misc/basic_script_HR.R

In agosiewska/shapper: Wrapper of Python Library 'shap'

library(reticulate)
#pd <- import("pandas")
shap <- import("shap")

library("DALEX")
library("randomForest")
Y_train <- factor(as.numeric(HR$status))
x_train <- HR[ , -6]
x_train$gender <- as.numeric(x_train$gender)

set.seed(123)
model_rf <- randomForest(x = x_train, y = Y_train)


pfun <- function(data){
 # data <- as.data.frame(as.matrix(data))
 # colnames(data) <- colnames(x_train)
 # data$gender <- factor(data$gender, levels = levels(x_train$gender))
 # res <- matrix(predict(model_rf, newdata = data, type = "prob"), ncol = 3)
 # colnames(res) <- c("fired", "ok","promoted")
 # res
  predict(model_rf, newdata = data, type = "prob")
}

pfun(x_train[2:10,])

x_train[,4] <- as.double(x_train[,4])

explainer = shap$KernelExplainer(pfun, x_train)
explainer = shap$kmeans(x_train, 100)

new_obs <- x_train[1,]
X_train <- r_to_py(new_obs)
shap_values = explainer$shap_values(new_obs, nsamples=100)
shap_values
shap_values = explainer$expected_value
shap_values