R语言中自定义函数foo的Bootstrap实现报错求助

The Root Cause #

Your error stems from inconsistent output length of your foo() function across Bootstrap resamples. Here's the breakdown:

• When you run foo(dat) on your original data, it returns a vector with length equal to the number of unique values in dat (1,2,3,4 → 4 values).
• But during Bootstrap resampling, some resampled datasets might miss one or more of these unique values. For example, a resample could have no rows with value 2, so foo(resample) would only return 3 values.
• The boot package requires the statistic function (boot_fun) to return a vector of fixed length every time. Mismatched lengths trigger that replacement error.

The Fix #

We need to modify foo() to always return a vector with length matching the unique values from your original dataset—even if a resample is missing some values. Here's how to adjust your code:

1. First, define the fixed set of unique values from your original data to use as reference levels.
2. Update foo() to accept a fixed_levels parameter, ensuring we always factorize values against this fixed set.
3. Modify boot_fun to pass these fixed levels to foo() during each resample.

Modified Code #

library(boot)

# Updated foo function with fixed levels parameter
foo <- function(X, fixed_levels) {
  X <- as.matrix(X)
  # Use fixed_levels instead of deriving from X to keep columns consistent
  tab <- table(row(X), factor(X, levels = fixed_levels))
  w <- diag(ncol(tab))
  rosum <- rowSums(tab)
  obs_oc <- tab * (t(w %*% t(tab)) - 1)
  obs_c <- colSums(obs_oc)
  max_oc <- tab * (rosum - 1)
  max_c <- colSums(max_oc)
  SA <- obs_c / max_c
  h <- names(SA)
  h[is.na(h)] <- "NA"
  setNames(SA, h)
}

# Original data
dat <- data.frame(a = 1:4, b = c(2,1, 3, 4))
# Extract fixed unique levels from original data
fixed_levels <- sort(unique(as.vector(dat)))

# Test the updated foo function
foo(dat, fixed_levels)

# Updated bootstrap function
boot_fun <- function(data, i, fixed_levels){
  resample <- data[i, ,drop = FALSE]
  foo(resample, fixed_levels)
}

# Run Bootstrap successfully
boot_result <- boot::boot(data = dat, statistic = boot_fun, R = 200, fixed_levels = fixed_levels)
print(boot_result)

Why This Works #

• By forcing factor(X, levels = fixed_levels), we ensure the tab table always has the same number of columns (matching the original data's unique values), even if a resample is missing some values.
• Missing values in resamples will result in NA in the corresponding position of the SA vector, but the length stays consistent—exactly what boot needs to run without errors.

内容的提问来源于stack exchange，提问作者rnorouzian