Struct TreeSlot 

pub struct TreeSlot { /* private fields */ }

Manages the lifecycle of a single tree in the ensemble.

When the drift detector signals DriftSignal::Warning, an alternate tree begins training alongside the active tree. When DriftSignal::Drift is confirmed, the alternate replaces the active tree (or a fresh tree is created if no alternate exists). The drift detector is then reset via clone_fresh to monitor the new tree.

Implementations

impl TreeSlot

pub fn new( tree_config: TreeConfig, detector: Box<dyn DriftDetector>, max_tree_samples: Option<u64>, ) -> Self

Create a new TreeSlot with a fresh tree and drift detector.

The active tree starts as a single-leaf tree (prediction = 0.0). No alternate tree is created until a Warning signal is received.

pub fn from_trees( active: HoeffdingTree, alternate: Option<HoeffdingTree>, tree_config: TreeConfig, detector: Box<dyn DriftDetector>, max_tree_samples: Option<u64>, ) -> Self

Reconstruct a TreeSlot from pre-built trees and a fresh drift detector.

Used during model deserialization to restore tree state without replaying the training stream.

pub fn train_and_predict( &mut self, features: &[f64], gradient: f64, hessian: f64, ) -> f64

Train the active tree (and alternate if it exists) on a single sample.

The absolute value of the gradient is fed to the drift detector as an error proxy (gradient = derivative of loss = prediction error signal).

Returns

The prediction from the active tree before training on this sample. This ensures the prediction reflects only previously seen data, which is critical for unbiased gradient computation in the boosting loop.

Drift handling

• DriftSignal::Stable: no action.
• DriftSignal::Warning: spawn an alternate tree if one is not already being trained. The alternate receives the same training sample.
• DriftSignal::Drift: replace the active tree with the alternate (or a fresh tree if no alternate exists). The drift detector is reset via clone_fresh so it monitors the new tree from a clean state.

pub fn predict(&self, features: &[f64]) -> f64

Predict without training.

Routes the feature vector through the active tree and returns the leaf value. Does not update any state.

pub fn predict_with_variance(&self, features: &[f64]) -> (f64, f64)

Predict with variance for confidence estimation.

Returns (leaf_value, variance) where variance = 1 / (H_sum + lambda).

pub fn n_leaves(&self) -> usize

Number of leaves in the active tree.

pub fn n_samples_seen(&self) -> u64

Total samples the active tree has seen.

pub fn has_alternate(&self) -> bool

Whether an alternate tree is currently being trained.

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

Accumulated split gains per feature from the active tree.

pub fn active_tree(&self) -> &HoeffdingTree

Immutable access to the active tree.

pub fn alternate_tree(&self) -> Option<&HoeffdingTree>

Immutable access to the alternate tree (if one is being trained).

pub fn tree_config(&self) -> &TreeConfig

Immutable access to the tree configuration.

pub fn detector(&self) -> &dyn DriftDetector

Immutable access to the drift detector.

pub fn detector_mut(&mut self) -> &mut dyn DriftDetector

Mutable access to the drift detector.

pub fn alt_detector(&self) -> Option<&dyn DriftDetector>

Immutable access to the alternate drift detector (always None in the current architecture – the alternate tree shares the main detector). Reserved for future use.

pub fn alt_detector_mut(&mut self) -> Option<&mut dyn DriftDetector>

Mutable access to the alternate drift detector.

pub fn reset(&mut self)

Reset to a completely fresh state: new tree, no alternate, reset detector.

Trait Implementations

impl Clone for TreeSlot

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TreeSlot

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

Formats the value using the given formatter.