Implementation:Online ml River Tree Nodes HTC

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

Overview

Hoeffding Tree Classifier (HTC) leaf node implementations providing different prediction strategies for classification tasks.

Description

This module provides three types of leaf nodes for Hoeffding Tree classifiers. LeafMajorityClass predicts using the majority class and monitors class distributions to determine split promise. LeafNaiveBayes uses Naive Bayes models for prediction when sufficient samples are available. LeafNaiveBayesAdaptive dynamically selects between majority class and Naive Bayes based on which performs better on the observed data.

Usage

Use LeafMajorityClass for simple and fast predictions, LeafNaiveBayes when features provide strong probabilistic information, and LeafNaiveBayesAdaptive when uncertain which strategy will perform better on your data stream.

Code Reference

Source Location

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

Signature

class LeafMajorityClass(HTLeaf):
    def __init__(self, stats, depth, splitter, **kwargs):
        ...
    def prediction(self, x, *, tree=None):
        ...
    def calculate_promise(self):
        ...
    def observed_class_distribution_is_pure(self):
        ...

class LeafNaiveBayes(LeafMajorityClass):
    def prediction(self, x, *, tree=None):
        ...
    def disable_attribute(self, att_index):
        pass

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

Import

from river.tree.nodes.htc_nodes import LeafMajorityClass
from river.tree.nodes.htc_nodes import LeafNaiveBayes
from river.tree.nodes.htc_nodes import LeafNaiveBayesAdaptive

I/O Contract

Input	Type	Description
stats	dict	Class observation counts
depth	int	Node depth in tree
splitter	Splitter	Numeric attribute observer

Output	Type	Description
prediction	dict	Normalized class probability distribution
promise	float	Split promise value (higher means more likely to split)

Usage Examples

from river.tree.nodes.htc_nodes import LeafNaiveBayesAdaptive
from river.tree.splitter import GaussianSplitter

# Create adaptive leaf
leaf = LeafNaiveBayesAdaptive(
    stats={'cat': 10, 'dog': 8},
    depth=2,
    splitter=GaussianSplitter()
)

# Learn from samples
x = {'size': 0.5, 'weight': 12.5}
leaf.learn_one(x, y='cat', w=1.0, tree=None)

# Get prediction (automatically chooses best strategy)
pred = leaf.prediction(x, tree=None)
# Returns: {'cat': 0.55, 'dog': 0.45}

# Check if node should split
promise = leaf.calculate_promise()

Related Pages

• Environment:Online_ml_River_Python_Runtime_Environment