Package 'networkformat'

Convert to tbl_graph

Description

S3 methods for converting tree-based model objects into tbl_graph objects. Each method wraps the corresponding as.igraph method. These methods are registered against the as_tbl_graph generic from tidygraph via delayed S3 registration and are available whenever tidygraph is loaded.

Usage

## S3 method for class 'tree'
as_tbl_graph(x, ...)

## S3 method for class 'randomForest'
as_tbl_graph(x, treenum = NULL, ...)

## S3 method for class 'rpart'
as_tbl_graph(x, ...)

## S3 method for class 'xgb.Booster'
as_tbl_graph(x, treenum = NULL, ...)

## S3 method for class 'gbm'
as_tbl_graph(x, treenum = NULL, ...)

Arguments

x	An object to convert (currently tree, randomForest, rpart, xgb.Booster, or gbm).
...	Additional arguments passed to as.igraph.
treenum	Integer vector of tree numbers to extract. Default NULL returns all trees combined into one graph with disconnected components. Pass a single integer (e.g. 1) to extract one tree.

Value

A tbl_graph object.

Examples

if (requireNamespace("rpart", quietly = TRUE) &&
    requireNamespace("tidygraph", quietly = TRUE) &&
    requireNamespace("igraph", quietly = TRUE)) {
  fit <- rpart::rpart(Sepal.Length ~ ., data = iris)
  tg <- tidygraph::as_tbl_graph(fit)
  tg
}

---

Convert to igraph

Description

S3 methods for converting tree-based model objects into igraph graph objects. Each method calls edgelist and nodelist internally and handles column reconciliation so you get a ready-to-use graph. These methods are registered against the as.igraph generic from igraph via delayed S3 registration and are available whenever igraph is loaded.

Usage

## S3 method for class 'tree'
as.igraph(x, ...)

## S3 method for class 'randomForest'
as.igraph(x, treenum = NULL, ...)

## S3 method for class 'rpart'
as.igraph(x, ...)

## S3 method for class 'xgb.Booster'
as.igraph(x, treenum = NULL, ...)

## S3 method for class 'gbm'
as.igraph(x, treenum = NULL, ...)

Arguments

x	An object to convert (tree, randomForest, rpart, xgb.Booster, or gbm).
...	Additional arguments passed to methods.
treenum	Integer vector of tree numbers to extract. Default NULL returns all trees combined into one graph with disconnected components. Pass a single integer (e.g. 1) to extract one tree.

Value

An igraph object. For randomForest with multiple trees, the graph contains disconnected components (one per tree) and a treenum vertex/edge attribute.

Examples

if (requireNamespace("rpart", quietly = TRUE) &&
    requireNamespace("igraph", quietly = TRUE)) {
  fit <- rpart::rpart(Sepal.Length ~ ., data = iris)
  g <- igraph::as.igraph(fit)
  igraph::vcount(g)
  igraph::ecount(g)
}

---

Course prerequisite and crosslisting network

Description

A small dataset of university courses with prerequisite and crosslisting relationships, suitable for demonstrating edgelist() and nodelist().

Usage

courses

Format

A data.frame with 13 rows and 7 columns:

dept

Department code (STAT, MATH, DATA, CS)

course

Course identifier, used as node ID

prereq

Prerequisite course (NA if none)

prereq2

Second prerequisite course (NA if none)

crosslist

Crosslisted equivalent course (NA if none)

credits

Number of credit hours (integer)

level

Course level: 100, 200, or 300 (integer)

Examples

# Prerequisite edgelist (course -> prereq)
edgelist(courses, source_cols = course, target_cols = prereq)

# Node list with course as ID column
nodelist(courses, id_col = course)

---

Extract Edgelist from Various Object Types

Description

Generic function to extract a network edgelist from various object types. Methods exist for atomic vectors (sequential edges), data frames (column-pair edges), lists (recursive parent-child edges), and tree-based model objects (randomForest, tree, rpart, xgb.Booster, gbm). The specific columns returned depend on the input type.

