R kernelshap 包，带有 tidymodels 和分类

try 使用tidyModels为分类问题生成Shap值时出现问题.

当我在网站https://github.com/ModelOriented/kernelshap上按照步骤训练我的模型后，try 计算Shap值时，我不能为分类问题报告它.我的目标变量必须是因子.它总是返回:

#check_pred(pred_Fun(Object，X，...)，n=n)中出错: 运行函数kernelshap后，预测必须是数字

我用矩阵代替数据帧来寻找解决方案，并用Extra Fit Parnip提取模型.但问题仍然存在，除了分类，有没有办法重现这个例子. 示例代码如下

library(tidyverse)
library(tidymodels)


Default <- ISLR::Default


Default = Default %>%
  mutate(
    default = factor(case_when(
      default == "Yes" ~ 1,
      default == "No" ~ 0
    ), levels = c(1,0)
  ) 
)# changing to factor otherwise model will not work
  

# model fitting
split <- initial_split(Default)
  tr <- training(split)
  te <- testing(split)
  

  
  
rec <- recipe(default~., data = Default) %>% 
  step_dummy(all_nominal_predictors())




spec <- boost_tree() %>% 
  set_mode("classification") %>% 
  set_engine("xgboost")


wf <- workflow() %>% 
  add_model(spec) %>% 
  add_recipe(rec)

# model fit
mod <- fit(wf,tr)



library(kernelshap)

x <- rec %>% 
  prep %>% 
  bake(te %>% 
         slice_sample(n = 50)) %>% 
  select(-default) %>% 
  as.data.frame()
bg <- rec %>% prep %>% 
  bake(te %>% slice_sample(n = 10)) %>% 
  mutate(default = as.numeric(as.character(default))) %>% 
  as.data.frame()

# test for prediction
predict(mod, te)

# extract model form tidymodels
md <- extract_fit_parsnip(mod)

# this version works
kernelshap(md$fit, 
           X = x %>% as.matrix(), # if i do it with matrix structure then it works
           bg_X =bg %>% as.matrix()
           
           )

# this version does not work
kernelshap(mod, 
           X = te %>% 
             select(-default) %>%  # remove target var
             slice_sample(n = 50) %>% 
             as.data.frame(), 
           bg_X = te %>% 
             slice_sample(n = 50) %>% 
             as.data.frame()
           
)

################################### error messgae:

#Error in check_pred(pred_fun(object, X, ...), n = n) : 
#  Predictions must be numeric
######################################
              kernelshap(mod, 
                          X = te %>% 
                            select(-default) %>%  # remove target var
                            slice_sample(n = 50) %>% 
                            as.data.frame(), 
                          bg_X = te %>% 
                            slice_sample(n = 50) %>% 
                            as.data.frame()
              )

                          

# toy example from github page using tidymodels


library(tidymodels)
library(kernelshap)

iris_recipe <- iris %>%
  recipe(Sepal.Length ~ .)

reg <- linear_reg() %>%
  set_engine("lm")

iris_wf <- workflow() %>%
  add_recipe(iris_recipe) %>%
  add_model(reg)

fit <- iris_wf %>%
  fit(iris)

ks <- kernelshap(fit, iris[, -1], bg_X = iris)
ks