自定义recipes步骤step_hai_fourier prep阶段类型校验失败求助

问题描述 #

在开发自定义recipes步骤时，step_hai_fourier在prep阶段抛出类型不匹配错误，但测试用例中传入的变量b明确是数值类型（runif()生成的double类型）。错误提示如下：

Error in `prep()`:
! All variables for `step_hai_fourier` must be `numeric``integer` `double` classes.

问题根源 #

查看prep.step_hai_fourier中的类型检查逻辑：

if (any(value_data$type != "numeric")) {
  rlang::abort(
    paste0(
      "All variables for `step_hai_fourier` must be `numeric`",
      "`integer` `double` classes."
    )
  )
}

在recipes >=1.0.0版本中，info$type字段对数值类型做了细分：

• integer类型变量的type为"integer"
• double类型变量的type为"double"
• 仅通过as.numeric()转换的变量type才是"numeric"

测试用例中的b是runif()生成的double类型，因此value_data$type等于"double"，不满足== "numeric"的条件，触发错误。

修复方案 #

方案1：修正类型判断逻辑 #

将类型检查改为判断变量类型是否属于允许的数值类型集合：

#' @export
prep.step_hai_fourier <- function(x, training, info = NULL, ...) {
  col_names <- recipes::recipes_eval_select(x$terms, training, info)

  value_data <- info[info$variable %in% col_names, ]

  allowed_types <- c("numeric", "integer", "double")
  if (any(!value_data$type %in% allowed_types)) {
    rlang::abort(
      paste0(
        "All variables for `step_hai_fourier` must be of type ",
        paste(allowed_types, collapse = ", "), "."
      )
    )
  }

  step_hai_fourier_new(
    terms      = x$terms,
    role       = x$role,
    trained    = TRUE,
    columns    = col_names,
    scale_type = x$scale_type,
    period     = x$period,
    order      = x$order,
    skip       = x$skip,
    id         = x$id
  )
}

方案2：使用recipes内置类型检查函数（推荐） #

使用recipes::check_type()函数，这是recipes包提供的标准类型检查工具，能自动处理类型判断并抛出规范错误：

#' @export
prep.step_hai_fourier <- function(x, training, info = NULL, ...) {
  col_names <- recipes::recipes_eval_select(x$terms, training, info)

  # 用recipes内置函数检查类型，支持numeric/integer/double
  recipes::check_type(training[, col_names], types = c("numeric", "integer", "double"))

  step_hai_fourier_new(
    terms      = x$terms,
    role       = x$role,
    trained    = TRUE,
    columns    = col_names,
    scale_type = x$scale_type,
    period     = x$period,
    order      = x$order,
    skip       = x$skip,
    id         = x$id
  )
}

验证修复 #

修改后重新运行测试用例：

len_out <- 10
by_unit <- "month"
start_date <- as.Date("2021-01-01")

data_tbl <- tibble(
  date_col = seq.Date(from = start_date, length.out = len_out, by = by_unit),
  a = rnorm(len_out),
  b = runif(len_out)
)

rec_obj <- recipe(a ~ ., data = data_tbl) %>%
  step_hai_fourier(b, scale_type = "sin") %>%
  step_hai_fourier(b, scale_type = "cos") %>%
  step_hai_fourier(b, scale_type = "sincos")

# 现在prep可以正常执行
prepped_rec <- prep(rec_obj)
# 烘焙数据
baked_data <- bake(prepped_rec, data_tbl)

内容的提问来源于stack exchange，提问作者MCP_infiltrator