compute_tree {treeheatr}

Compute decision tree from data set

Description

Compute decision tree from data set

Usage

compute_tree(
  x,
  data_test = NULL,
  target_lab = NULL,
  task = c("classification", "regression"),
  feat_types = NULL,
  label_map = NULL,
  clust_samps = TRUE,
  clust_target = TRUE,
  custom_layout = NULL,
  lev_fac = 1.3,
  panel_space = 0.001
)

Arguments

x

Dataframe or a 'party' or 'partynode' object representing a custom tree. If a dataframe is supplied, conditional inference tree is computed. If a custom tree is supplied, it must follow the partykit syntax: https://cran.r-project.org/web/packages/partykit/vignettes/partykit.pdf

data_test

Tidy test dataset. Required if 'x' is a 'partynode' object. If NULL, heatmap displays (training) data 'x'.

target_lab

Name of the column in data that contains target/label information.

task

Character string indicating the type of problem, either 'classification' (categorical outcome) or 'regression' (continuous outcome).

feat_types

Named vector indicating the type of each features, e.g., c(sex = 'factor', age = 'numeric'). If feature types are not supplied, infer from column type.

label_map

Named vector of the meaning of the target values, e.g., c(‘0' = ’Edible', ‘1' = ’Poisonous').

clust_samps

Logical. If TRUE, hierarchical clustering would be performed among samples within each leaf node.

clust_target

Logical. If TRUE, target/label is included in hierarchical clustering of samples within each leaf node and might yield a more interpretable heatmap.

custom_layout

Dataframe with 3 columns: id, x and y for manually input custom layout.

lev_fac

Relative weight of child node positions according to their levels, commonly ranges from 1 to 1.5. 1 for parent node perfectly in the middle of child nodes.

panel_space

Spacing between facets relative to viewport, recommended to range from 0.001 to 0.01.

Value

A list of results from 'partykit::ctree' or provided custom tree, including fit, estimates, smart layout and terminal data.

Examples

fit_tree <- compute_tree(penguins, target_lab = 'species')
fit_tree$fit
fit_tree$layout
dplyr::select(fit_tree$term_dat, - contains('nodedata'))