caugi

In caugi: Causal Graph Interface

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

# load the package
library(caugi)

In this vignette, we will walk you through how to create a caugi, query it, modify it, and then compare it to other caugis.

The caugi object

You can create a caugi graph object using the caugi() function along with infix operators to define edges. Let's create a directed acyclic graph (DAG) with 5 nodes and 5 edges.

cg <- caugi(
  A %-->% B %-->% C + D,
  A %-->% C,
  class = "DAG"
)
cg

You might scratch your head a bit, when looking at the call above. To clarify, A %-->% B creates the edge --> from A to B. The syntax B %-->% C + D is equivalent to B %-->% C and B %-->% D. Notice that the graph prints two tibbles. The first is equivalent to cg@nodes and the second cg@edges. Besides that, the caugi holds other properties. Let's check the other properties.

Properties

session

cg@session

This is the session pointer to the Rust graph object that caugi utilizes for performance. The session manages lazy compilation and caching of graph operations internally.

simple

cg@simple

This indicates whether the graph is simple or not. Let's try to create a non-simple graph:

caugi(A %-->% B, B %-->% A)

This cannot be done unless you initialize the graph with simple = FALSE. Note that, currently, all of the supported graph classes only support simple = TRUE unless the class is UNKNOWN.

graph_class

cg@graph_class

This is the graph's class. As you can see here, it is a DAG.

Querying the caugi

We can query the caugi graph object with the built-in queries provided in the package. Let's try to find the descendants of all the parents of the node C:

descendants(cg, parents(cg, "C"))

First note that the output is a list of named character vectors. How come? Since the parents of C is c(A, B):

parents(cg, "C")

So for each parent of C we have a named vector in the list that represents the descendants of that parent node.

Modifying the caugi

Let's try to modify the graph from before, so we get a new DAG.

cg_modified <- cg |>
  remove_edges(A %-->% B, B %-->% C + D) |>
  add_edges(B %-->% A, D %-->% C)
cg_modified

Would you like to add nodes? Then use add_nodes().

Graph metrics

Now that we have two different graphs, we can use different metrics to measure the difference between the two graphs. Here, we use the adjustment identification distance (AID) and the structural Hamming distance (SHD):

aid(cg, cg_modified)
shd(cg, cg_modified, normalized = TRUE)

There you go!

You have now created a graph, inspected it, modified it, and measured the difference between the two graphs -- both structurally and interventionally.

For further reading, we recommend the vignettes vignette("package_use") and performance article on our website to see how to use caugi in your own packages, and to see how caugi performs compared to other graph packages in R. For the interested reader, we also recommend vignette("motivation"), which goes deeper into the motivation behind caugi and what we aspire to do with caugi.

Appendix

Advanced properties

name_index_map

session

cg@session

This is the Rust GraphSession pointer that holds the canonical graph state. The session handles lazy compilation and caching of graph operations internally. You should avoid accessing or modifying the session directly. Instead, use the provided accessor properties (@nodes, @edges, @simple, @graph_class) and verb functions (add_edges(), remove_nodes(), etc.).

caugi documentation built on Oct. 28, 2026, 1:07 a.m.