How to hack stargazer ci.custom with a multinom model?

Question

I'm trying to change my confidence intervals in stargazer() with ci.custom but it's not possible.

Using this exemple (used here):

library(dplyr)
library(nnet)

# create sample data
data <- tibble(
  choice = rep(c("strawberry", "mango", "orange"), 100),
  likes_kiwi = rbinom(300, size = 1, prob = 0.3),
  inner_beauty = rnorm(n = 300, mean = 5, sd = 2)
)

m1 <- multinom(choice ~ likes_kiwi + inner_beauty, data = data)

I have :

stargazer(m1, type = "text", ci = TRUE)

# =================================================
#   Dependent variable:      
#   -------------------------------
#                         orange        strawberry   
#                           (1)             (2)      
# -------------------------------------------------
#   likes_kiwi             0.166          -0.571*    
#                     (-0.403, 0.735) (-1.180, 0.038)
# 
# inner_beauty          -0.029          -0.062     
#                     (-0.178, 0.120) (-0.212, 0.087)
# 
# Constant               0.085           0.493     
#                     (-0.767, 0.936) (-0.345, 1.332)
# 
# -------------------------------------------------
#   Akaike Inf. Crit.     664.198         664.198    
# =================================================
#   Note:                 *p<0.1; **p<0.05; ***p<0.01

But if I try to change the intervals with ci.custom, it shows :

stargazer(m1, type = "text", ci.custom = list(confint(m1)))
# Error in ci.custom[[1]][, 1] : incorrect number of dimensions

Looking into .stargazer.wrap() function, I found (~L 3891) :

user.ci.lb=ci.custom[[1]][,1], user.ci.rb=ci.custom[[1]][,2]

and then :

.get.ci.lb.1 <- function(object.name, user.given=NULL, model.num=1) {
        if (!is.null(user.given)) { 
          if (.model.identify(object.name) == "multinom") {
            if (!is.null(nrow(user.given))) { user.given <- as.vector(user.given[model.num,]) }
          } 
          return(user.given) 
        }
        return(NULL)
      }
.get.ci.lb <- function(object.name, user.given=NULL, model.num=1) {
            
            out <- .get.ci.lb.1(object.name, user.given, model.num)
            
            coef.vars <- .coefficient.variables(object.name)
            if (is.null(names(out))) {  
              
              if (length(out) < length(coef.vars)) {
                out.temp <- rep(NA, times=length(coef.vars)-length(out))
                out <- c(out, out.temp)
              }
              else if (length(out) > length(coef.vars)) {
                out <- out[1:length(coef.vars)]
              }
              
              names(out) <- coef.vars   
            }
            else {
              out.temp <- rep(NA, times = length(coef.vars))
              names(out.temp) <- coef.vars
              for (i in 1:length(out)) {
                name <- names(out)[i]
                if (name %in% coef.vars) {
                  out.temp[name] <- out[i]
                }
              }
              out <- out.temp
            }
            return(out)
          }

(and the same form rb) where user.given takes user.ci.lb value ie ci.custom[[1]][,1].

The problem is that confidence intervals are arrays with 3 dimensions in multinom:

confint(m1)
# , , orange
# 
#                  2.5 %    97.5 %
# (Intercept)  -0.7666906 0.9364189
# likes_kiwi   -0.4032031 0.7350409
# inner_beauty -0.1779730 0.1200524
# 
# , , strawberry
# 
#                  2.5 %     97.5 %
# (Intercept)  -0.3452642 1.33184380
# likes_kiwi   -1.1797179 0.03782021
# inner_beauty -0.2115865 0.08737243

but it wants only 2 dimensions. By the way, stargazer uses .get.ci.lb and .get.ci.lb.1 for one part of the model at the time (orange, strawberry), so I cannot give a big matrix to it.

At first I thought that making array transpositions would help but it doesn't. Do you have any idea of how to resolve the problem ?

Thanks,