plotNetwork: Plot an interaction variable network.

In aruaud/endoR: Interpret and Visualize Stable Tree Ensemble Models

Plot an interaction variable network.

Description

Returns a plot of nodes and edges. Plots are created with the ggraph and igraph packages.

Usage

plotNetwork(
  decision_ensemble,
  path_length = Inf,
  layout = "stress",
  colour_edge_low = "#E69F00",
  colour_edge_mid = "grey87",
  colour_edge_high = "#0072B2",
  colour_node_low = "#E69F00",
  colour_node_mid = "grey87",
  colour_node_high = "#0072B2",
  text_size = 4,
  hide_isolated_nodes = TRUE,
  seed = 0
)

Arguments

decision_ensemble	stable decision ensemble: list with edges, nodes, etc.
path_length	maximal number of edges between 2 nodes, default = Inf.
layout	network layout, default is 'stress' (see ggraph package)
colour_edge_low	colour for the negative interaction influence values (i.e., edges; default: yellowish)
colour_edge_mid	colour for the null interaction influence values (i.e., edges; default: light grey)
colour_edge_high	colour for the positive interaction influence values (i.e., edges; default: blue)
colour_node_low	colour for the negative feature influence values (i.e., nodes; default: yellowish)
colour_node_mid	colour for the null feature influence values (i.e., nodes; default: light grey)
colour_node_high	colour for the positive feature influence values (i.e., nodes; default: blue)
text_size	size of node labels.
hide_isolated_nodes	logical, default = TRUE (= nodes without any edge are not shown).
seed	the seed to use for generating the network.

Value

a ggraph object

Examples

library(randomForest)
library(caret)

# import data and fit model
data(iris)
mod <- randomForest(Species ~ ., data = iris)

# Fit a decision ensemble to predict the species setosa (vs versicolor and
# virginica): no regularization (no decision pruning, discretization,
# bootstrapping, or decision filtering)
endo_setosa <- model2DE(model = mod, model_type = "rf", data = iris[, -5]
    , target = iris$Species, classPos = "setosa"
    , filter = FALSE, discretize = FALSE, prune = FALSE)

# Only decision pruning (default = TRUE) and discretization (default in 2
# categories, we want 3 categories so we change K); no bootstrapping or
# decision filtering.
endo_setosa <- model2DE(model = mod, model_type = "rf", data = iris[, -5]
    , target = iris$Species, classPos = "setosa"
    , filter = FALSE, discretize = TRUE, K = 3)

# idem but run it in parallel on 2 threads
endo_setosa <- model2DE(model = mod, model_type = "rf", data = iris[, -5]
    , target = iris$Species, classPos = "setosa"
    , filter = FALSE, discretize = TRUE, K = 3
    , in_parallel = TRUE, n_cores = 2)

# Plot the decision ensemble:
# Plants from the setosa species have small petal and narrow long sepals.
plotFeatures(endo_setosa, levels_order = c("Low", "Medium", "High"))
plotNetwork(endo_setosa, hide_isolated_nodes = FALSE, layout = "fr")

aruaud/endoR documentation built on Jan. 25, 2025, 2:20 a.m.