r - 如何解决 predict() : subscript out of bounds in R when doing binary classification? 中的错误

Question

我正在尝试使用 R 中的二进制分类来计算测试指标。我想我有正确的代码，但是我一直遇到错误。
目标是创建一个分类器，通过训练决策树模型来帮助检测给定其他变量的糖尿病。使用 0.50 的概率临界值为正类创建一个分类器。
以下是我的代码:

# load packages
library("mlbench")
library("tibble")
library("rpart")

# set seed 
set.seed(457)

# load data, remove NA rows, coerce to tibble
data("PimaIndiansDiabetes2")
diabetes = as_tibble(na.omit(PimaIndiansDiabetes2))

# split data
dbt_trn_idx = sample(nrow(diabetes), size = 0.8 * nrow(diabetes))
dbt_trn = diabetes[dbt_trn_idx, ]
dbt_tst = diabetes[-dbt_trn_idx, ]

# check data
dbt_trn

# fit models
mod_tree = rpart(diabetes ~ ., dbt_trn)

# get predicted probabilities for "positive" class, always use second alphabetically for +
prob_tree = predict(mod_tree, dbt_trn)[,"glucose"]

# create tibble of results for tree
results = tibble(
  actual   = dbt_tst$diabetes,
  prob_tree  = prob_tree,
)

# evaluate knn with various metrics
tree_eval = evaluate(
  data = results,
  target_col = "actual",
  prediction_cols = "prob_tree",
  positive = "diabetes",
  type = "binomial",
  metrics = list("Accuracy" = TRUE)
  cutoff = 0.5)

tree_eval 

我不断收到错误:
“预测错误(mod_tree，dbt_trn)[，“葡萄糖”]:
下标越界”
我不确定如何解决这个问题。任何帮助都会很棒!

Answer 1

predict的结果通话有栏目pos和 neg .此外，看起来您打算在测试集上进行预测。prob_tree应该这样定义:

prob_tree = predict(mod_tree, dbt_tst)[,"pos"]