# Lecture 6

## Welcome!

• Welcome back to CS50’s Introduction to Programming with R!
• Today, we will be learning about testing programs. We will see how our programs can go wrong, how we can handle things when they do, and how to methodically test our programs to ensure they behave as we expect!

## Exceptions

• Consider the following program that calculates an average:

# Define function to calculate average value in a vector

average <- function(x) {
sum(x) / length(x)
}


Notice how this program attempts to take as its input a vector of numbers and output the average.

• You can imagine how your user may accidentally pass characters instead of numbers, resulting in our average function outputting an error.
• These errors are called exceptions. Could there be a way by which we could check for potential such exceptions? Consider the following update to average:

# Handle non-numeric input

average <- function(x) {
if (!is.numeric(x)) {
return(NA)
}
sum(x) / length(x)
}


Notice how a conditional, an if statement, checks to see if the vector x is not full of numbers. By convention in the R world, returning a value NA is appropriate in such an instance.

## message

• While this allows our program to run silently, we may wish to let the user know that an exception has occurred. One way to alert the user is via the message function:

# Message about returning NA

average <- function(x) {
if (!is.numeric(x)) {
message("x must be a numeric vector. Returning NA instead.")
return(NA)
}
sum(x) / length(x)
}


Notice how a message is sent to the user about why the program is returning NA instead.

• Traditionally, message is intended for when something has not gone wrong: message is purely for informational purposes. Thus, we can escalate the importance of this information through a warning.

## warning

• We can escalate the importance of our message to a warning as follows:

# Warn about returning NA

average <- function(x) {
if (!is.numeric(x)) {
warning("x must be a numeric vector. Returning NA instead.")
return(NA)
}
sum(x) / length(x)
}


Notice how the output is now a warning message.

• A warning doesn’t stop a program altogether, but it does let a programmer know that something has gone wrong.

## stop

• You can imagine situations where you don’t simply want to warn the user; You may want to completely stop the function. Consider the following:

# Stop instead of warn

average <- function(x) {
if (!is.numeric(x)) {
stop("x must be a numeric vector.")
}
sum(x) / length(x)
}


Notice how stop tells the user we cannot proceed given the input they have provided us.

• It is also possible to combine both possibilities. For example, the following code looks at both situations where x contains non-numeric elements. Similarly, this code accommodates situations where there are NA values:

# Handle NA values

average <- function(x) {
if (!is.numeric(x)) {
stop("x must be a numeric vector.")
}
if (any(is.na(x))) {
warning("x contains one or more NA values.")
return(NA)
}
sum(x) / length(x)
}


Notice how two if statements are provided.

## Unit Tests

• Unit tests are used to test our functions and programs.
• Consider the following test function for average in a separate file:

# Write test function

source("average6.R")

test_average <- function() {
if (average(c(1, 2, 3)) == 2) {
cat("average passed test :)\n")
} else {
cat("average failed test :(\n")
}
}

test_average()


Notice that this function provides a test case where the numbers 1, 2, and 3 are provided to the average function. Then, some feedback is provided. Notice how, in the first line, source ensures this test file has access to the average function.

• It would also be wise to test negative numbers:

# Add test cases

source("average6.R")

test_average <- function() {
if (average(c(1, 2, 3)) == 2) {
cat("average passed test :)\n")
} else {
cat("average failed test :(\n")
}

if (average(c(-1, -2, -3)) == -2) {
cat("average passed test :)\n")
} else {
cat("average failed test :(\n")
}

if (average(c(-1, 0, 1)) == 0) {
cat("average passed test :)\n")
} else {
cat("average failed test :(\n")
}
}

test_average()


Notice how additional tests are provided for positive and negative numbers and zero.

• We have already written 21 lines of code! Thankfully, programmers have already created various test packages or libraries that can be used to test our code.

## testthat

• testthat is a package for testing R code. It can be loaded by typing library(testthat) into your console.
• testthat includes a function called test_that can be used to test our function:

# Test warning about NA values

source("average6.R")

test_that("average calculates mean", {
expect_equal(average(c(1, 2, 3)), 2)
expect_equal(average(c(-1, -2, -3)), -2)
expect_equal(average(c(-1, 0, 1)), 0)
expect_equal(average(c(-2, -1, 1, 2)), 0)
})

test_that("average warns about NAs in input", {
expect_warning(average(c(1, NA, 3)))
expect_warning(average(c(NA, NA, NA)))
})


Notice how the test_that function can be told to expect that that the average of various numbers will equal a certain value, thanks to expect_equal. Similarly, we can provide the test_that function instructions to expect_warning when the average calculation includes NA values. Further, notice how the test is divided into various sections. One section tests the calculation of the mean, while another tests the warnings.

• Running the above test, we discover that the order of our if statements in our average function may be out of order:

# Fix ordering of error handling

average <- function(x) {
if (any(is.na(x))) {
warning("x contains one or more NA values.")
return(NA)
}
if (!is.numeric(x)) {
stop("x must be a numeric vector.")
}
sum(x) / length(x)
}


Notice how the order of the conventional statements is altered.

• We should still test that average returns NA when given an NA value in its input, not just that average raises a warning!

