R中glm自定义单变量逻辑回归函数无法运行问题求助

Let's walk through the issues in your current code and get it working smoothly.

Key Problems in Your Original Code #

1. Incorrect lapply syntax: You’re assigning the function univariate_logistic inside the lapply call, which isn’t the standard or reliable way to pass a function to lapply. While R might try to evaluate this, it’s prone to unexpected behavior.
2. Manual variable list: Typing A1 to A50 by hand is error-prone and unnecessary—we can generate this list programmatically in one line.
3. Not storing results: If you don’t assign the output of lapply to a variable, you won’t be able to access or review the regression summaries later.

Corrected Code #

First, generate your predictor list cleanly to avoid typos:

# Create list of predictor variables (A1 to A50)
predictors <- paste0("A", 1:50)

Then, use one of these two approaches to run your regressions:

Option 1: Anonymous Function Directly in lapply

This is concise for one-off use:

# Run regressions and store all summaries
univariate_results <- lapply(predictors, function(var) {
  # Build the formula dynamically
  formula <- as.formula(paste("churn_flag ~", var))
  # Fit the logistic regression model
  model <- glm(formula, data = split_train, family = binomial)
  # Return the model summary (this becomes the list element)
  summary(model)
})

# Name list elements so you can easily reference specific results
names(univariate_results) <- predictors

Option 2: Separate Reusable Function

If you want to reuse this logic later, define a dedicated function first:

# Define a function to run univariate logistic regression
run_univariate_logistic <- function(var, data) {
  formula <- as.formula(paste("churn_flag ~", var))
  model <- glm(formula, data = data, family = binomial)
  return(summary(model))
}

# Apply the function to all predictors
univariate_results <- lapply(predictors, run_univariate_logistic, data = split_train)
names(univariate_results) <- predictors

Quick Checks to Avoid Further Issues #

• Ensure churn_flag is binary: The binomial family in glm requires the response to be either a factor with two levels (e.g., "Yes"/"No") or a numeric vector of 0s and 1s. If your churn_flag is coded differently, convert it first:

  split_train$churn_flag <- factor(split_train$churn_flag)
  

• Handle missing data: glm drops rows with missing values by default. If you need to keep those rows (or want explicit handling), add na.action = na.exclude to your glm call.
• Access results easily: You can now pull specific summaries using variable names, like univariate_results[["A1"]] to view the output for predictor A1.

内容的提问来源于stack exchange，提问作者James