Usage

edgelist(input_object, ...)

Arguments

input_object	An object to extract an edgelist from
...	Additional arguments passed to specific methods

Value

A data.frame representing an edgelist. The specific columns depend on the input object type.

Examples

# Vector --- sequential edges
edgelist(c("A", "B", "C", "D"))

# Data.frame example using bundled dataset
edgelist(courses, source_cols = course, target_cols = prereq)

# RandomForest example
if (requireNamespace("randomForest", quietly = TRUE)) {
  rf <- randomForest::randomForest(Species ~ ., data = iris, ntree = 5)
  edges_rf <- edgelist(rf)
  head(edges_rf)
}

# Tree example
if (requireNamespace("tree", quietly = TRUE)) {
  tr <- tree::tree(Species ~ ., data = iris)
  edges_tr <- edgelist(tr)
  head(edges_tr)
}

---

Extract Edgelist from Data Frame

Description

Converts a data.frame to network edgelist format by specifying which columns represent source and target nodes. This is useful for creating edgelists from tabular data where node relationships are stored in columns.

Usage

## S3 method for class 'data.frame'
edgelist(
  input_object,
  source_cols = 1,
  target_cols = 2,
  attr_cols = NULL,
  na.rm = TRUE,
  symmetric_cols = NULL,
  dedupe = TRUE,
  weights = FALSE,
  ...
)

Arguments

input_object	A data.frame containing network information
source_cols	Column(s) for source nodes (default: 1). Accepts tidyselect expressions: bare names, strings, numeric indices, or helpers like starts_with().
target_cols	Column(s) for target nodes (default: 2). Same syntax as source_cols.
attr_cols	Columns to carry as edge attributes (default: NULL keeps all columns not used as source or target). Pass an empty c() to keep only from, to, and metadata columns. Accepts the same tidyselect syntax as source_cols.
na.rm	Logical; if TRUE (the default), rows where from or to is NA are removed from the result. Set to FALSE to keep all rows including those with NA endpoints.
symmetric_cols	Target column names that represent undirected (symmetric) relationships. When non-NULL, a directed column is added: FALSE for edges from target columns named in symmetric_cols, TRUE otherwise. Accepts the same tidyselect syntax as target_cols; names must be a subset of target_cols.
dedupe	Logical; if TRUE (the default) and symmetric_cols is non-NULL, duplicate undirected edges are removed by keeping only rows where from <= to (lexicographic comparison). Set to FALSE to preserve both directions.
weights	Logical; if TRUE, fully identical rows (same from, to, and all attribute columns) are collapsed and a weight column is added with the count. Rows that share (from, to) but differ in any attribute are kept separate. Applied after NA removal and symmetric deduplication. Defaults to FALSE.
...	Additional arguments (currently unused)

Value

A data.frame with columns:

from

Source node values

to

Target node values

from_col

Name of the original column each source value came from

to_col

Name of the original column each target value came from

directed

(only when symmetric_cols is provided) Logical: FALSE for edges from symmetric target columns, TRUE otherwise

weight

(only when weights = TRUE) Integer count of how many times each unique (from, to) pair occurred

...

Additional attribute columns selected by attr_cols

Examples

# Basic usage --- all non-source/target columns kept by default
edgelist(courses, source_cols = course, target_cols = prereq)

# Multiple target columns --- to_col identifies the relationship type
edgelist(courses, source_cols = course, target_cols = c(prereq, crosslist))

# Keep only the edgelist (no attribute columns)
edgelist(courses, source_cols = course, target_cols = prereq, attr_cols = c())

# Select specific attribute columns
edgelist(courses, source_cols = course, target_cols = prereq,
         attr_cols = c(dept, credits))

# NA removal (default) vs preservation
edgelist(courses, source_cols = course, target_cols = c(prereq, crosslist))
edgelist(courses, source_cols = course, target_cols = c(prereq, crosslist),
         na.rm = FALSE)

# Mark symmetric (undirected) target columns
edgelist(courses, source_cols = course,
         target_cols = c(prereq, crosslist),
         symmetric_cols = crosslist)

---

Extract Edgelist from a Vector or Unsupported Object

Description

When called on an atomic vector (character, numeric, integer, logical, factor), creates a sequential edgelist connecting each element to the next: element i is connected to element i + 1. For all other unsupported types, an informative error is raised.

