Implementation:Online ml River Tree Nodes HATC

Knowledge Sources	Online_ml_River
Domains	Online_Learning, Decision_Trees
Last Updated	2026-02-08 16:00 GMT

Overview

Hoeffding Adaptive Tree Classifier (HATC) node implementations with drift detection and alternate subtree management for adapting to concept drift.

Description

This module implements adaptive nodes for Hoeffding Adaptive Tree classifiers. The AdaLeafClassifier uses Naive Bayes Adaptive prediction with ADWIN drift detection to monitor classification errors. The AdaBranchClassifier maintains alternate subtrees that are grown when drift is detected and can replace the main subtree if they perform better. Bootstrap sampling via Poisson distribution is supported for improved learning.

Usage

Use HATC nodes when building classification trees that need to adapt to concept drift in non-stationary data streams. The drift detection mechanism automatically creates alternate subtrees and switches to them when beneficial.

Code Reference

Source Location

• Repository: Online_ml_River
• File: river/tree/nodes/hatc_nodes.py

Signature

class AdaLeafClassifier(LeafNaiveBayesAdaptive):
    def __init__(self, stats, depth, splitter, drift_detector, rng, **kwargs):
        ...
    def learn_one(self, x, y, *, w=1.0, tree=None, parent=None, parent_branch=None):
        ...
    def prediction(self, x, *, tree=None):
        ...

class AdaBranchClassifier(DTBranch):
    def __init__(self, stats: dict, *children, drift_detector: base.DriftDetector, **attributes):
        ...
    def traverse(self, x, until_leaf=True) -> list[HTLeaf]:
        ...
    def learn_one(self, x, y, *, w=1.0, tree=None, parent=None, parent_branch=None):
        ...
    def kill_tree_children(self, tree):
        ...

Import

from river.tree.nodes.hatc_nodes import AdaLeafClassifier
from river.tree.nodes.hatc_nodes import AdaBranchClassifier

I/O Contract

Input	Type	Description
x	dict	Feature dictionary
y	int/str	Target class label
w	float	Sample weight (default 1.0)
drift_detector	DriftDetector	Drift detection algorithm instance

Output	Type	Description
prediction	dict	Class probabilities weighted by error estimates
found_nodes	list[HTLeaf]	Leaves including alternate subtree leaves

Usage Examples

from river.tree.nodes.hatc_nodes import AdaLeafClassifier
from river.drift import ADWIN
from river.tree.splitter import GaussianSplitter
import random

# Create adaptive leaf with drift detection
rng = random.Random(42)
leaf = AdaLeafClassifier(
    stats={0: 5, 1: 5},
    depth=1,
    splitter=GaussianSplitter(),
    drift_detector=ADWIN(),
    rng=rng
)

# Update with samples (uses Poisson bootstrap)
x = {'feature1': 0.5, 'feature2': 1.2}
leaf.learn_one(x, y=1, w=1.0, tree=None)

# Predictions are weighted by error estimates
pred = leaf.prediction(x, tree=None)

Related Pages

• Environment:Online_ml_River_Python_Runtime_Environment