Struct GradientBooster 

pub struct GradientBooster { /* private fields */ }

Main gradient boosting implementation.

GradientBooster is the core ensemble model that trains a sequence of decision trees to minimize a given loss function. It supports various training strategies and provides comprehensive monitoring capabilities.

Key Features

• Loss Functions: MSE, MAE, Huber (robust regression), LogLoss (classification)
• Regularization: Learning rate shrinkage, L2 regularization (lambda)
• Stochastic Boosting: Row subsampling for variance reduction
• Early Stopping: Configurable patience and improvement tolerance
• Feature Importance: Gain-based importance scores

Serialization

This struct derives Serialize and Deserialize, allowing trained models to be saved and loaded using serde-compatible formats.

Implementations

impl GradientBooster

pub fn new(config: GBRTConfig) -> BoostingResult<Self>

Creates a new gradient booster with the given configuration.

Validates the configuration before returning. The booster is untrained and ready for [fit()] to be called.

Parameters

• config: GBRTConfig with hyperparameters and training options.

Errors

Returns BoostingError::ConfigError if configuration validation fails.

pub fn fit( &mut self, train_data: &Dataset, validation_data: Option<&Dataset>, ) -> BoostingResult<()>

Trains the model on a dataset with optional validation data.

This is the main training entry point. It performs the complete gradient boosting algorithm including:

• Data validation and initialization
• Iterative tree building with gradient optimization
• Optional stochastic subsampling
• Validation loss monitoring and early stopping
• Feature importance computation

Parameters

• train_data: Training dataset with features and targets.
• validation_data: Optional validation dataset for early stopping and monitoring.

Errors

Returns errors for:

• Invalid training data (empty, mismatched dimensions)
• Configuration issues detected during training
• Tree building failures
• Loss computation problems (e.g., NaN values)

Early Stopping

If config.early_stopping_rounds is set and validation_data is provided, training stops when validation loss doesn’t improve significantly (by more than config.early_stopping_tolerance) for the specified number of rounds.

pub fn predict(&self, features: &FeatureMatrix) -> BoostingResult<Vec<f64>>

Makes predictions for a batch of samples.

Calculates predictions by summing contributions from all trees with shrinkage applied. Applies loss-specific transformation (e.g., sigmoid for LogLoss).

Parameters

• features: Feature matrix with shape (n_samples, n_features).

Returns

Vector of predictions of length n_samples.

Errors

• PredictionError::ModelNotTrained if called before [fit()]
• PredictionError::FeatureMismatch if features.n_features() doesn’t match training data

pub fn predict_single(&self, features: &[f64]) -> BoostingResult<f64>

Makes a prediction for a single sample.

More efficient than predict() for single-sample inference.

Parameters

• features: Slice of feature values of length n_features.

Returns

Single prediction value.

Errors

• PredictionError::ModelNotTrained if called before [fit()]
• PredictionError::FeatureMismatch if features.len() doesn’t match training data

pub fn feature_importance(&self) -> &[f64]

Returns feature importance scores from the trained model.

Importance is computed as the normalized total gain contributed by splits on each feature across all trees. Scores sum to 1.0.

Returns

Slice of importance scores of length n_features. Returns zeros if config.compute_feature_importance is false.

Panics

Will panic if called before any model has been trained (no features are known).

pub fn training_state(&self) -> Option<&TrainingState>

Returns the training history and validation state.

Contains per-iteration metrics and the best iteration if early stopping was used. This is useful for plotting learning curves or analyzing training dynamics.

Returns

Some(TrainingState) after training, None before training.

pub fn n_trees(&self) -> usize

Returns the number of trees in the ensemble.

Note: This may be fewer than config.n_estimators if early stopping triggered.

Returns

Number of trees after training.

pub fn config(&self) -> &GBRTConfig

Returns the configuration used to create this booster.

pub fn is_trained(&self) -> bool

Checks whether the model has been trained.

Trait Implementations

impl Clone for GradientBooster

fn clone(&self) -> GradientBooster

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for GradientBooster

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for GradientBooster

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Display for GradientBooster

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Serialize for GradientBooster

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.