Usage

## Default S3 method:
edgelist(input_object, weights = FALSE, ...)

Arguments

input_object	An atomic vector, or an object for which no specific edgelist method exists.
weights	Logical; if TRUE, duplicate edges are collapsed and a weight column is added with the count of each unique (from, to) pair. Defaults to FALSE.
...	Additional arguments (currently unused)

Value

A data.frame with columns from and to (and weight when weights = TRUE). For unsupported types, an error is raised.

Examples

# Character vector
edgelist(c("A", "B", "C", "D"))

# Numeric vector
edgelist(1:5)

# With duplicate counting
edgelist(c("A", "B", "A", "B", "C"), weights = TRUE)

# Unsupported type
try(edgelist(as.formula(y ~ x)))

---

Extract Edgelist from GBM Model

Description

Converts a gbm model object into a network edgelist representation using pretty.gbm.tree(). Missing-sentinel nodes (the phantom routing nodes gbm creates for NA handling) are excluded — only real LeftNode/RightNode edges are returned.

Usage

## S3 method for class 'gbm'
edgelist(input_object, treenum = NULL, ...)

Arguments

input_object	A gbm model object from the gbm package
treenum	Integer vector of 1-based tree numbers to extract (default: NULL extracts all trees). For multinomial models, this indexes physical trees (not boosting iterations).
...	Additional arguments (currently unused)

Details

Node IDs are 0-based integers per tree. For models with num.classes > 1 (multinomial), physical trees are stored as n.trees * num.classes entries; treenum indexes into this flat sequence.

Value

A data.frame with the following columns:

from

Parent node ID (0-based integer)

to

Child node ID (0-based integer)

split_var

0-based variable index

split_point

Split threshold for continuous variables, or c.splits index for categorical variables

prediction

Prediction value at the child node

treenum

1-based tree number

split_var_name

Human-readable variable name

Examples

if (requireNamespace("gbm", quietly = TRUE)) {
  set.seed(1)
  fit <- gbm::gbm(mpg ~ ., data = mtcars,
                   distribution = "gaussian", n.trees = 5,
                   interaction.depth = 3, n.minobsinnode = 3)
  el <- edgelist(fit)
  head(el)

  # Single tree
  el1 <- edgelist(fit, treenum = 1)
}

---

Extract Edgelist from a List

Description

Recursively traverses a nested list structure, producing a parent-child edgelist. Each element of the list becomes a node; nested lists create deeper edges. Path-style IDs ensure unique node names even when element names repeat at different levels.

Usage

## S3 method for class 'list'
edgelist(input_object, name_root = "root", max_depth = NULL, ...)

Arguments

input_object	A list (or S3 object falling through to this method).
name_root	Character; label for the root node. Defaults to "root".
max_depth	Integer or NULL; maximum node depth to include (root is depth 0, its children are depth 1). NULL (the default) means unlimited. max_depth = 0 returns an empty edgelist (root only).
...	Additional arguments (currently unused).

Details

When called on an S3 object that has no dedicated edgelist method (e.g. an lm object), the object is treated as a plain list and a diagnostic message is emitted. To force structural decomposition of an object that has its own method (e.g. a tree), use edgelist(unclass(x)).

Value

A data.frame with columns from, to, and depth (integer depth of the child node, root children are depth 1). An empty list returns a zero-row data.frame with the same columns.

Examples

edgelist(list(a = 1, b = list(c = 2, d = 3)))

# Unnamed elements use positional indices
edgelist(list(1, 2, list(3, 4)))

# Limit depth (root = 0, children = 1, ...)
edgelist(list(a = list(b = list(c = 1))), max_depth = 2)

---

Extract Edgelist from RandomForest Model

Description

Converts a randomForest model object into a network edgelist representation by extracting parent-child relationships from one or more trees in the forest. Each edge represents a split in the decision tree, with additional attributes including split variable, split point, and prediction values.

