scattered examples

[bbolker]

I'm opening this issue to consider the fragmentation of TMB examples. I recently created an example of vector-valued random effects (mostly for handling of unstructured covariance matrices, which I now see is handled by one of the examples below), and was thinking about adding it here. Then I realized there are at least three different places where examples are tabulated:

• https://github.com/admb-project/tmb-examples : cpp + R + README, fairly detailed comments in code
• https://github.com/skaug/tmb-case-studies: this repo. cpp+R+ full explanatory Rmd.
• https://kaskr.github.io/adcomp/_book/Examples.html#example-overview : minimal examples. Text says there are both .R and .cpp files for each example, and refers to '"Examples” tab on the top of the page' — to find that tab you have to go here to see a Doxygen-style list, but the .R files aren't linked from there -- you have to go to the corresponding directory in the GitHub repo to find them)
• https://gist.github.com/dsjohnson/9d66aa47557ad56438aaf75dd25910ea: a useful gist (about incorporating mgcv-derived spline bases in TMB models) that I don't want to lose track of

At a minimum it would probably be useful for all three of these sources to point at each other. Don't know whether it's worth going further (having one of these places collect the examples from the other, create a unified list ...)

[danielinteractive]

Thanks @bbolker for raising this, I think as a new comer to TMB I would definitely benefit from a one-stop-example shop :-)

[dsjohnson]

Given the fact that it's been duplicated several times, in slightly different contexts, it might be useful to have a more unified version. But, do you mean {glmmTMB} examples or {TMB} in general? Because all the spatial and time series examples qualify as 'vector-valued' random effects.

[bbolker]

Should have been more specific: "simple linear mixed model with vector-valued random effects, unstructured (general positive-definite) covariance matrix". I found two examples that involve unstructured covariance matrices:

• Arni's plain multivariate normal example ("multivariate normal 2" on the ADMB project repo here) (also sparsely commented, although there is a YouTube video [!]; also, this one doesn't deal with latent variables/random effects)
• Stochastic volatility model (Skaug & Yu) on the TMB repo here (very sparsely commented)
  I'm not

[skaug]

I can put together something with emphasis on @bbolker last suggestion, but also serve the purpose of pointing to existing examples/literature. Unless anyone have a suggestion for an appropriate dataset, I will use simulated data. I feel that
seeing the R-code for the simulation nails down exactly what the model is.

[bbolker]

Sorry for any confusion. I have written such an example here; it uses an existing real data set, but I can easily simulate a data set instead/in addition. While I was working on writing the description I went looking for existing examples and got distracted by the fact that they are scattered about.

(I actually came to the issues list to ask about the markup system you're using, but I'll ask that in a separate issue ...)

[skaug]

@bbolker : are you aiming at pushing it back into my repository? In case I would reformat the example slightly (probably) to make it be more in line with the other case-studies.

[bbolker]

I was thinking I would edit to match the other case studies, but I hadn't gotten around to it yet ...