Crate irithyll 

Irithyll

Streaming Gradient Boosted Trees for evolving data streams.

Irithyll implements the SGBT algorithm (Gunasekara et al., 2024) in pure Rust, providing incremental gradient boosted tree ensembles that learn one sample at a time. Trees are built using Hoeffding-bound split decisions and automatically replaced when concept drift is detected, making the model suitable for non-stationary environments where the data distribution shifts over time.

Key Capabilities

• True online learning – train_one() processes samples individually, no batching
• Concept drift adaptation – automatic tree replacement via Page-Hinkley, ADWIN, or DDM
• Async streaming – tokio-native AsyncSGBT with bounded channels and concurrent prediction
• Pluggable losses – squared, logistic, softmax, Huber, or custom via the Loss trait
• Multi-class – one-vs-rest committees with softmax normalization (MulticlassSGBT)
• Serialization – checkpoint/restore via JSON or bincode for zero-downtime deployments
• Production-grade – SIMD acceleration, parallel training, Arrow/Parquet I/O, ONNX export

Feature Flags

Feature	Default	Description
serde-json	Yes	JSON model serialization
serde-bincode	No	Bincode serialization (compact, fast)
parallel	No	Rayon-based parallel tree training (ParallelSGBT)
simd	No	AVX2 histogram acceleration
kmeans-binning	No	K-means histogram binning strategy
arrow	No	Apache Arrow RecordBatch integration
parquet	No	Parquet file I/O
onnx	No	ONNX model export
neural-leaves	No	Experimental MLP leaf models
full	No	Enable all features

Quick Start

use irithyll::{SGBTConfig, SGBT, Sample};

let config = SGBTConfig::builder()
    .n_steps(100)
    .learning_rate(0.0125)
    .build()
    .unwrap();

let mut model = SGBT::new(config);

// Stream samples one at a time
let sample = Sample::new(vec![1.0, 2.0, 3.0], 0.5);
model.train_one(&sample);
let prediction = model.predict(&sample.features);

Algorithm

The ensemble maintains n_steps boosting stages, each owning a streaming Hoeffding tree and a drift detector. For each sample (x, y):

1. Compute the ensemble prediction F(x) = base + lr * sum(tree_s(x))
2. For each boosting step, compute gradient/hessian of the loss at the residual
3. Update the tree’s histogram accumulators and evaluate splits via Hoeffding bound
4. Feed the standardized error to the drift detector
5. If drift is detected, replace the tree with a fresh alternate

This enables continuous learning without storing past data, with statistically sound split decisions and automatic adaptation to distribution shifts.

Re-exports

pub use drift::DriftDetector;

pub use drift::DriftSignal;

pub use ensemble::bagged::BaggedSGBT;

pub use ensemble::config::FeatureType;

pub use ensemble::config::SGBTConfig;

pub use ensemble::multi_target::MultiTargetSGBT;

pub use ensemble::multiclass::MulticlassSGBT;

pub use ensemble::quantile_regressor::QuantileRegressorSGBT;

pub use ensemble::DynSGBT;

pub use ensemble::SGBT;

pub use error::ConfigError;

pub use error::IrithyllError;

pub use histogram::BinnerKind;

pub use histogram::BinningStrategy;

pub use loss::Loss;

pub use loss::LossType;

pub use sample::Observation;

pub use sample::Sample;

pub use sample::SampleRef;

pub use tree::StreamingTree;

pub use explain::importance_drift::ImportanceDriftMonitor;

pub use explain::streaming::StreamingShap;

pub use explain::treeshap::ShapValues;

pub use stream::AsyncSGBT;

pub use stream::Prediction;

pub use stream::PredictionStream;

pub use stream::Predictor;

pub use stream::SampleSender;

pub use metrics::conformal::AdaptiveConformalInterval;

pub use metrics::ewma::EwmaClassificationMetrics;

pub use metrics::ewma::EwmaRegressionMetrics;

pub use metrics::rolling::RollingClassificationMetrics;

pub use metrics::rolling::RollingRegressionMetrics;

pub use metrics::ClassificationMetrics;

pub use metrics::FeatureImportance;

pub use metrics::MetricSet;

pub use metrics::RegressionMetrics;

Modules

drift

Concept drift detection algorithms.

ensemble

SGBT ensemble orchestrator — the core boosting loop.

error

Error types for Irithyll.

explain

TreeSHAP explanations for streaming gradient boosted trees.

histogram

Histogram-based feature binning for streaming tree construction.

loss

Loss functions for gradient boosting.

metrics

Online metric tracking for streaming model evaluation.

sample

Core data types for streaming samples.

serde_support

Model serialization and deserialization support.

stream

Async streaming infrastructure for tokio-native sample ingestion.

tree

Streaming decision trees with Hoeffding-bound split decisions.