Usage

## S3 method for class 'randomForest'
edgelist(input_object, treenum = NULL, ...)

Arguments

input_object	A randomForest model object from the randomForest package
treenum	Integer vector of tree numbers to extract (default: NULL extracts all trees). Values must be between 1 and input_object$ntree.
...	Additional arguments (currently unused)

Value

A data.frame with the following columns:

from

Parent node index within the tree

to

Child node index (left or right daughter)

split_var

Numeric index of the variable used for splitting

split_point

Threshold value for the split

prediction

Prediction value at the child node

direction

Branch direction: "left" or "right"

treenum

Tree number within the forest

split_var_name

Character vector with human-readable variable names

Examples

if (requireNamespace("randomForest", quietly = TRUE)) {
  set.seed(12)
  rf_model <- randomForest::randomForest(
    Species ~ .,
    data = iris,
    ntree = 5,
    maxnodes = 10
  )

  # Extract all trees
  rf_edges <- edgelist(rf_model)
  head(rf_edges)

  # Extract specific trees
  rf_edges_1 <- edgelist(rf_model, treenum = 1)
  rf_edges_13 <- edgelist(rf_model, treenum = c(1, 3))
}

---

Extract Edgelist from rpart Model

Description

Converts an rpart model object into a network edgelist representation. The rpart package uses binary heap numbering for node IDs (root = 1, left child of k = 2k, right child of k = 2k + 1), so parent-child edges are derived directly from the node IDs in rownames(input_object$frame).

Usage

## S3 method for class 'rpart'
edgelist(input_object, ...)

Arguments

input_object	An rpart model object from the rpart package
...	Additional arguments (currently unused)

Details

Edge labels come from labels(input_object, collapse = TRUE), which correctly handles the ncat sign that controls split direction (left child does not always get "<").

Value

A data.frame with the following columns:

from

Parent node ID (binary heap index)

to

Child node ID (binary heap index)

label

Split condition label from labels()

split_var

Variable name used for the split

split_op

Operator: "<" or ">=" for numeric splits, NA for categorical splits

split_point

Numeric threshold for the split (NA for categorical splits)

Examples

if (requireNamespace("rpart", quietly = TRUE)) {
  fit <- rpart::rpart(Species ~ ., data = iris)
  el <- edgelist(fit)
  head(el)

  # Parsed split components
  el[, c("split_var", "split_op", "split_point")]
}

---

Extract Edgelist from Tree Model

Description

Converts a tree model object (from the tree package) into a network edgelist representation. The tree package uses binary heap numbering for node IDs (root = 1, left child of k = 2k, right child of k = 2k + 1), so parent-child edges are derived directly from the node IDs in rownames(input_object$frame).

Usage

## S3 method for class 'tree'
edgelist(input_object, ...)

Arguments

input_object	A tree model object from the tree package
...	Additional arguments (currently unused)

Value

A data.frame with the following columns:

from

Parent node ID (binary heap index)

to

Child node ID (binary heap index)

label

Split condition label (variable and threshold)

split_var

Variable name used for the split

split_op

Operator: "<" or ">" for numeric splits, NA for categorical splits

split_point

Numeric threshold for the split (NA for categorical splits)

Examples

if (requireNamespace("tree", quietly = TRUE)) {
  # Fit a classification tree
  tree_model <- tree::tree(Species ~ ., data = iris)

  # Extract edgelist
  tree_edges <- edgelist(tree_model)
  head(tree_edges)

  # Parsed split components
  tree_edges[, c("split_var", "split_op", "split_point")]
}

---

Extract Edgelist from XGBoost Model

Description

Converts an xgboost model object into a network edgelist representation using xgb.model.dt.tree(). Each edge connects a split node to one of its two children (yes/no branches). Node IDs are globally unique strings in "Tree-Node" format (e.g., "0-3").

