R6类调用stats::lm()时weights与subset参数无法识别问题求助

问题原因 #

当调用stats::lm()并指定data参数时，函数会默认在data对应的环境中解析weights和subset参数，而非当前R6方法的局部函数环境。你在方法内定义的weight_vec和subset_vec属于局部变量，不在private$data的环境中，因此lm无法找到这些变量。

方案1：临时将变量加入data中（推荐） #

直接把权重或子集向量临时添加到private$data中，让lm能在指定数据集里找到对应变量，用完后再删除临时列避免污染数据：

gen_reg_data <- R6::R6Class(
  classname = "regressionClass", 
  public = list(
    initialize = function(data){
      private$data <- data
    }, 
    lm = function(formula){
      stats::lm(formula = formula, data = private$data)
    }, 
    weighted_lm = function(formula, weight_col){
      # 临时添加权重列到data
      private$data$.temp_weight <- abs(private$data[, weight_col])
      print(head(private$data$.temp_weight))
      result <- tryCatch(
        stats::lm(formula = formula,
                  weights = .temp_weight, 
                  data = private$data), 
        error = function(e) e
      )
      # 用完删除临时列
      private$data$.temp_weight <- NULL
      result
    }, 
    subset_lm = function(formula){
      # 临时添加子集列到data
      private$data$.temp_subset <- private$data$X10 >= 0
      print(head(private$data$.temp_subset))
      result <- tryCatch(
        stats::lm(formula = formula,
                  subset = .temp_subset,
                  data = private$data), 
        error = function(e) e
      )
      # 删除临时列
      private$data$.temp_subset <- NULL
      result
    }
  ), 
  private = list(
    data = NULL
  )
)

方案2：手动指定lm的求值环境 #

将公式的环境设置为当前方法的局部环境，让lm优先在局部环境中查找变量：

gen_reg_data <- R6::R6Class(
  classname = "regressionClass", 
  public = list(
    initialize = function(data){
      private$data <- data
    }, 
    lm = function(formula){
      stats::lm(formula = formula, data = private$data)
    }, 
    weighted_lm = function(formula, weight_col){
      weight_vec <- abs(private$data[, weight_col])
      print(head(weight_vec))
      # 将公式的环境设置为当前函数环境
      formula_obj <- as.formula(formula)
      environment(formula_obj) <- environment()
      tryCatch(
        stats::lm(formula = formula_obj,
                  weights = weight_vec, 
                  data = private$data), 
        error = function(e) e
      )
    }, 
    subset_lm = function(formula){
      subset_vec <- private$data$X10 >= 0
      print(head(subset_vec))
      formula_obj <- as.formula(formula)
      environment(formula_obj) <- environment()
      tryCatch(
        stats::lm(formula = formula_obj,
                  subset = subset_vec,
                  data = private$data), 
        error = function(e) e
      )
    }
  ), 
  private = list(
    data = NULL
  )
)

方案3：使用with()包裹lm调用 #

通过with()函数将data环境和局部环境结合，让lm能同时访问data和局部变量：

gen_reg_data <- R6::R6Class(
  classname = "regressionClass", 
  public = list(
    initialize = function(data){
      private$data <- data
    }, 
    lm = function(formula){
      stats::lm(formula = formula, data = private$data)
    }, 
    weighted_lm = function(formula, weight_col){
      weight_vec <- abs(private$data[, weight_col])
      print(head(weight_vec))
      tryCatch(
        with(private$data, 
             stats::lm(formula = formula, weights = weight_vec)), 
        error = function(e) e
      )
    }, 
    subset_lm = function(formula){
      subset_vec <- private$data$X10 >= 0
      print(head(subset_vec))
      tryCatch(
        with(private$data, 
             stats::lm(formula = formula, subset = subset_vec)), 
        error = function(e) e
      )
    }
  ), 
  private = list(
    data = NULL
  )
)

内容的提问来源于stack exchange，提问作者Zhenning