Implementation:Online ml River Metrics MicroAverage

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

Overview

Micro-average wrapper aggregating all outputs into a single metric by updating one shared metric instance.

Description

MicroAverage wraps any single-output metric to work with multi-output problems by maintaining just one metric instance that is updated with values from all outputs. Unlike MacroAverage which computes per-output metrics and averages them, MicroAverage treats all predictions across all outputs as coming from a single task. This weights the metric by the total number of predictions, making frequent outputs contribute more to the final score.

Usage

Use MicroAverage when you want the overall score to be weighted by the number of predictions per output, giving more importance to frequently occurring outputs. This is appropriate when outputs have different frequencies and you want the metric to reflect overall system performance weighted by actual usage. In multi-label classification, MicroAverage emphasizes performance on common labels over rare ones.

Code Reference

Source Location

Repository: Online_ml_River
File: river/metrics/multioutput/micro.py

Signature

class MicroAverage(MultiOutputMetric, metrics.base.WrapperMetric):
    def __init__(self, metric):
        # metric: Any classification or regression metric
        pass

Import

from river import metrics

I/O Contract

Method	Parameters	Returns	Description
update	y_true (dict), y_pred (dict), [w]	None	Updates single shared metric with all output values
get	-	float	Returns the shared metric's value

Usage Examples

from river import metrics

# Wrap F1 score for multi-output micro-averaging
micro_f1 = metrics.multioutput.MicroAverage(metrics.F1())

y_true = [
    {0: False, 1: True, 2: True},
    {0: True, 1: True, 2: False},
    {0: True, 1: False, 2: True},
]

y_pred = [
    {0: False, 1: True, 2: True},   # All correct
    {0: True, 1: False, 2: False},  # Label 1 wrong
    {0: False, 1: False, 2: True},  # Label 0 wrong
]

for yt, yp in zip(y_true, y_pred):
    micro_f1.update(yt, yp)

print(micro_f1)
# Single F1 score computed across all outputs globally

# Compare Micro vs Macro averaging
macro_f1 = metrics.multioutput.MacroAverage(metrics.F1())

for yt, yp in zip(y_true, y_pred):
    macro_f1.update(yt, yp)

print(f"Micro F1: {micro_f1.get():.2%}")
print(f"Macro F1: {macro_f1.get():.2%}")
# Micro: Weighted by total predictions
# Macro: Equal weight to each output

# Regression example
micro_mae = metrics.multioutput.MicroAverage(metrics.MAE())

y_true_reg = [
    {0: 1.0, 1: 2.0, 2: 3.0},
    {0: 2.0, 1: 3.0, 2: 4.0},
]

y_pred_reg = [
    {0: 1.1, 1: 2.2, 2: 2.9},
    {0: 2.1, 1: 2.8, 2: 4.2},
]

for yt, yp in zip(y_true_reg, y_pred_reg):
    micro_mae.update(yt, yp)

print(f"Micro MAE: {micro_mae.get():.3f}")
# Global MAE across all outputs and samples

# Access the underlying metric
print(f"Underlying metric: {micro_mae.metric}")

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment