Currently thinking about the best way to get the credibility regions/MAP estimates.
If anyone wants to do a serious analysis using this type of simulation model, it’s possible to code a much more efficient and elegant simulation by making use of continuous time—having the data structure store times and locations of events rather than discrete snapshots. However, the continuous-time model would be harder to tinker with, so I opted for the more transparent discrete-time model for this tutorial.
The package “feature” on CRAN does multivariate density estimation. You might be able to use that for MAP estimation by running it on the posterior samples, and it also might be useful for HPD credible regions.
Currently thinking about the best way to get the credibility regions/MAP estimates.
If anyone wants to do a serious analysis using this type of simulation model, it’s possible to code a much more efficient and elegant simulation by making use of continuous time—having the data structure store times and locations of events rather than discrete snapshots. However, the continuous-time model would be harder to tinker with, so I opted for the more transparent discrete-time model for this tutorial.
The package “feature” on CRAN does multivariate density estimation. You might be able to use that for MAP estimation by running it on the posterior samples, and it also might be useful for HPD credible regions.