partialResponse constant output for categorical variables in GBM

[AMBarbosa]

It's awesome that predicts::partialResponse() works for different types of models, but there are still some glitches. For categorial variables in a GBM, the result is a constant, unlike with gbm::plot.gbm():

library(gbm)
library(predicts)

N <- 1000
set.seed(1)
X1 <- runif(N)
set.seed(2)
X2 <- runif(N)
set.seed(3)
X3 <- ordered(sample(letters[1:4], N, replace = TRUE), levels = letters[4:1])
mu <- c(-1, 0, 1, 2)[as.numeric(X3)]
Y <- X1 ^ 1.5 + 2 * (X2 ^ 0.5) + mu

data <- data.frame(Y, X1, X2, X3)
head(data)

mod_gbm <- gbm(Y ~ X1 + X2 + X3, data = data)

partialResponse(mod_gbm, var = "X3", data = data)
#   X3        p
# 1  d 2.304339
# 2  c 2.304339
# 3  b 2.304339
# 4  a 2.304339

plot.gbm(mod_gbm, i.var = "X3", return.grid = TRUE)
#   X3         y
# 1  d 0.8268557
# 2  c 1.7450856
# 3  b 2.7247473
# 4  a 3.6364649

Categorical variables seem to be analysed correctly e.g. for GLM:

mod_glm <- glm(Y ~ X1 + X2 + X3, data = data)
partialResponse(mod_glm, var = "X3", data = data)
#   X3         p
# 1  d 0.7212272
# 2  c 1.7217522
# 3  b 2.7182770
# 4  a 3.7356409

[rhijmans]

Thank you. I now get:

partialResponse(mod_gbm, var = "X3", data = data, n.trees=100)
#  X3         p
#1  d 0.8230262
#2  c 1.7412560
#3  b 2.7209178
#4  a 3.6326354