使用cv.glmertree时出现family$family符号不可子集化报错求助

我使用R语言的glmertree包运行广义线性混合模型树，能成功拟合glmertree模型，但调用cv.glmertree函数获取标准误与推断结果时触发报错：

Error in family$family : object of type 'symbol' is not subsettable

我的代码如下：

train <- sample(1:nrow(non_miss), size =2382 , replace = TRUE)
test <- sample(1:nrow(non_miss), size =2382 , replace = TRUE)

tree1 <- glmertree(event ~ 1 | (1| cohort) | ageatenroll + hiv + racecat  + visitsite 
                  + edu + smkgrp_hist + percent_drug + comorbid_hist + percent_depress, 
                 data=non_miss[train, ], 
                 family=binomial, 
                 nAGQ = 0,
                 minsize=200)

tree2 <- cv.glmertree(tree1, newdata=non_miss[test, ], reference=13)

报错回溯信息：

> tree2 <- cv.glmertree(tree1, newdata = non_miss[test, ], reference = 13)
Error in family$family : object of type 'symbol' is not subsettable
> traceback()
6: family$family %in% c("quasibinomial", "quasipoisson", "quasi")
5: lme4::glFormula(formula = ff, data = newdata, family = family, 
   control = glmer.control, nAGQ = nAGQ)
4: eval(mc, parent.frame(1L))
3: eval(mc, parent.frame(1L))
2: glmer(ff, data = newdata, nAGQ = nAGQ, family = family, control = glmer.control)
1: cv.glmertree(tree1, newdata = non_miss[test, ], reference = 13)

排查与解决建议 #

• 显式传递family对象给cv.glmertree：报错核心是family被当作符号而非family对象传递。在调用cv.glmertree时手动指定family = binomial()（注意带括号，返回的是family对象而非符号），替代默认从tree1中提取的参数：

  tree2 <- cv.glmertree(tree1, newdata=non_miss[test, ], reference=13, family = binomial())
  

• 验证数据集一致性：确认non_miss[test, ]的变量类型、因子水平与训练集non_miss[train, ]完全匹配，避免glFormula处理数据时出现异常
• 更新glmertree包：旧版本包可能存在参数传递的bug，运行update.packages("glmertree")更新到最新版后重试
• 显式指定glmer控制参数：在cv.glmertree中添加control = glmer.control()，确保控制参数正确传递给内部调用的glmer函数

内容的提问来源于stack exchange，提问作者DeWa0611