purrr map() for Data Wrangling: Read Files, Transform Lists & Fit Models

Q: When should I use map() vs a for loop?

Use map() when each iteration is independent and returns a value. Use a for loop when iterations depend on previous results or need break/next control flow.

Q: How does map() compare to lapply()?

They do the same thing. map() adds typed variants (map_dbl, map_chr), formula shorthand (~ .x + 1), and element extraction by name (map(list, "field")). Use lapply() for zero dependencies, map() for everything else.

purrr's map() family shines in data wrangling: reading multiple CSV files at once, extracting stats from nested data, fitting models per group, and safely handling messy inputs. These are the patterns you'll use daily.

Pattern 1: Read and Combine Multiple Files

library(purrr) library(readr) # Simulate 3 monthly CSV files csv_jan <- "date,sales\n2026-01-01,100\n2026-01-02,120" csv_feb <- "date,sales\n2026-02-01,150\n2026-02-02,130" csv_mar <- "date,sales\n2026-03-01,170\n2026-03-02,160" csvs <- list(jan = csv_jan, feb = csv_feb, mar = csv_mar) # Read all, add source name, combine combined <- imap(csvs, \(csv, name) { df <- read_csv(csv, show_col_types = FALSE) df$month <- name df }) |> list_rbind() print(combined)

Pattern 2: Extract from List Columns

library(purrr) library(dplyr) # Nested data: each row has a vector of scores students <- tibble( name = c("Alice", "Bob", "Carol"), scores = list(c(88, 92, 79), c(76, 81, 85), c(92, 95, 88)) ) students |> mutate( avg = map_dbl(scores, mean), best = map_dbl(scores, max), n = map_int(scores, length) )

Pattern 3: Split → Map → Combine (Model Fitting)

library(purrr) library(dplyr) # Fit a model per cylinder group models <- mtcars |> split(mtcars$cyl) |> map(\(df) lm(mpg ~ wt + hp, data = df)) # Extract R-squared per group r_sq <- map_dbl(models, \(m) summary(m)$r.squared) cat("R² per cyl group:\n") print(round(r_sq, 3))

Pattern 4: Safely Process Messy Data

library(purrr) # Some conversions will fail — safely catches errors raw <- list("42", "3.14", "abc", "100", "xyz") safe_num <- possibly(as.numeric, otherwise = NA) results <- map_dbl(raw, safe_num) cat("Parsed:", results, "\n") cat("Success:", sum(!is.na(results)), "/", length(results), "\n")

Pattern 5: Named Iteration with imap

library(purrr) quarterly <- list(Q1 = c(150,200,180), Q2 = c(250,300,275), Q3 = c(180,190,210), Q4 = c(350,400,380)) imap_chr(quarterly, \(sales, qtr) { sprintf("%s: total=$%d, avg=$%.0f", qtr, sum(sales), mean(sales)) }) |> walk(~ cat(.x, "\n"))

Practice Exercises

Exercise 1: Per-Species Summary

Split iris by Species, compute summary stats per group, and combine into one data frame.

library(purrr) library(dplyr) # Use split + imap_dfr to create a summary table

Click to reveal solution

```r

library(purrr) library(dplyr) iris |> split(iris$Species) |> imap_dfr(\(df, species) { tibble( species = species, n = nrow(df), mean_sl = round(mean(df$Sepal.Length), 2), sd_sl = round(sd(df$Sepal.Length), 2) ) })

Summary

Pattern	Code
Read multiple files	`map(files, read_csv) \	> list_rbind()`
Extract from list col	`mutate(stat = map_dbl(col, fn))`
Model per group	`split(df, group) \	> map(lm_fn)`
Safe processing	`map(x, possibly(fn, NA))`
Named iteration	`imap(list, \(val, name) ...)`

FAQ

When should I use map() vs a for loop?

Use map() when each iteration is independent and returns a value. Use a for loop when iterations depend on previous results or need break/next control flow.

How does map() compare to lapply()?

They do the same thing. map() adds typed variants (map_dbl, map_chr), formula shorthand (~ .x + 1), and element extraction by name (map(list, "field")). Use lapply() for zero dependencies, map() for everything else.

What's Next?

dplyr group_by & summarise — the parent tutorial
purrr map Variants — complete map() reference
R Joins — combine results from mapped operations

r-statistics.co by Selva Prabhakaran