Implementation:Online ml River Tree Nodes HTLeaf

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

Overview

Base abstract class for all leaf nodes in Hoeffding Trees, defining the core interface and functionality for learning nodes.

Description

HTLeaf is the foundational abstract class for all Hoeffding Tree leaf nodes. It manages target statistics, depth, attribute splitters, and provides methods for updating statistics, finding split candidates, and making predictions. Leaves can be activated or deactivated to manage memory. The class enforces implementation of abstract methods for total_weight calculation, prediction generation, promise calculation, and statistics updates.

Usage

Do not instantiate HTLeaf directly. Instead, extend it to create specific leaf node types for classification or regression tasks with different prediction strategies.

Code Reference

Source Location

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

Signature

class HTLeaf(Leaf, abc.ABC):
    def __init__(self, stats, depth, splitter, **kwargs):
        ...

    @property
    @abc.abstractmethod
    def total_weight(self) -> float:
        pass

    def is_active(self):
        ...

    def activate(self):
        ...

    def deactivate(self):
        ...

    @staticmethod
    @abc.abstractmethod
    def new_nominal_splitter():
        pass

    @abc.abstractmethod
    def update_stats(self, y, w):
        pass

    def update_splitters(self, x, y, w, nominal_attributes):
        ...

    def best_split_suggestions(self, criterion, tree) -> list[BranchFactory]:
        ...

    def disable_attribute(self, att_id):
        ...

    def learn_one(self, x, y, *, w=1.0, tree=None):
        ...

    @abc.abstractmethod
    def prediction(self, x, *, tree=None) -> dict:
        pass

    @abc.abstractmethod
    def calculate_promise(self) -> int:
        pass

Import

from river.tree.nodes.leaf import HTLeaf

I/O Contract

Input	Type	Description
stats	dict/Var	Target statistics (dict for classification, Var for regression)
depth	int	Node depth in tree
splitter	Splitter	Numeric attribute observer
x	dict	Feature dictionary
y	Target	Target value (class label or numeric)
w	float	Sample weight

Output	Type	Description
split_suggestions	list[BranchFactory]	List of candidate splits
prediction	dict/float	Prediction (probabilities for classification, value for regression)
promise	int	Node promise for splitting

Usage Examples

# HTLeaf is abstract, so use a concrete implementation
from river.tree.nodes.htc_nodes import LeafMajorityClass
from river.tree.splitter import GaussianSplitter

leaf = LeafMajorityClass(
    stats={'A': 10, 'B': 5},
    depth=1,
    splitter=GaussianSplitter()
)

# Check if active
assert leaf.is_active()

# Update with sample
x = {'feature1': 0.5}
leaf.learn_one(x, y='A', w=1.0, tree=None)

# Deactivate to save memory
leaf.deactivate()
assert not leaf.is_active()

# Get split suggestions
from river.tree.split_criterion import GiniSplitCriterion
criterion = GiniSplitCriterion()
suggestions = leaf.best_split_suggestions(criterion, tree=None)

Related Pages

• Environment:Online_ml_River_Python_Runtime_Environment