Open Martin-Jung opened 2 years ago
Some words seem to be missing (markdown issue?), but I get the main points, and the predict()
output suffers from this issue as well; there's currently no option to safely alter the quantiles (although one can do it via manual processing of generate()
output).
inla
and predict
broom
style tidy()
output option/postprocessing method added, even though the naming scheme from https://www.tidymodels.org/learn/develop/broom/#glossary has some terminology issues for Bayesian outputs)Part 1. has now been fixed; predict()
now takes a probs
argument for specifying quantile probabilities.
By default
inlabru
calculates a 0.025, 0.5 and 0.975 quantile and uses these measurements in summary functions and plots. However for a specific application I prefer to use other quantiles in the estimation. If I pass on the quantiles to be calculated as an explicit option ininlabru
, these are taken into account in the resulting inla call. E.g., via:inlabru::bru_options_set(quantiles = c(0.05, 0.5, 0.95)) # Set quantiles to be computed
However many of the inlabru functions still have the default quantiles hardcoded in. For instance here or here. Any model fitted via the custom bru options specified above subsequently results in errors. Calling for instance
inlabru:::plot.prediction
on a model object with other specified quantiles returns thisA solution would be to add
grep
on the posterior names to get those columns withquant
in its name.