Intermediate programming with R

Testing with testit

Using assertion statements is a good first step to writing more reliable code. Going the next step, we can pass inputs to a function and confirm that the result is what we expect. Tests that check a function works properly are called unit tests (because each function is a unit of the overall code we are writing). Writing tests gives us confidence that our code works in different situations, serves as explicit documentation of how a function is supposed to work, and alerts us to any changes due to software updates.

In this lesson we will use the simple testing framework in the package testit. There are other more elaborate testing frameworks such as RUnit and testthat if you need more complicated testing in the future. Also, as a caveat, R testing frameworks work best in the context of an R package. They will be less flexible in our context testing functions that are not part of an R package.

Using the testit package

Let’s first load the package.

library("testit")

In the last lesson, we wrote assertion statements using stopifnot. However, the error messages generated by stopifnot are cryptic unless we are familiar with the intricacies of a specific function. This can be difficult when writing lots of code or returning to code written long ago. The main function of the testit package is assert, which allows us to include a message that is printed if the assertion fails. This makes it easier to interpret what went wrong.

assert("one equals one", 1 == 1)
assert("two plus two equals five", 2 + 2 == 5)

assertion failed: two plus two equals five

Error: 2 + 2 == 5 is not TRUE

And similar to stopifnot, we can list multiple statements to check.

assert("these statements are TRUE", 1 == 1, 2 == 2, 3 == 3)

And if any of them fails, it throws an error.

assert("these statements are TRUE", 0 == 1, 2 == 2, 3 == 3)

assertion failed: these statements are TRUE

Error: 0 == 1 is not TRUE

Furthermore, the package has functions to that detect if an error or warning was produced. From their documentation:

# No warning
has_warning(1 + 1)

[1] FALSE

# Issues warning because vectors have different lengths
has_warning(1:2 + 1:3)

[1] TRUE

# No error
has_error(2 - 3)

[1] FALSE

# Throws error because cannot add numeric and character vectors
has_error(1 + "a")

[1] TRUE

We can combine these with assert to confirm that our functions throw errors or warnings when given certain inputs.

assert("Throws error", has_error(1 + "a"))

Writing unit tests for a function

In the challenge for the last lesson we added assertion statements using stopifnot to the function calc_sum_stat. Also we fixed the problem with NAs by adding the argument na.rm = TRUE to mean. The result should have looked something like the following:

calc_sum_stat <- function(df, cols) {
  stopifnot(dim(df) > 0,
            is.character(cols),
            cols %in% colnames(df))
  if (length(cols) == 1) {
    warning("Only one column specified. Calculating the mean will not change anything.")
  }
  df_sub <- df[, cols, drop = FALSE]
  stopifnot(is.data.frame(df_sub))
  sum_stat <- apply(df_sub, 1, mean, na.rm = TRUE)
  stopifnot(!is.na(sum_stat))
  return(sum_stat)
}

We checked that our function worked properly by testing some different inputs. Let’s convert these into formal units tests that we can automatically run to test our function.

When passing an empty data frame, we expect it to throw an error.

# Empty data frame
sum_stat <- calc_sum_stat(data.frame(), c("wosCountThru2010", "f1000Factor"))

Error: dim(df) > 0 are not all TRUE

Which we can convert to a test:

assert("Empty data frame throws error",
       has_error(calc_sum_stat(data.frame(), c("wosCountThru2010",
                                               "f1000Factor"))))

We also expect an error when passing a non-character vector for the argument cols.

# Non-character cols
sum_stat <- calc_sum_stat(counts_raw, 1:3)

Error: is.character(cols) is not TRUE

And the test looks like:

assert("Non-character vector input for columns throws error",
       has_error(calc_sum_stat(counts_raw, 1:3)))

Testing non-existent column names:

# Bad column names
sum_stat <- calc_sum_stat(counts_raw, c("a", "b"))

Error: cols %in% colnames(df) are not all TRUE

Converted to a test:

assert("Column names not in data frame throws error",
       has_error(calc_sum_stat(counts_raw, c("a", "b"))))

Issue a warning if only one column is given to the function.

# Issue warning since only one column
sum_stat <- calc_sum_stat(counts_raw, "mendeleyReadersCount")

Warning in calc_sum_stat(counts_raw, "mendeleyReadersCount"): Only one
column specified. Calculating the mean will not change anything.

And the test:

assert("Selecting only one column issues warning",
       has_warning(calc_sum_stat(counts_raw, "mendeleyReadersCount")))

Lastly, we passed a column that contains NAs. We fixed this so that mean ignores NAs and returns a numeric answer.

# NA output
sum_stat <- calc_sum_stat(counts_raw, c("wosCountThru2010", "facebookLikeCount"))
anyNA(sum_stat)

[1] FALSE

We can also test this:

assert("NA input does not result in NA output",
       !anyNA(calc_sum_stat(counts_raw, c("wosCountThru2010",
                                          "facebookLikeCount"))))

We have now built an entire test suite that we can automatically run whenever we update the function or update the version of R. This way we always know that the function still works as we initially planned.

assert("Empty data frame throws error",
       has_error(calc_sum_stat(data.frame(), c("wosCountThru2010",
                                               "f1000Factor"))))
assert("Non-character vector input for columns throws error",
       has_error(calc_sum_stat(counts_raw, 1:3)))
assert("Column names not in data frame throws error",
       has_error(calc_sum_stat(counts_raw, c("a", "b"))))
assert("Selecting only one column issues warning",
       has_warning(calc_sum_stat(counts_raw, "mendeleyReadersCount")))
assert("NA input does not result in NA output",
       !anyNA(calc_sum_stat(counts_raw, c("wosCountThru2010",
                                          "facebookLikeCount"))))

Challenge

my_mean <- function(x) {
  result <- sum(x) / length(x)
  return(result)
}