# Test NA return values

source("average7.R")

test_that("average calculates mean", {
expect_equal(average(c(1, 2, 3)), 2)
expect_equal(average(c(-1, -2, -3)), -2)
expect_equal(average(c(-1, 0, 1)), 0)
expect_equal(average(c(-2, -1, 1, 2)), 0)
})

test_that("average returns NA with NAs in input", {
expect_equal(suppressWarnings(average(c(1, NA, 3))), NA)
expect_equal(suppressWarnings(average(c(NA, NA, NA))), NA)
})

test_that("average warns about NAs in input", {
expect_warning(average(c(1, NA, 3)))
expect_warning(average(c(NA, NA, NA)))
})


Notice how we have two separate tests that pass NA values as input to average. One tests for the right return value, while the other tests for a warning to be raised.

• test_that has other functions that can assist us in testing, including expect_error and expect_no_error.
• Using expect_error we can modify our code as follows:

# Test stop if argument is non-numeric

source("average7.R")

test_that("average calculates mean", {
expect_equal(average(c(1, 2, 3)), 2)
expect_equal(average(c(-1, -2, -3)), -2)
expect_equal(average(c(-1, 0, 1)), 0)
expect_equal(average(c(-2, -1, 1, 2)), 0)
})

test_that("average returns NA with NAs in input", {
expect_equal(suppressWarnings(average(c(1, NA, 3))), NA)
expect_equal(suppressWarnings(average(c(NA, NA, NA))), NA)
})

test_that("average warns about NAs in input", {
expect_warning(average(c(1, NA, 3)))
expect_warning(average(c(NA, NA, NA)))
})

test_that("average stops if x is non-numeric", {
expect_error(average(c("quack!")))
expect_error(average(c("1", "2", "3")))
})


Notice how this code expects an error when the input is “quack!” or when characters are provided instead of numbers.

## Testing Floating-Point Values

• We may wish to provide floating-point values (i.e., decimal values) as input to average:

  # Test doubles

source("average7.R")

test_that("average calculates mean", {
expect_equal(average(c(1, 2, 3)), 2)
expect_equal(average(c(-1, -2, -3)), -2)
expect_equal(average(c(-1, 0, 1)), 0)
expect_equal(average(c(-2, -1, 1, 2)), 0)
expect_equal(average(c(0.1, 0.5)), 0.3)
})

test_that("average returns NA with NAs in input", {
expect_equal(suppressWarnings(average(c(1, NA, 3))), NA)
expect_equal(suppressWarnings(average(c(NA, NA, NA))), NA)
})

test_that("average warns about NAs in input", {
expect_warning(average(c(1, NA, 3)))
expect_warning(average(c(NA, NA, NA)))
})

test_that("average stops if x is non-numeric", {
expect_error(average(c("quack!")))
expect_error(average(c("1", "2", "3")))
})


Notice how a test is added at the end of the first set of tests.

## Tolerance

• Floating-point values are unique, in that they are subject to floating-point imprecision.
• Let’s understand floating-point imprecision by example:

# Demonstrates floating-point imprecision

print(0.3)
print(0.3, digits = 17)


Notice how we see that 0.3 is not represented as precisely 0.3 in R. This is a common phenomenon across programming languages, given that there are an infinite number of floating-point values and a finite number of bits to represent them.

• Because of floating-point imprecision, tests of equality involving floating-point values need to allow for some tolerance. Tolerance refers to a range of values, above or below the expected value, that will be considered—for the sake of the test—to be equal to the expected value. Tolerance is often specified in absolute terms, such as ± .000001.
• The expect_equal function already provides a level of tolerance that is generally acceptable for most use cases. This default can be changed with the tolerance argument.
• You and your team should decide upon what level of precision is expected in your calculations.

## Test-Driven Development

• One philosophy of development is called test-driven development. In this mindset, the belief is that it is best to create a test first before even writing the source code that will be tested. Consider the following test:

# Test greet

source("greet1.R")

test_that("greet says hello to a user", {
expect_equal(greet("Carter"), "hello, Carter")
})


Notice how you can imagine that a greet function should greet a user provided as input.

• Looking at the test, we could create code that responds to the test:

  # Greets a user

greet <- function(to) {
return(paste("hello,", to))
}


Notice how this code says hello to a user by name or to the world as default.

• In test-driven development, writing tests allows programmers to know what functionality they should implement. The benefit is that this functionality is then immediately testable. Further modifications should always pass the tests one has already written.

## Behavior-Driven Development

• Behavior-driven development is similar in spirit to test-driven development, with a greater focus on the behavior of a function in context. In behavior-driven development, one might describe what we want the function to do by explicitly naming what it should do.
• testthat comes with two functions to implement behavior-driven development, describe and it:

# Describe greet

source("greet2.R")

describe("greet()", {
it("can say hello to a user", {
name <- "Carter"
expect_equal(greet(name), "hello, Carter")
})
it("can say hello to the world", {
expect_equal(greet(), "hello, world")
})
})


Notice how describe includes several code-based descriptions of what it (the function!) should be able to do.

## Test Coverage

• As you go off and write tests for your code, consider how comprehensive these tests are. Define what is critical for your code to accomplish and create tests that exemplify those critical tasks.

## Summing Up

In this lesson, you learned how to visualize data in R. Specifically, you learned about:

• Exceptions
• message
• warning
• stop
• Unit Tests
• testthat
• Testing Floating-Point Values
• Tolerance
• Test-Driven Development
• Behavior-Driven Development
• Test Coverage

See you next time when we can package our code and share it with the world.