在R中标准化数据框数值变量时保留因子变量的技术问询

Hey there! I totally get why your factor variable APPL_SITE_NONSITE got dropped—likely your previous code was only selecting numeric/integer columns and overwriting the entire data frame, leaving out non-numeric types. Let's use a straightforward for loop with if/else logic to fix this, so we keep your factor variable intact while standardizing the numeric ones.

Step-by-Step Solution Code #

First, we'll work with a copy of your data frame to avoid messing up the original dcc:

# Create a copy of your data frame to preserve the original
dcc_scaled <- dcc

# Loop through each column in the data frame
for (col_name in colnames(dcc_scaled)) {
  # Check if the column is numeric or integer type
  if (is.numeric(dcc_scaled[[col_name]]) || is.integer(dcc_scaled[[col_name]])) {
    # Standardize: (value - mean) / standard deviation
    col_mean <- mean(dcc_scaled[[col_name]], na.rm = TRUE)
    col_sd <- sd(dcc_scaled[[col_name]], na.rm = TRUE)
    dcc_scaled[[col_name]] <- (dcc_scaled[[col_name]] - col_mean) / col_sd
  } else {
    # Leave non-numeric/factor columns as they are
    next
  }
}

What This Does #

• Preserves Original Data: By working on dcc_scaled (a copy), we don't alter your original dcc data frame—always a good practice!
• Targeted Standardization: The if statement checks if a column is numeric or integer. Only those columns get the (value - mean)/sd treatment.
• Keeps Factor Variables: The else clause uses next to skip over non-numeric columns (like your APPL_SITE_NONSITE factor), so they stay in the data frame untouched.
• Robust to NAs: We added na.rm = TRUE to mean() and sd() just in case—even though you said you filtered complete observations, this prevents errors if any sneaky NAs are left.

Verify the Result #

Run this to check that your factor variable is still there and numeric columns are scaled:

# Check variable types
str(dcc_scaled)

# Check summary stats of a scaled numeric column (mean ~0, sd ~1)
summary(dcc_scaled$your_numeric_column_here)

Bonus: Alternative (No Loop) #

If you ever want a more concise approach without loops, you can use dplyr (but since you asked for loops, this is just extra):

library(dplyr)
dcc_scaled <- dcc %>%
  mutate_if(is.numeric, ~(.x - mean(.x, na.rm = TRUE)) / sd(.x, na.rm = TRUE))

内容的提问来源于stack exchange，提问作者lambertj