purrr Exercises: 10 Functional Programming Practice Problems

Reading about map() is fast. Writing map_dfr(), map2(), pmap(), safely(), and the \(x) lambda shortcut fluently takes practice, and that's what these ten runnable problems give you.

How do you apply a function to every column of a data frame?

The typed map_*() family is the workhorse of purrr. You hand it a list (or a data frame, which is a list of columns) and a function, and it applies the function to every element, returning a guaranteed-type atomic vector instead of a list. Reach for map_dbl() when the answer is numeric, map_chr() for text, map_int() for counts, and map_lgl() for yes/no. The first three warm-ups all fit this one idea, starting with the simplest: one number per column.

That one line replaces sapply(mtcars, mean) and a ritual as.numeric() call. map_dbl() inspects each column, computes mean(), and guarantees the output is a named numeric vector, so downstream code that does round(..., 2) or sort() just works. If any column returned something non-numeric (say, you accidentally passed a character column), you'd get a clear error instead of a silent list-of-mixed-types.

How do you iterate over logic, row-bind results, and walk parallel inputs?

The next tier of purrr covers four patterns that come up constantly in real analysis code. map_lgl() answers TRUE/FALSE questions about every element. map_dfr() stacks data frames returned by each iteration into one tidy frame, the purrr answer to do.call(rbind, ...). map2() walks two inputs in parallel, useful for pairwise arithmetic. And pmap() generalises to three or more inputs by iterating over the rows of a data frame or list of vectors.

Every column gets the same check, all(x > 0), and the result is a named logical vector you can immediately plug into mtcars[, pos_cols] to subset. vs and am come back FALSE because both contain zeros. That's the pattern: map_lgl() + a predicate gives you a keep/drop mask in one line, no loop required.

How do you handle list-columns, errors, and lambda shortcuts?

The hardest purrr patterns show up when iteration meets three real-world wrinkles: grouped data that you want to keep in one tidy frame (list-columns via nest() + map()), iterations that might fail on some inputs (safely()), and short one-off functions that would clutter your code if you had to name them (\(x) lambdas). The three closing problems cover exactly these.

The pipeline reads left-to-right: split iris into one data frame per species, then map() over that list applying the lambda. The result is a named list with one element per group. You'd use this shape when the per-group answer is richer than a single number, for example, a fitted model or a multi-row tidy table.

Practice Exercises

Three capstone problems that combine multiple purrr concepts from above. These are harder than the warm-ups and meant to consolidate what you've just learned. Every solution uses variables prefixed with my_ so they don't collide with the warm-up state.

Exercise 1: Grouped summary table in one pipeline

For mtcars, nest by cyl and compute three per-group aggregates in one mutate() + map_dbl() sweep: mean mpg, max hp, and row count. Return a single tidy frame with one row per cylinder group and columns cyl, mean_mpg, max_hp, n_rows. Save the result as my_summary.

Exercise 2: Safely parse many character vectors

You receive a list of three character vectors. Two are cleanly convertible to numeric, one contains a non-numeric token. Wrap as.numeric in safely(), map it over the list, and build a tidy data frame called my_parse with two columns: input_id (which vector) and status ("ok" if the conversion produced no warnings, "failed" otherwise). Hint: use purrr::quietly() if you want warnings captured alongside errors.

Exercise 3: All pairwise correlations with pmap_dbl

You have a tibble with three numeric columns x, y, w. Compute the three pairwise Pearson correlations, cor(x,y), cor(x,w), cor(y,w), in a single pmap_dbl() call over a helper tibble that lists the column pairs. Save the result as my_corrs (named numeric vector with entries x_y, x_w, y_w).

Complete Example

Here's what a real-world purrr pipeline looks like: per-cylinder multiple regression of mpg ~ wt + hp, extracting both the tidy coefficient tables and the R² values in one flow. Every step uses a concept from the warm-ups above.

Three things happen in one pipeline. First, nest() collapses each cyl group into a list-column. Second, map() fits a model per group and stashes the fit object in a new column. Third, map_dbl() extracts R² into a plain numeric column while map() pulls the tidy coefficient frame into another list-column that unnest() expands into the final 9-row table. One read-through tells you which coefficients are significant in which groups, all without a single explicit for-loop.

Summary

The ten problems above cover the purrr vocabulary you'll use in 90% of real analysis code. If you can fluently pick the right verb from this table, you're past the "purrr is confusing" phase.

References

1. Wickham, H. & Grolemund, G., R for Data Science, 2nd ed. Chapter 27: Iteration. Link
2. purrr package reference and articles. Link
3. Wickham, H., Advanced R, 2nd ed. Chapter 9: Functionals. Link
4. tidyr nest() documentation. Link
5. Tidyverse blog, purrr 1.0.0 release notes. Link

Continue Learning

• Functional Programming in R, the parent tutorial covering why functional style works so well for data analysis.
• purrr map() Variants, a function-by-function deep dive on every typed map_*() in the package.
• Reduce, Filter, Map in Base R, the same ideas using only base R, useful when you can't add tidyverse as a dependency.