使用pdp包获取所有类的概率标度偏相关图

Question

I have been following the example here to create partial dependence plots but I would like to combine the approach used to get plots for all levels in a multiclass with the one to get predictions on the probability scale (see pages 430-431). This is my approach but it doesn't work because pred.fun is not allowed to have a third arguement

library(e1071)

iris.svm <- svm(Species ~ ., data = iris, kernel = "radial", gamma = 0.75,
                cost = 0.25, probability = TRUE)

pred.prob <- function(object, newdata,i) { # see ?predict.svm
  pred <- predict(object, newdata, probability = TRUE)
  prob.class <- attr(pred, which = "probabilities")[, i]
  mean(prob.class)
}

pred.prob(iris.svm,iris,"setosa")

pd <- NULL

for (i in 1:3) {
  tmp <- partial(iris.svm, pred.var = c("Petal.Width", "Petal.Length"),
                 pred.fun = pred.prob,
                 which.class = i, grid.resolution = 101, progress = "text")
  pd <- rbind(pd, cbind(tmp, Species = levels(iris$Species)[i]))
}

Any recommendations for how to get around this requirement or a different approach?

Answer 1

It looks like the package has actually been updated since the article I referred to was published. Now all you need to do is set the prob argument to TRUE and it will predict on the probability scale.

pd <- NULL    
for (i in 1:3) {
  tmp <- partial(iris.svm, pred.var = c("Petal.Width", "Petal.Length"),
                 prob = T,
                 which.class = i, grid.resolution = 101, progress = "text")
  pd <- rbind(pd, cbind(tmp, Species = levels(iris$Species)[i]))
}

I hope this helps someone else to avoid wasting an afternoon!