DecisionTreeRegressor
#include <Skigen/Tree>
template <typename Scalar = double>
class Skigen::DecisionTreeRegressor(max_depth=-1, min_samples_split=2, max_features_mode=0, max_features_value=0.0, random_state=std::nullopt)
A decision tree regressor.
Uses MSE (Mean Squared Error) reduction as the splitting criterion. The prediction for a leaf node is the mean of the target values.
Mirrors sklearn.tree.DecisionTreeRegressor.
Attributes:
-
feature_importances : RowVectorType
-
n_targets : int
Methods
fit(X, y)
Build a decision tree regressor from the training set.
predict(X)
Predict using a sparse design matrix (densifies internally).
fit_with_indices(X, y, sample_indices)
fit(X, y)
Fit natively on a sparse design matrix without densifying.
Split finding runs through a CSC column accessor that materialises one feature column at a time (implicit zeros filled as 0), so the full dense matrix is never built. Results match a dense fit exactly.
fit_columns(cols, y, sample_indices, n_rows, n_cols)
predict(X)
Predict target values for X.
score(X, y)
Return the coefficient of determination.
fit_multi(X, Y)
Fit on a multi-target response matrix Y (n_samples × n_targets).
Implementation: fits one shared tree whose split criterion is the sum of target variances across all output columns, matching sklearn's joint multi-output tree structure.
The single-target API (fit(X, y) / predict(X)) is unchanged; after fit_multi, the first target column's tree backs predict() for callers that expect a vector output.
fit_with_indices_multi(X, Y, sample_indices)
Multi-target fit on a specific row index subset.
Mirrors fit_with_indices for the single-target path but builds one shared joint tree with summed-variance split scoring. Used by ensemble bootstraps that need per-tree multi-output sample subsets.
predict_multi(X)
Multi-target predict (n_samples × n_targets).