Pipelines with targets
RaukR 2026 • Data Science With R
18-Aug-2026
What is targets?
targets is a pipeline toolkit for R that:
- Declares steps of your analysis as targets
- Tracks which steps of your analysis are up to date
- Skips steps whose inputs have not changed
- Makes the dependency graph of your analysis explicit
- Provides reproducibility by design
The problem targets solves
- You have a 3-hour data processing step
- You change a plotting function
- Do you need to re-run data processing? No — but how do you know?
- With a flat script, you re-run everything
- With targets, only the changed and downstream steps re-run
Exploring the pipeline
tar_manifest()Structure & dependencies
# A tibble: 4 × 2
name command
<chr> <chr>
1 data_raw "datasets::mtcars"
2 data_clean "transform(data_raw, am = factor(am, labels = c(\"automatic\", …
3 model "stats::lm(mpg ~ wt + am, data = data_clean)"
4 model_summary "summary(model)" tar_outdated()targets status
[1] "data_clean" "model_summary" "data_raw" "model" tar_meta()Metadata about targets, status, and dependencies
tar_visnetwork()Visualize dependency graph
Opens an interactive graph in the viewer:
| Colour | Meaning |
|---|---|
| Grey | Outdated: will run |
| Green | Up to date: will be skipped |
| Blue | Functions / objects |
| Red | Error |
tar_glimpse()Visualize simpler dependency graph
Running the pipeline
Run specific targets
tar_make(names = "data_clean")
+ data_raw dispatched
✔ data_raw completed [0ms, 1.23 kB]
+ data_clean dispatched
✔ data_clean completed [0ms, 1.26 kB]
✔ ended pipeline [209ms, 2 completed, 0 skipped]
Run all targets
tar_make()
+ model dispatched
✔ model completed [3ms, 3.42 kB]
+ model_summary dispatched
✔ model_summary completed [1ms, 2.33 kB]
✔ ended pipeline [193ms, 2 completed, 2 skipped]
Simplifying the pipeline
Normal pipeline
Place custom functions in R/
And source them at the top of _targets.R:
Working with files
Normal pipeline
Create a function to save the plot and return the file path:
Extending targets with tarchetypes
tarchetypes provides high-level target archetypes that reduce boilerplate in targets pipelines.
A collection of factory functions that create common target patterns — especially for:
- Rendering Quarto / R Markdown documents
- Iterating over combinations of parameters
- Time-based cues and polling
- Grouping dynamic branches
Static branching
When you know every case before the pipeline runs.
library(targets)
library(tarchetypes)
species_names <- c(
"setosa",
"versicolor",
"virginica"
)
list(
tar_target(data, datasets::iris),
tar_map(
values = list(
species_name = species_names
),
names = species_name,
tar_target(
data_summary,
dplyr::summarise(
dplyr::filter(data, Species == species_name),
species = dplyr::first(Species),
observations = dplyr::n(),
mean_sepal_length = mean(Sepal.Length)
)
)
)
)
Dynamic branching
When the number of cases is unknown until the pipeline runs.
library(targets)
list(
tar_target(data, datasets::iris),
tar_target(
species_names,
as.list(sort(unique(data$Species))),
iteration = "list"
),
tar_target(
data_summary,
dplyr::summarise(
dplyr::filter(data, Species == species_names),
species = dplyr::first(Species),
observations = dplyr::n(),
mean_sepal_length = mean(Sepal.Length)
),
pattern = map(species_names),
iteration = "list"
),
tar_target(
data_summary_combined,
dplyr::bind_rows(data_summary)
)
)
Branching with groups
Split a data frame into groups and process each group as a branch:
library(targets)
library(tarchetypes)
list(
tar_group_by(
split_data,
datasets::iris,
Species
),
tar_target(
data_summary,
dplyr::summarise(
split_data,
species = dplyr::first(Species),
observations = dplyr::n(),
mean_sepal_length = mean(Sepal.Length)
),
pattern = map(split_data),
iteration = "list"
),
tar_target(
data_summary_combined,
dplyr::bind_rows(data_summary)
)
)
A complete pipeline
library(targets)
library(tarchetypes)
library(crew)
fn_clean <- function(data) {
dplyr::filter(data,
!is.na(flipper_length_mm), !is.na(body_mass_g)
)
}
fn_summarize <- function(data) {
dplyr::summarise(
data,
penguins = dplyr::n(),
mean_flipper_mm = mean(flipper_length_mm),
mean_body_mass_g = mean(body_mass_g)
)
}
fn_plot <- function(data) {
path <- "output/penguin-flipper-mass.png"
dir.create(dirname(path), showWarnings = FALSE)
plot <- ggplot2::ggplot(
data,
ggplot2::aes(
flipper_length_mm,
body_mass_g,
colour = species
)
) +
ggplot2::geom_point() +
ggplot2::labs(
x = "Flipper length (mm)",
y = "Body mass (g)"
)
ggplot2::ggsave(
path, plot, width = 7, height = 5
)
path
}
tar_option_set(
packages = c("dplyr", "ggplot2", "palmerpenguins"),
format = "rds",
seed = 42,
controller = crew::crew_controller_local(workers = 2)
)
list(
tar_target(data_raw, palmerpenguins::penguins),
tar_target(data_clean, fn_clean(data_raw)),
# Run one summary branch for each species.
tar_group_by(data_grouped, data_clean, species),
tar_target(
species_summary,
fn_summarize(data_grouped),
pattern = map(data_grouped),
iteration = "list"
),
tar_target(
data_combined,
dplyr::bind_rows(species_summary)
),
tar_target(
plot_file,
fn_plot(data_clean),
format = "file"
)
)
