Formula parsing error with t2() interaction terms

[hsbadr]

varsel()/make_formula fails with t2() interaction terms (see paul-buerkner/brms#1133).

Here's a simple reproducible example:

library(brms)
library(projpred)

dat <- mgcv::gamSim(1, n = 400, scale = 2)
fit1 <- brm(y ~ x1 + t2(x0, x2), data = dat)

refmod1 <- get_refmodel(fit1)
vs1 <- varsel(refmod1)

Error in str2lang(x) : :1:29: unexpected ','
1: . ~ 1 + t2(x0, x2) + x1 + x0,
^

Here's the traceback:

> traceback()
9: str2lang(x)
8: formula.character(object, env = baseenv())
7: formula(object, env = baseenv())
6: as.formula(paste0(". ~ ", paste(terms_, collapse = " + "))) at formula.R#595
5: make_formula(list_of_terms) at formula.R#729
4: count_terms_chosen(search_terms, duplicates = TRUE) at varsel.R#327
3: parse_args_varsel(refmodel, method, cv_search, intercept, nterms_max,
       nclusters, ndraws, nclusters_pred, ndraws_pred, search_terms) at varsel.R#118
2: varsel.refmodel(refmod1) at varsel.R#93
1: varsel(refmod1)

A quick debug reveals that projpred:::make_formula() uses a term ("x0, x2") generated by projpred:::split_formula() that includes the two variables (x0 and x2) separated by a comma, which causes the failure:

> projpred:::split_formula(refmod1$formula)
[1] "1"          "t2(x0, x2)" "x1"         "x0, x2"

> paste0(". ~ ", paste(projpred:::split_formula(refmod1$formula), collapse = " + "))
[1] ". ~ 1 + t2(x0, x2) + x1 + x0, x2"

Either the added term ("x0, x2") needs to be removed or the comma replaced by +.

[AlejandroCatalina]

Exactly, we did not implement more complex splines. Thanks for the PR, I'll look into it now!

[AlejandroCatalina]

Fixed in latest develop.