Usage

## S3 method for class 'xgb.Booster'
edgelist(input_object, treenum = NULL, ...)

Arguments

input_object	An xgboost model object (xgb.Booster)
treenum	Integer vector of 1-based tree numbers to extract (default: NULL extracts all trees).
...	Additional arguments (currently unused)

Value

A data.frame with the following columns:

from

Parent node ID string ("Tree-Node" format)

to

Child node ID string

feature

Name of the split variable (or 0-based index string when feature names are absent)

split

Numeric split threshold

quality

Information gain at the split

cover

Number of observations covered

treenum

1-based tree number

Examples

if (requireNamespace("xgboost", quietly = TRUE)) {
  data(agaricus.train, package = "xgboost")
  bst <- xgboost::xgboost(
    x = agaricus.train$data,
    y = factor(agaricus.train$label),
    max_depth = 2, nrounds = 2, nthreads = 1
  )
  el <- edgelist(bst)
  head(el)
}

---

Extract Node List from Various Objects

Description

Generic function to extract a node list (with attributes) from various object types including atomic vectors (unique values with frequencies), data frames, lists (recursive node decomposition), and tree models (randomForest, tree, rpart, xgb.Booster, gbm). The node list provides metadata about each node in the network.

Usage

nodelist(input_object, ...)

Arguments

input_object	An object containing node information (vector, data.frame, tree model, etc.)
...	Additional arguments passed to specific methods

Value

A data.frame where the first column is the node ID and subsequent columns contain node attributes

Examples

# Vector --- unique values with frequency counts
nodelist(c("A", "B", "A", "C"))

# Node list with course as ID (column 2)
nodelist(courses, id_col = 2)

---

Extract Node List from Data Frame

Description

Reorders a data.frame to place the ID column first, creating a proper node list format. All other columns are treated as node attributes.

Usage

## S3 method for class 'data.frame'
nodelist(input_object, id_col = 1, ...)

Arguments

input_object	A data.frame containing node information
id_col	Column for the node ID (default: 1). Accepts tidyselect expressions: a bare name, string, or numeric index.
...	Additional arguments (currently unused)

Value

A data.frame with the ID column first, followed by all attribute columns

Examples

# Default: first column (dept) is ID
nodelist(courses)

# Use bare column name
nodelist(courses, id_col = course)

# Numeric index still works
nodelist(courses, id_col = 2)

---

Extract Node List from a Vector or Unsupported Object

Description

When called on an atomic vector (character, numeric, integer, logical, factor), returns the unique values as a node list with a name column (preserving order of first appearance) and an n column giving the frequency of each value. For all other unsupported types, an informative error is raised.

Usage

## Default S3 method:
nodelist(input_object, ...)

Arguments

input_object	An atomic vector, or an object for which no specific nodelist method exists.
...	Additional arguments (currently unused)

Value

A data.frame with columns name (unique values in order of first appearance) and n (frequency count). For unsupported types, an error is raised.

Examples

# Character vector
nodelist(c("A", "B", "C", "A", "B"))

# Numeric vector
nodelist(c(1, 2, 3, 2, 1))

# Unsupported type
try(nodelist(as.formula(y ~ x)))

---

Extract Node List from GBM Model

Description

Extracts node-level attributes from a gbm model via pretty.gbm.tree(). Missing-sentinel nodes are excluded by keeping only nodes reachable through LeftNode/RightNode edges from the root.

