Feature request: for a multivariate ordinal model, [add]_epred_draws returns response variables somewhere

[Fumeng-Yang]

For a multivariate ordinal model, brms still uses .category for ordinal responses; therefore, it seems that response variables
are not included in [add]_epred_draws (and [add]_epred_rvars?)

It would be nice to have the response variables somewhere, at least when resp = ... is specified.

A minimal reproducible example

set.seed(1234)
simulated_data <- 
  tibble(
    likert1 = sample(1:5, 100, replace = TRUE),
    likert2 = sample(1:5, 100, replace = TRUE)
  ) 

mvo <- brm(
       bf(mvbind(
          likert1, 
          likert2
           ) ~ 1) 
       + set_rescor(FALSE), # set correlation in residuals to FALSE
       data = simulated_data, 
       refresh = 2000, # don't want lots of messages
       backend = 'cmdstanr',
       family = cumulative('logit')
       )

add_epred_draws(object = mvo,
                                   # need some placeholders
                newdata = tibble(likert1 = -1, likert2 = -1),
                resp = c('likert1', 'likert2'),
                draw_ids = 100:102
                )

The output is

likert1	likert2	.row	.chain	.iteration	.draw	.category	.epred
-1	-1	1	NA	NA	1	1	0.1159525
-1	-1	1	NA	NA	2	1	0.1254349
-1	-1	1	NA	NA	3	1	0.2474628
-1	-1	1	NA	NA	1	2	0.1834422
-1	-1	1	NA	NA	2	2	0.2435328
-1	-1	1	NA	NA	3	2	0.1387139
-1	-1	1	NA	NA	1	3	0.2417583
-1	-1	1	NA	NA	2	3	0.2418615
-1	-1	1	NA	NA	3	3	0.1800300
-1	-1	1	NA	NA	1	4	0.3103983
-1	-1	1	NA	NA	2	4	0.1766087
-1	-1	1	NA	NA	3	4	0.2802388
-1	-1	1	NA	NA	1	5	0.1484487
-1	-1	1	NA	NA	2	5	0.2125622
-1	-1	1	NA	NA	3	5	0.1535545

Verify the output of brms:posterior_epred

dimnames(brms::posterior_epred(
    mvo,
    newdata = simulated_data[1,],
    draw_ids = 100:102
  ))

The output is

[[1]]
[1] "1" "2" "3"

[[2]]
NULL

[[3]]
 [1] "1" "2" "3" "4" "5" "1" "2" "3" "4" "5"

[Fumeng-Yang]

[add_][predicted|linpred]_draws appears to return response variables.

add_predicted_draws(object = mvo,
                # need some placeholders
                newdata = tibble(likert1 = -1, likert2 = -1),
                resp = c('likert1', 'likert2'),
                draw_ids = 100:102
                )

The output is

likert1	likert2	.row	.chain	.iteration	.draw	.category	.prediction
-1	-1	1	NA	NA	1	likert1	3
-1	-1	1	NA	NA	2	likert1	5
-1	-1	1	NA	NA	3	likert1	4
-1	-1	1	NA	NA	1	likert2	3
-1	-1	1	NA	NA	2	likert2	1
-1	-1	1	NA	NA	3	likert2	5