I’ve trained a BART model on a binomial classification task. Among the predictors is a factor with 18 levels. They are as follows:

levels(poke$Type.1)
 [1] "Bug"      "Dark"     "Dragon"   "Electric" "Fairy"    "Fighting"
 [7] "Fire"     "Flying"   "Ghost"    "Grass"    "Ground"   "Ice"     
[13] "Normal"   "Poison"   "Psychic"  "Rock"     "Steel"    "Water" 

When I read the BART object, specifically the varcount.mean (an array), to determine variable importance, I get the following difficult-to-interpret factor names (note the Type ones):

bart1$varcount.mean
   Type.11    Type.12    Type.13    Type.14    Type.15    Type.16 
     1.930      1.825      1.782      1.804      1.983      1.864 
   Type.17    Type.18    Type.19   Type.110   Type.111   Type.112 
     1.913      0.000      1.950      1.977      1.983      1.987 
  Type.113   Type.114   Type.115   Type.116   Type.117   Type.118 
     2.105      2.004      1.871      2.102      1.906      2.342 
     Total         HP     Attack    Defense    Sp..Atk    Sp..Def 
     4.400      2.266      2.415      2.175      2.508      2.652 
     Speed Generation 
     2.711      2.281 

My question is – how would you solve this issue, short of renaming each row manually? Is there an argument I can pass to bart, or a convenient function I can use to rename the output rows in the bart object?

Here is the program I have written:

rm(list=ls())
graphics.off

library(BART)
library(tidyr)
library(dplyr)

library(ggplot2)
library(ROCR)

pokeB<-read.csv("~/Downloads/Pokemon.csv", header=T)

Legend<-vector(length=800) %>% rep(0, 800)
pokeB<-data.frame(pokeB, Legend)
pokeB$Legend<-as.integer(
  ifelse(pokeB$Legendary=="True","1","0")
)

poke<-pokeB %>% select(Type.1,Total,HP,Attack,Defense,
                       Sp..Atk,Sp..Def,Speed,Generation,
                       Legend)
poke$Type.1<-as.factor(poke$Type.1)

set.seed(1)
train<-sample(1:nrow(poke), nrow(poke)/2)

x<-poke %>% select(-Legend)
y<-poke[,"Legend"]
xtrain<-x[train,]
ytrain<-y[train]
xtest<-x[-train,]
ytest<-y[-train]

bart1<-mc.gbart(xtrain, ytrain, x.test=xtest, type='pbart',
                mc.cores=4)

ord1<-order(bart1$varcount.mean,decreasing=T)
vars1<-as.data.frame(bart1$varcount.mean[ord1])
pred1<-ifelse(bart1$prob.test.mean>0.5, 1, 0)
tab1<-table(ytest, as.factor(pred1))

To see the messy factor names output in the bart object, just view vars1. My goal is to either have the output of the bart object preserve original factor level names, or to edit the vars1 dataframe to restore original level names in a convenient manner (i.e. not going manually row-by-row).

Thank you for reading