Skip to main content

DecisionTreeClassifier

#include <Skigen/Tree>

template <typename Scalar = double>
class Skigen::DecisionTreeClassifier(max_depth=-1, min_samples_split=2, max_features_mode=0, max_features_value=0.0, random_state=std::nullopt)

A decision tree classifier.

A non-parametric supervised learning method used for classification. The model predicts the value of a target variable by learning simple decision rules inferred from the data features. Uses Gini impurity as the splitting criterion.

Mirrors sklearn.tree.DecisionTreeClassifier.

Attributes:

  • classes : const Eigen::VectorXi

  • n_classes : int

  • feature_importances : RowVectorType

Methods​

fit(X, y)​

Build a decision tree classifier from the training set.

fit(X, y)​

Fit on a sparse design matrix.

The sparse input is converted to a dense MatrixType internally and forwarded to the regular split-finding path. The numerical results match a dense fit; a native sparse split finder (skip-zero column scans accumulating per-class counts only over explicit nonzeros) would avoid the densification but is not implemented.

predict(X)​

Predict using a sparse design matrix (densifies internally).

fit_with_indices(X, y, sample_indices)​

Fit using a specific row index subset (used by RandomForest bootstraps).

Parameters:

  • sample_indices : const std::vector< Eigen::Index > & If empty, uses all rows of X. Otherwise, builds the tree from the rows specified (with possible repetitions).

predict(X)​

Predict class labels for samples in X.

predict_proba(X)​

Per-class probability estimates from leaf class distributions.

build_tree(X, y, indices, depth)​

fit(X, y)​

predict(X)​

Example​

// Best model
Skigen::DecisionTreeClassifier<double> best(5);
best.fit(split.X_train, split.y_train);
auto best_pred = best.predict(split.X_test);

std::cout << "\n=== Confusion Matrix (depth=5) ===\n";
auto cm = Skigen::Metrics::confusion_matrix(split.y_test, best_pred);
std::cout << cm << "\n";