Usage

## S3 method for class 'gbm'
nodelist(input_object, treenum = NULL, ...)

Arguments

input_object	A gbm model object from the gbm package
treenum	Integer vector of 1-based tree numbers to extract (default: NULL extracts all trees).
...	Additional arguments (currently unused)

Value

A data.frame with one row per real node and the following columns:

name

0-based node ID within the tree

is_leaf

Logical: TRUE for terminal nodes

split_var

0-based variable index (NA for leaves)

split_var_name

Variable name (NA for leaves)

split_point

Split threshold (NA for leaves)

prediction

Prediction value at the node

error_reduction

Impurity reduction from the split (0 for leaves); per-node variable importance from ErrorReduction

weight

Number of training observations reaching the node, from Weight

treenum

1-based tree number

label

Display label: "<var>\n< <threshold>" for splits, rounded prediction for leaves

Examples

if (requireNamespace("gbm", quietly = TRUE)) {
  set.seed(1)
  fit <- gbm::gbm(mpg ~ ., data = mtcars,
                   distribution = "gaussian", n.trees = 5,
                   interaction.depth = 3, n.minobsinnode = 3)
  nl <- nodelist(fit)
  head(nl)
}

---

Extract Node List from a List

Description

Recursively traverses a nested list structure, producing one row per node with metadata about each element.

Usage

## S3 method for class 'list'
nodelist(input_object, name_root = "root", max_depth = NULL, ...)

Arguments

input_object	A list (or S3 object falling through to this method).
name_root	Character; label for the root node. Defaults to "root".
max_depth	Integer or NULL; maximum node depth to include (root is depth 0, its children are depth 1). NULL (the default) means unlimited. max_depth = 0 returns the root node only.
...	Additional arguments (currently unused).

Details

When called on an S3 object that has no dedicated nodelist method, the object is treated as a plain list and a diagnostic message is emitted. To force structural decomposition of an object that has its own method (e.g. a tree), use nodelist(unclass(x)).

Value

A data.frame with columns name (path-style ID), depth (integer), type (character class of the element), n_children (integer, 0 for leaves), and label (element name or positional index). An empty list returns a one-row data.frame for the root node only.

Examples

nodelist(list(a = 1, b = list(c = 2, d = 3)))

# Unnamed elements
nodelist(list(1, 2, list(3, 4)))

---

Extract Node List from RandomForest Model

Description

Extracts node-level attributes from every tree in a randomForest model. Node IDs match the from/to indices produced by edgelist.randomForest, so the two outputs can be passed directly to igraph::graph_from_data_frame() (after filtering to a single treenum).

Usage

## S3 method for class 'randomForest'
nodelist(input_object, treenum = NULL, ...)

Arguments

input_object	A randomForest model object from the randomForest package
treenum	Integer vector of tree numbers to extract (default: NULL extracts all trees). Values must be between 1 and input_object$ntree.
...	Additional arguments (currently unused)

Value

A data.frame with one row per node (across all trees) and the following columns:

name

Integer node ID within the tree (matches edgelist from/to)

is_leaf

Logical: TRUE for terminal nodes

split_var

Numeric index of the split variable (NA for leaves)

split_var_name

Name of the split variable (NA for leaves)

split_point

Split threshold (NA for leaves)

prediction

Predicted value (numeric for regression, integer class index for classification)

treenum

Integer identifying which tree the node belongs to

label

Display label: "<var>\n< <threshold>" for internal nodes, predicted class name (classification) or numeric value (regression) for leaves

Examples

if (requireNamespace("randomForest", quietly = TRUE)) {
  set.seed(12)
  rf <- randomForest::randomForest(Species ~ ., data = iris, ntree = 3)
  nodes <- nodelist(rf)
  head(nodes)

  # Extract a single tree
  nodes1 <- nodelist(rf, treenum = 1)
}

---

Extract Node List from rpart Model

Description

Extracts node-level attributes from an rpart model object. Node IDs are the binary heap indices from rownames(input_object$frame) and match the from/to values produced by edgelist.rpart, so the two outputs can be passed directly to igraph::graph_from_data_frame().

