5 Functions

targets expects users to adopt a function-oriented style of programming. User-defined R functions are essential to express the complexities of data generation, analysis, and reporting. This chapter explains what makes functions useful and how to leverage them in your pipelines. It is based on the example from the walkthrough chapter.

5.1 Problems with script-based workflows

Traditional data analysis projects consist of imperative scripts, often with with numeric prefixes.

01-data.R
02-model.R
03-plot.R

To run the project, the user runs each of the scripts in order.

source("01-data.R")
source("02-model.R")
source("03-plot.R")

Each script executes a different part of the workflow.

# 01-data.R
library(tidyverse)
data <- "data.csv" %>%
  read_csv(col_types = cols()) %>%
  filter(!is.na(Ozone))
write_csv(data, "data.rds")

# 02-model.R
library(biglm)
library(tidyverse)
data <- read_rds("data.rds", col_types = cols())
model <- lm(Ozone ~ Temp, data) %>%
  coefficients()
saveRDS(model, "model.rds")

# 03-plot.R
library(tidyverse)
model <- readRDS("model.rds")
data <- readRDS("data.rds")
plot <- ggplot(data) +
  geom_point(aes(x = Temp, y = Ozone)) +
  geom_abline(intercept = model[1], slope = model[2]) +
  theme_gray(24)
ggsave("plot.png", plot)

Although this approach may feel convenient at first, it scales poorly for medium-sized workflows. These imperative scripts are monolithic, and they grow too large and complicated to understand or maintain.

5.2 Functions

Functions are the building blocks of most computer code. They make code easier to think about, and they break down complicated ideas into small manageable pieces. Out of context, you can develop and test a function in isolation without mentally juggling the rest of the project. In the context of the whole workflow, functions are convenient shorthand to make your work easier to read.

In addition, functions are a nice mental model to express data science. A data analysis workflow is a sequence of transformations: datasets map to analyses, and analyses map to summaries. In fact, a function for data science typically falls into one of three categories:

Process a dataset.
Analyze a dataset.
Summarize an analysis.

The example from the walkthrough chapter is a simple instance of this structure.

5.3 Writing functions

Let us begin with our imperative code for data processing. Every time you look at it, you need to read it carefully and relearn what it does. And test it, you need to copy the entire block into the R console.

data <- "data.csv" %>%
  read_csv(col_types = cols()) %>%
  filter(!is.na(Ozone))

It is better to encapsulate this code in a function.

get_data <- function(file) {
  read_csv(file, col_types = cols()) %>%
    as_tibble() %>%
    filter(!is.na(Ozone))
}

Now, instead of invoking a whole block of text, all you need to do is type a small reusable command. The function name speaks for itself, so you can recall what it does without having to mentally process all the details again.

get_data("data.csv")

As with the data, we can write a function to fit a model,

fit_model <- function(data) {
  lm(Ozone ~ Temp, data) %>%
    coefficients()
}

and another function to plot the model against the data.

plot_model <- function(model, data) {
  ggplot(data) +
    geom_point(aes(x = Temp, y = Ozone)) +
    geom_abline(intercept = model[1], slope = model[2]) +
    theme_gray(24)
}

5.4 Functions in pipelines

Without those functions, our pipeline in the walkthrough chapter would look long, complicated, and difficult to digest.

# _targets.R
library(targets)
library(tarchetypes)
tar_source()
tar_option_set(packages = c("tibble", "readr", "dplyr", "ggplot2"))
list(
  tar_target(file, "data.csv", format = "file"),
  tar_target(
    data,
    read_csv(file, col_types = cols()) %>%
      filter(!is.na(Ozone))
  ),
  tar_target(
    model,
    lm(Ozone ~ Temp, data) %>%
      coefficients()
  ),
  tar_target(
    plot,
    ggplot(data) +
      geom_point(aes(x = Temp, y = Ozone)) +
      geom_abline(intercept = model[1], slope = model[2]) +
      theme_gray(24)
  )
)

But if we write our functions in R/functions.R and source() them into the target script file (default: _targets.R) the pipeline becomes much easier to read. We can even condense out raw_data and data targets together without creating a large command.

# _targets.R
library(targets)
library(tarchetypes)
tar_source()
tar_option_set(packages = c("tibble", "readr", "dplyr", "ggplot2"))
list(
  tar_target(file, "data.csv", format = "file"),
  tar_target(data, get_data(file)),
  tar_target(model, fit_model(data)),
  tar_target(plot, plot_model(model, data))
)

5.5 Tracking changes

To help figure out which targets to rerun and which ones to skip, the targets package tracks changes to the functions you define. To track changes to a function, targets computes a hash. This hash fingerprints the deparsed function (body and arguments) together with the hashes of all global functions and objects called that function. So if the function’s body, arguments, or dependencies change nontrivially, that change will be detected.

This hashing system is not perfect. For example, functions created by Rcpp::cppFunction() do not show the state of the underlying C++ code. As a workaround, you can use a wrapper that inserts the C++ code into the R function body so targets can track it for meaningful changes.

cpp_function <- function(code) {
  out <- Rcpp::cppFunction(code)
  body(out) <- rlang::call2("{", code, body(out))
  out
}

your_function <- cpp_function(
  "int your_function(int x, int y, int z) {
     int sum = x + y + z;
     return sum;
   }"
)

Functions produced by Vectorize() and purrr::safely() suffer similar issues because the actual function code is in the closure of the function instead of the body. In addition, functions from packages are not automatically tracked, and extra steps documented in the packages chapter are required to enable this.

It is impossible to eliminate every edge case, so before running the pipeline, please run the dependency graph and other utilities to check your understanding of the state of project.