`as.matrix.projection()` for GAMs: Naming scheme of parameters

[fweber144]

For GAMs, as.matrix.projection() does not follow brms's naming scheme (discussed in #75):

On branch develop (commit 96d8e11), this reprex:

options(mc.cores = parallel::detectCores(logical = FALSE))
library(brms)
data(kidiq, package = "rstanarm")
fit_gauss <- brm(formula = kid_score ~ s(mom_iq),
                 data = kidiq,
                 seed = 469230)

library(projpred)
prj <- project(fit_gauss, solution_terms = c("s(mom_iq)"),
               ndraws = 25)
colnames(as.matrix(prj))

gives

 [1] "b_Intercept"   "b_s(mom_iq).1" "b_s(mom_iq).2" "b_s(mom_iq).3"
 [5] "b_s(mom_iq).4" "b_s(mom_iq).5" "b_s(mom_iq).6" "b_s(mom_iq).7"
 [9] "b_s(mom_iq).8" "b_s(mom_iq).9" "sigma"

whereas

colnames(as.matrix(fit_gauss))

gives

 [1] "b_Intercept"    "bs_smom_iq_1"   "sds_smom_iq_1"  "sigma"         
 [5] "s_smom_iq_1[1]" "s_smom_iq_1[2]" "s_smom_iq_1[3]" "s_smom_iq_1[4]"
 [9] "s_smom_iq_1[5]" "s_smom_iq_1[6]" "s_smom_iq_1[7]" "s_smom_iq_1[8]"
[13] "lp__"

Just a side note: As may be seen in issues #150 and #151, rstanarm uses a different naming scheme. It might make sense to have separate naming schemes depending on whether the reference model inherits from a "stanreg", a "brmsfit", or some different object. (In the latter case, one could use either rstanarm's or brms's naming scheme.) But that's probably a larger change and may be deferred to the future. I think for now, using brms naming scheme consistently should be enough.