Struct DecisionTreeLearner 

pub struct DecisionTreeLearner { /* private fields */ }

ID3/C4.5-style decision tree learner.

use ipfrs_tensorlogic::{DecisionTreeLearner, DtlSample, DtlLearnerConfig, DtlCriterion};

let mut learner = DecisionTreeLearner::new(
    DtlLearnerConfig { max_depth: 5, ..Default::default() },
    vec!["petal_len".into(), "petal_wid".into()],
);

let samples = vec![
    DtlSample::new(vec![1.0, 0.5], "setosa"),
    DtlSample::new(vec![4.5, 1.5], "versicolor"),
    DtlSample::new(vec![5.0, 2.0], "virginica"),
    DtlSample::new(vec![1.2, 0.4], "setosa"),
    DtlSample::new(vec![4.8, 1.8], "versicolor"),
];
learner.fit(&samples).expect("example: should succeed in docs");

let pred = learner.predict(&[1.1, 0.45]).expect("example: should succeed in docs");
assert_eq!(pred.label, "setosa");

Implementations

impl DecisionTreeLearner

pub fn new(config: DtlLearnerConfig, feature_names: Vec<String>) -> Self

Create a new (untrained) learner.

feature_names is optional context: if non-empty its length must match the feature dimension of the first fit call.

pub fn with_defaults(feature_names: Vec<String>) -> Self

Convenience constructor with default configuration.

impl DecisionTreeLearner

pub fn fit(&mut self, samples: &[DtlSample]) -> Result<(), DtlError>

Train the decision tree on samples.

Clears any previously trained tree. Feature names, if provided at construction, are validated against the sample dimension.

impl DecisionTreeLearner

pub fn predict(&self, features: &[f64]) -> Result<DtlPrediction, DtlError>

Predict the class label for a single feature vector.

pub fn predict_batch( &self, samples: &[Vec<f64>], ) -> Vec<Result<DtlPrediction, DtlError>>

Predict a batch of feature vectors.

impl DecisionTreeLearner

pub fn feature_importance(&self) -> Vec<(String, f64)>

Compute normalised feature importances (weighted impurity reduction).

Returns a vector of (feature_name, importance) pairs sorted by importance descending. Values are normalised to sum to 1.0.

impl DecisionTreeLearner

pub fn depth(&self) -> usize

Return the depth of the trained tree (0 if not trained).

pub fn n_leaves(&self) -> usize

Return the number of leaf nodes (0 if not trained).

pub fn n_nodes(&self) -> usize

Return the total number of nodes (0 if not trained).

impl DecisionTreeLearner

pub fn prune(&mut self, min_samples: usize)

Post-hoc pruning: collapse split nodes where both children are leaves and neither has at least min_samples samples in the majority class.

This is a conservative bottom-up pruning pass that replaces such split nodes with a single leaf carrying the combined class distribution.

impl DecisionTreeLearner

pub fn learner_stats(&self) -> DtlLearnerStats

Return a snapshot of learner statistics.

pub fn history(&self) -> &VecDeque<DtlTrainingRecord>

Reference to the training history ring-buffer.

pub fn config(&self) -> &DtlLearnerConfig

Reference to the current configuration.

pub fn feature_names(&self) -> &[String]

Reference to the feature names.

pub fn class_labels(&self) -> &[String]

Reference to the discovered class labels.

pub fn root(&self) -> Option<&DtlNode>

Access the raw tree root (for serialisation / inspection).