Ordinal models no longer include level names in predict() output

[mjskay]

In previous versions of brms (up to 2.10 at least), output of predict() on an ordinal model would include the level names from the response variable in column names when summary = TRUE and would include a "levels" attribute for column names when summary = FALSE, which is very useful for plotting / interpreting model output. E.g.:

library(brms)
#> Loading required package: Rcpp
#> Loading 'brms' package (version 2.10.0). Useful instructions
#> can be found by typing help('brms'). A more detailed introduction
#> to the package is available through vignette('brms_overview').

d = data.frame(
  x = ordered(c(rep("a", 20), rep("b", 10), rep("c", 5)))
)

m = brm(x ~ 1, data = d, family = cumulative)
#> Compiling the C++ model
#> Start sampling
#> 
#> [...]

str(predict(m))
#>  num [1:35, 1:3] 0.563 0.568 0.571 0.559 0.568 ...
#>  - attr(*, "dimnames")=List of 2
#>   ..$ : NULL
#>   ..$ : chr [1:3] "P(Y = a)" "P(Y = b)" "P(Y = c)"
str(predict(m, summary = FALSE))
#>  num [1:4000, 1:35] 1 2 2 1 3 1 3 2 1 2 ...
#>  - attr(*, "dimnames")=List of 2
#>   ..$ : NULL
#>   ..$ : NULL
#>  - attr(*, "levels")= chr [1:3] "a" "b" "c"

^{Created on 2020-01-25 by the reprex package (v0.3.0)}

As of the version of CRAN (2.11) and the current github version (reprex below), I no longer see level names in the output, only numbers. E.g.:

library(brms)
#> Loading required package: Rcpp
#> Loading 'brms' package (version 2.11.4). Useful instructions
#> can be found by typing help('brms'). A more detailed introduction
#> to the package is available through vignette('brms_overview').
#> 
#> Attaching package: 'brms'
#> The following object is masked from 'package:stats':
#> 
#>     ar

d = data.frame(
  x = ordered(c(rep("a", 20), rep("b", 10), rep("c", 5)))
)

m = brm(x ~ 1, data = d, family = cumulative)
#> Compiling the C++ model
#> Start sampling
#> 
#> [...]

str(predict(m))
#>  num [1:35, 1:3] 0.575 0.562 0.577 0.561 0.563 ...
#>  - attr(*, "dimnames")=List of 2
#>   ..$ : NULL
#>   ..$ : chr [1:3] "P(Y = 1)" "P(Y = 2)" "P(Y = 3)"
str(predict(m, summary = FALSE))
#>  num [1:4000, 1:35] 1 1 2 1 1 1 2 1 1 2 ...
#>  - attr(*, "dimnames")=List of 2
#>   ..$ : NULL
#>   ..$ : NULL

^{Created on 2020-01-25 by the reprex package (v0.3.0)}

Is this a bug, or is there another way to retrieve the level names? Thanks!

[paul-buerkner]

Thanks! In 2.11 I decoupled the handling of categorical and ordinal models much more than before, which now makes the internals easier to handle. Unfortunately, I overlooked that response category names are still beneficial to store in ordinal models. This should be fixed now.

[mjskay]

Awesome, thanks!