使用gbm.plot（dismo）和partial（pdp）时部分依赖图的差异，为什么？

Question

I am using boosted regression trees (gbm and dismo packages) to build habitat models and use partial dependence plots to look at the influence each predictor variable has on the response variable. I initially was using the gbm.plot function from the dismo package, but found the partial function from the pdp package which creates nicer looking plots however there are discrepancies between the two plots. See example images for latitude The lines look similar but the y-axis is different on both plots, and I am wondering why. I tried looking at the functions using print() but it does not yield much for partial.

final.model : fitted BRT

ceta_dd_final : data used to build BRT

Code for gbm.plot:

gbm.plot(final.model, n.plots = 17, write.title = T,
  show.contrib = T, y.label = "fitted function")

Code for partial:

final.model %>%
  partial(pred.var = "lat", plot=FALSE, train = ceta_dd_final, n.trees = 2400) %>%
  plotPartial(smooth = TRUE, train = ceta_dd_final, rug = TRUE,
      lwd = 2, ylab = expression(f(lat)))

Answer 1

不同之处在于gbm.plot使用普通logit作为绘图的结果，而partial使用以类为中心的logit进行此选择，以使y轴在二项式和多项式模型中具有相同的比例。