Partial dependence plot appears in the opposite way

[mcchiu]

I build the xgboost model.
The partial dependence plot (red line) appears in the opposite way.
Could you please help out?

set.seed(1278)
xgb_f <- xgboost(data = data, 
                 label = type, 
                 eta = 0.1,
                 max_depth = 15, 
                 nround = 35, 
                 subsample = 0.5,
                 colsample_bytree = 0.5,
                 objective = "reg:squarederror",
                 nthread = 4,
                 verbose = FALSE
)

 xgb_f %>%
    partial(pred.var = "airTEMP", grid.resolution = 100, 
            train = data, ice = TRUE, type = "regression") %>%
    autoplot(rug = TRUE, train = data, alpha = .1, center = TRUE)

[bgreenwell]

@mcchiu do you have the code to reproduce data and type?

[mcchiu]

@bgreenwell
Here is the code to reproduce data and type.

dataAll <- read.csv("rawdatasub1.csv")
type <- data.matrix(dataAll$aquarticflux)
data <- data.matrix(dataAll[,! names(dataAll) %in% "aquarticflux"])

[bgreenwell]

Thanks @mcchiu, I'll take a look. For now, plotPartial(airTEMP, alpha = 0.1, center = TRUE) (which uses lattice) seems to work fine.

[bgreenwell]

Still not sure what the exact issue is, but here's a quick fix if you still want to use ggplot2

d <- pdp:::center_ice_curves.ice(airTEMP)  # center manually
ggplot(d, aes(airTEMP, yhat, group = yhat.id)) +
  geom_line(alpha = 0.1) +
  geom_line(data = pdp:::average_ice_curves.ice(d), color = "red",
            aes(airTEMP, yhat), inherit.aes = FALSE)

[mcchiu]

@bgreenwell Thanks for the help. Either works. I look forward to your new release.