Implementation:Online ml River Tree Splitter TEBST

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

Overview

Truncated E-BST splitter that rounds feature values before storage to reduce memory and processing time.

Description

TEBSTSplitter is a memory-optimized variant of EBSTSplitter that rounds incoming feature values to a specified number of decimal places before inserting them into the binary search tree. This reduces the number of unique nodes by merging small variations into the same value. The trade-off is slightly coarser split candidates in exchange for improved memory efficiency and faster processing.

Usage

Use TEBSTSplitter when memory or processing time is constrained and you can tolerate rounding numerical features. Set digits parameter based on required precision versus memory constraints.

Code Reference

Source Location

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

Signature

class TEBSTSplitter(EBSTSplitter):
    def __init__(self, digits: int = 1):
        ...

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

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

Import

from river.tree.splitter import TEBSTSplitter

I/O Contract

Input	Type	Description
digits	int	Decimal places for rounding (default 1)
att_val	float	Numerical feature value to be rounded
target_val	float/dict	Target value (supports multi-target)
w	float	Sample weight

Output	Type	Description
split_suggestion	BranchFactory	Best split using rounded values

Usage Examples

from river.tree.splitter import TEBSTSplitter
from river.tree.split_criterion import VarianceRatioSplitCriterion
from river.stats import Var

# Create truncated E-BST with 2 decimal places
splitter = TEBSTSplitter(digits=2)

# Update with precise values (will be rounded)
splitter.update(5.5234, 10.2, 1.0)
splitter.update(5.5189, 12.5, 1.0)  # Rounded to same value as above
splitter.update(4.8123, 9.1, 1.0)

# The tree has fewer nodes due to rounding
# 5.5234 and 5.5189 both become 5.52

# Get best split
criterion = VarianceRatioSplitCriterion()
pre_split = Var()
pre_split.update(10.2, 1.0)
pre_split.update(12.5, 1.0)
pre_split.update(9.1, 1.0)

suggestion = splitter.best_evaluated_split_suggestion(
    criterion=criterion,
    pre_split_dist=pre_split,
    att_idx='feature1',
    binary_only=True
)

print(f"Threshold: {suggestion.split_info}")

# Compare with regular EBST
from river.tree.splitter import EBSTSplitter
regular = EBSTSplitter()
# Would create more nodes for same data

Related Pages

• Environment:Online_ml_River_Python_Runtime_Environment