Implementation:Online ml River Tree Splitter QO

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

Overview

Quantization Observer (QO) splitter that uses hash-based quantization to discretize numerical features for efficient split evaluation in regression trees.

Description

QOSplitter implements hash-based feature quantization for regression trees. It discretizes incoming feature values into equal-sized intervals defined by a radius parameter. Split candidates are the midpoints between consecutive hash slots. The splitter supports both binary and multiway splits. Memory and computation scale with the radius: smaller radius means more slots and finer discretization but higher costs. Assumes features are scaled to similar ranges.

Usage

Use QOSplitter for regression when you want efficient memory-bounded split evaluation. Scale features first using preprocessing.StandardScaler, and set radius as a proportion of standard deviation (default 0.25).

Code Reference

Source Location

• Repository: Online_ml_River
• File: river/tree/splitter/qo_splitter.py

Signature

class QOSplitter(Splitter):
    def __init__(self, radius: float = 0.25, allow_multiway_splits=False):
        ...

    def update(self, att_val, target_val, w):
        ...

    def cond_proba(self, att_val, target_val):
        raise NotImplementedError

    def best_evaluated_split_suggestion(self, criterion, pre_split_dist, att_idx, binary_only=True):
        ...

    @property
    def is_target_class(self) -> bool:
        return False


class FeatureQuantizer:
    def __init__(self, radius: float):
        ...

    def update(self, x: float, y: float | utils.VectorDict, weight: float):
        ...

    def __iter__(self):
        ...

Import

from river.tree.splitter import QOSplitter

I/O Contract

Input	Type	Description
att_val	float	Numerical feature value
target_val	float/dict	Target value (supports multi-target)
w	float	Sample weight
radius	float	Quantization radius (default 0.25)
allow_multiway_splits	bool	Enable multiway splits (default False)

Output	Type	Description
split_suggestion	BranchFactory	Best split (binary or multiway) with statistics

Usage Examples

from river.tree.splitter import QOSplitter
from river.tree.split_criterion import VarianceRatioSplitCriterion
from river.stats import Var
from river import preprocessing

# Scale features first
scaler = preprocessing.StandardScaler()

# Create QO splitter with custom radius
splitter = QOSplitter(radius=0.5, allow_multiway_splits=True)

# Update with scaled observations
for x_val, y_val in [(0.5, 25.5), (0.8, 30.2), (0.3, 22.1)]:
    splitter.update(x_val, y_val, w=1.0)

# Get best split
criterion = VarianceRatioSplitCriterion()
pre_split = Var()
for y in [25.5, 30.2, 22.1]:
    pre_split.update(y, 1.0)

suggestion = splitter.best_evaluated_split_suggestion(
    criterion=criterion,
    pre_split_dist=pre_split,
    att_idx='scaled_feature',
    binary_only=False
)

print(f"Split: {suggestion.split_info}, Merit: {suggestion.merit}")

Related Pages

• Environment:Online_ml_River_Python_Runtime_Environment