Usage

## S3 method for class 'rpart'
nodelist(input_object, ...)

Arguments

input_object	An rpart model object from the rpart package
...	Additional arguments (currently unused)

Value

A data.frame with one row per node and the following columns:

name

Integer node ID (binary heap index, matches edgelist from/to)

var

Split variable name, or "<leaf>" for terminal nodes

n

Number of observations routed to this node

dev

Deviance (impurity) at this node

yval

Predicted value (numeric for regression, character class label for classification)

is_leaf

Logical: TRUE for terminal nodes

depth

Integer tree depth (root = 0)

wt

Weighted observation count

complexity

CP pruning parameter at this node

ncompete

Number of competing splits considered

nsurrogate

Number of surrogate splits used

dev_improvement

Numeric deviance reduction from this node's split (NA for leaves)

n_*

(Classification only) One column per class with the count of observations of that class at the node, named n_<classname>. Class names are sanitized via make.names() and lowercased.

prob_*

(Classification only) One column per class with the class probability at that node, named prob_<classname>. Class names are sanitized via make.names() and lowercased.

nodeprob

(Classification only) Proportion of training data reaching this node

label

Display label: "<var>\nn=<n>" for internal nodes, "<yval>\nn=<n>" for leaves

Examples

if (requireNamespace("rpart", quietly = TRUE)) {
  fit <- rpart::rpart(Species ~ Sepal.Length + Sepal.Width, data = iris)
  nodelist(fit)

  # Labels ready for plotting
  nodelist(fit)$label
}

---

Extract Node List from Tree Model

Description

Extracts node-level attributes from a tree model object. Node IDs are the binary heap indices from rownames(input_object$frame) and match the from/to values produced by edgelist.tree, so the two outputs can be passed directly to igraph::graph_from_data_frame().

Usage

## S3 method for class 'tree'
nodelist(input_object, ...)

Arguments

input_object	A tree model object from the tree package
...	Additional arguments (currently unused)

Value

A data.frame with one row per node and the following columns:

name

Integer node ID (binary heap index, matches edgelist from/to)

var

Split variable name, or "<leaf>" for terminal nodes

n

Number of observations routed to this node

dev

Deviance (impurity) at this node

yval

Predicted value (numeric for regression, character for classification)

is_leaf

Logical: TRUE for terminal nodes

depth

Integer tree depth (root = 0)

dev_improvement

Numeric deviance reduction from this node's split (NA for leaves)

prob_*

(Classification only) One column per class with the class probability at that node, named prob_<classname>. Class names are sanitized via make.names() and lowercased.

label

Display label: "<var>\nn=<n>" for internal nodes, "<yval>\nn=<n>" for leaves

Examples

if (requireNamespace("tree", quietly = TRUE)) {
  tr <- tree::tree(Species ~ Sepal.Length + Sepal.Width, data = iris)
  nodelist(tr)

  # Labels ready for plotting
  nodelist(tr)$label
}

---

Extract Node List from XGBoost Model

Description

Extracts node-level attributes from an xgboost model via xgb.model.dt.tree(). Node IDs are globally unique strings in "Tree-Node" format and match the from/to columns produced by edgelist.xgb.Booster.

Usage

## S3 method for class 'xgb.Booster'
nodelist(input_object, treenum = NULL, ...)

Arguments

input_object	An xgboost model object (xgb.Booster)
treenum	Integer vector of 1-based tree numbers to extract (default: NULL extracts all trees).
...	Additional arguments (currently unused)

Value

A data.frame with one row per node and the following columns:

name

Node ID string ("Tree-Node" format, matches edgelist from/to)

is_leaf

Logical: TRUE for leaf nodes

feature

Split variable name (NA for leaves)

split

Split threshold (NA for leaves)

quality

Information gain for splits, leaf score for leaves

cover

Cover (sum of second-order gradient)

missing

Node ID ("Tree-Node" format) where observations with missing values are routed; NA for leaves

treenum

1-based tree number

label

Display label: "<feature>\n< <threshold>" for splits, rounded leaf score for leaves

Examples

if (requireNamespace("xgboost", quietly = TRUE)) {
  data(agaricus.train, package = "xgboost")
  bst <- xgboost::xgboost(
    x = agaricus.train$data,
    y = factor(agaricus.train$label),
    max_depth = 2, nrounds = 2, nthreads = 1
  )
  nl <- nodelist(bst)
  head(nl)
}

---