如果在 R 中的 Function 中发生错误，则跳过行

Question

I am currently trying to solve a bug but believe the data I am working with may be too complex and cause errors that shouldn't normally occur. I've written a function, and was hoping to add a try or tryCatch statement to skip the error if it occurs. I currently have:

library(glmnet)
foo <- function(data, ols_ps = TRUE, index) {
  # index is the bootstrap sample index
  x <- data[index, -1]
  y <- data[index, 1]
  ridge <- cv.glmnet(x, y, alpha = 0)
  ## The intercept estimate should be dropped.
  weights <- as.numeric(coef(ridge, s = ridge$lambda.min))[-1]
  # alpha=1, lasso
  alasso <- cv.glmnet(x, y, alpha = 1,
                      penalty.factor = 1 / abs(weights))
  # Select nonzero coefficients
  coef <- as.vector(coef(alasso, s = alasso$lambda.min, 
                         exact = TRUE, x = x, y = y,
                         penalty.factor = 1 / abs(weights)))[-1]
  if (ols_ps == TRUE) {
    coef_nonzero <- coef != 0
    new_x <- tryCatch(x[, coef_nonzero, drop = FALSE], 
                      error=function(e) NA)
    if (!any(is.na(new_x)) & ncol(new_x) > 0) {
      ls.obj <- lm(y ~ new_x)
      ls_coef <- (ls.obj$coefficients)[-1]
      coef[coef_nonzero] <- ls_coef
    } else {
      coef <- coef
    }
  } else {
    coef <- coef
  }
  return(coef)
}

which normally works and works on most datasets. I think the error may be coming from a complex dataset. Is it possible to skip OLS if I get the below error?

"Error in x[, coef_nonzero, drop = FALSE]: \n (subscript) logical subscript too long\n" attr(,"class")

Here is a minimal working example per request.

set.seed(123)
matrix <- matrix(runif(1000), ncol=10)
boot(matrix,foo,R=50)

Thanks in advance.

Answer 1

Maybe like this?

foo <- function(data, index) {
  # index is the bootstrap sample index
  x <- data[index, -1]
  y <- data[index, 1]
  ridge <- cv.glmnet(x, y, alpha = 0)
  ## The intercept estimate should be dropped.
  weights <- as.numeric(coef(ridge, s = ridge$lambda.min))[-1]
  # alpha=1, lasso
  alasso <- cv.glmnet(x, y, alpha = 1,
                      penalty.factor = 1 / abs(weights))
  # Select nonzero coefficients
  coef <- as.vector(coef(alasso, s = alasso$lambda.min, 
                         exact = TRUE, x = x, y = y,
                         penalty.factor = 1 / abs(weights)))[-1]
  coef_nonzero <- coef != 0
  new_x <- tryCatch(x[, coef_nonzero, drop = FALSE], 
                    error=function(e) NA)
  if (!any(is.na(new_x))) {
    ls.obj <- lm(y ~ new_x)
    ls_coef <- (ls.obj$coefficients)[-1]
    coef[coef_nonzero] <- ls_coef
  }
  return(coef)
}

The problem is that we have no case when it fails so far.