Error for `as.matrix()` applied to GAMM projection

[fweber144]

Applying as.matrix() to a GAMM projection does not work:

# Source: Example section of `?rstanarm::stan_gamm4`.
library(projpred)
library(rstanarm)
options(mc.cores = parallel::detectCores(logical = FALSE))
dat <- mgcv::gamSim(1, n = 400, scale = 2) ## simulate 4 term additive truth
## Now add 20 level random effect`fac'...
dat$fac <- fac <- as.factor(sample(1:20, 400, replace = TRUE))
dat$y <- dat$y + model.matrix(~ fac - 1) %*% rnorm(20) * .5
br <- stan_gamm4(y ~ s(x0) + x1 + s(x2), data = dat, random = ~ (1 | fac),
                 chains = 1,
                 iter = 500, # for example speed
                 seed = 1140350788)

myproj <- project(br,
                  solution_terms = c("x1", "s(x2)", "(1 | fac)"),
                  nclusters = 2)
myprjmat <- as.matrix(myproj)

On branch develop (for example), that last line raises the following error:

Error in names(population_effects) <- replace_intercept_name(names(population_effects)) :
  attempt to set an attribute on NULL

The issue is probably that as.matrix.lm() (which is used for submodels of class "gamm4") uses coef() which in turn is NULL for objects of class "gamm4":

coef(myproj$sub_fit[[1]])
## --> NULL

So for submodels of class "gamm4", the as.matrix.lm() method probably can't be used.

[AlejandroCatalina]

Yes, I see. Why would gamm4 objects not implement coef? It seems it's a fairly generic method that could be used in this case as well, but anyway. I'll look at the PR.

[fweber144]

Yes, a corresponding coef() method would be helpful. But the documentation of gamm4::gamm4() says:

Returns a list with two items:

• gam: an object of class gam. [...]
• mer: the fitted model object returned by lmer or glmer. [...]

Perhaps you can exploit that?

[AlejandroCatalina]

Yes. I think we can possibly merge both coefficients matrices together. I'll play with it locally and see what comes out.