance explained

[James-Thorson-NOAA]

@seananderson I was thinking about adding deviance_explained as function to tinyVAST. Similar to VAST, it would use the formula for relative deviance conditional on ML and EB estimators. I have the relative-deviance formula for most but not all likelihoods, and it matches output from mgcv for those cases (where mgcv typically uses plug-in estimators, e.g., conditional calculations). Is it worth discussing how to add a similar function to sdmTMB?

[seananderson]

Yes, I'd be very interested in getting something like this in the package.

[James-Thorson-NOAA]

In case it's helpful, I went ahead and prototyped it in tinyVAST@dev here: https://github.com/vast-lib/tinyVAST/blob/dev/R/utility.R$L333

I have several checks that confirm that it matches glm / mgcv in the vignettes and CI checks.

[James-Thorson-NOAA]

Do you wanna look at the tinyVAST demo and respond with some thoughts on interface, or would you rather I dive in and propose something via branch and PR?

I'd of course prefer the former, just to avoid the extra work in case you have improvements upon the tinyVAST implementation that you'd catch early in the process :)