Element Replacement in R Programming

Introduction

Replacement changes existing values; classification creates a new interpretation of those values. Keep those operations separate. Mixing “High” and “Low” labels into a numeric temperature vector changes its storage type and can change the comparisons that follow.

Tested with R 4.5.1. Run the following code blocks in order. All inputs are included.

Preserve the numeric measurements

Derive a character category alongside the original numeric column. Missing temperatures remain missing categories.

readings <- data.frame(temperature = c(23, 28, 100, 9, NA_real_))
original <- readings$temperature
readings$category <- ifelse(readings$temperature > 25, "High", "Low")
print(readings)
stopifnot(identical(readings$temperature, original), is.numeric(readings$temperature))
stopifnot(identical(readings$category, c("Low", "High", "High", "Low", NA_character_)))

This remains correct for 100 and 9, where lexicographic comparisons after character coercion could give a misleading result. Do not overwrite the measured quantity merely because a category is useful for display.

Make a targeted numeric correction

The fictional rule below marks a sentinel value of -999 as missing. It does not assert that every negative temperature is invalid.

raw <- c(-5, -999, 20, NA_real_)
clean <- raw
sentinel <- !is.na(clean) & clean == -999
clean[sentinel] <- NA_real_
print(clean)  # -5 NA 20 NA
stopifnot(identical(clean, c(-5, NA_real_, 20, NA_real_)))
stopifnot(identical(raw, c(-5, -999, 20, NA_real_)))
changed_positions <- which(sentinel)
stopifnot(identical(changed_positions, 2L))

The explicit nonmissing mask prevents ambiguity in the replacement index. Preserve the original vector and record which positions changed. A missing numeric value is NA_real_, not the text “NA”.

Replace selected cells without replacing the whole object

Column selection keeps an identifier separate from the measurement. A matrix is homogeneous, so inserting text can coerce all of its values.

df <- data.frame(id = c("A", "B", "C"), score = c(10, -999, 30))
mask <- !is.na(df$score) & df$score == -999
df$score[mask] <- NA_real_
stopifnot(identical(df$id, c("A", "B", "C")))
m <- matrix(c(1, 2, 3, 4, 5, 6), nrow = 2, byrow = TRUE)
m[1, 2] <- 20
m[2, ] <- c(40, 50, 60)
print(m)  # rows: 1 20 3; 40 50 60
stopifnot(identical(dim(m), c(2L, 3L)), all(m[2, ] == c(40, 50, 60)))

A replacement vector should match the selected cells unless repetition is intentional. Do not let recycling repair a mismatch you have not explained; see matrix replacement-length checks.

Check empty selections and types

Empty selection is a legitimate no-op. Missing-value replacement should have an explicit policy; an all-missing column has no observed median to impute.

empty <- numeric(0)
empty[!is.na(empty) & empty < 0] <- 0
stopifnot(identical(empty, numeric(0)))
no_match <- c(1, 2)
no_match[no_match < 0] <- 0
stopifnot(identical(no_match, c(1, 2)))
coerced <- c(10, 20)
coerced[1] <- "High"
stopifnot(is.character(coerced))
all_missing <- c(NA_real_, NA_real_)
stopifnot(is.na(median(all_missing, na.rm = TRUE)))

For factors, add a permitted level before assigning a new label or deliberately convert labels to character. Avoid broad error handlers that substitute zeros: a programming error should not turn into an apparently valid measurement.

Primary references

• R extraction and replacement
• R ifelse