Implementation:Evidentlyai Evidently Legacy Metric Preset Base

Knowledge Sources	Evidentlyai_Evidently
Domains	Machine Learning, Model Monitoring, Framework Architecture
Last Updated	2026-02-14 12:00 GMT

Overview

The MetricPreset class is the abstract base class for all metric presets in the Evidently legacy module, defining the interface that concrete presets must implement to generate a list of metrics.

Description

MetricPreset extends BasePreset and declares a single abstract method, generate_metrics, which concrete subclasses must implement. This method receives a DataDefinition object (describing the structure of the dataset) and an optional additional_data dictionary, and must return a list of AnyMetric instances.

The AnyMetric type alias is defined as Union[Metric, BaseGenerator[Metric]], allowing presets to return both individual metric instances and metric generators (which produce multiple metrics dynamically based on data characteristics).

The class is configured with is_base_type = True in its Config, indicating it is a base type in the Evidently type system and should not be instantiated directly.

Usage

This class should be subclassed when creating new metric presets. It provides the contract that the Evidently report and test suite infrastructure relies on to collect metrics from presets. Users do not instantiate this class directly; instead, they use concrete subclasses such as ClassificationPreset, DataDriftPreset, DataQualityPreset, or RecsysPreset.

Code Reference

Source Location

Repository: Evidentlyai_Evidently
File: src/evidently/legacy/metric_preset/metric_preset.py

Signature

AnyMetric = Union[Metric, BaseGenerator[Metric]]

class MetricPreset(BasePreset):
    """Base class for metric presets"""

    class Config:
        is_base_type = True

    @abc.abstractmethod
    def generate_metrics(
        self, data_definition: DataDefinition,
        additional_data: Optional[Dict[str, Any]]
    ) -> List[AnyMetric]:
        raise NotImplementedError()

Import

from evidently.legacy.metric_preset.metric_preset import MetricPreset
from evidently.legacy.metric_preset.metric_preset import AnyMetric

I/O Contract

Inputs

Name	Type	Required	Description
data_definition	DataDefinition	Yes	Describes the structure of the dataset (columns, types, prediction columns, etc.)
additional_data	Optional[Dict[str, Any]]	No	Additional context data (e.g., training data references) for metric generation

Outputs

Name	Type	Description
metrics	List[AnyMetric]	A list of Metric instances or metric generators to be evaluated by the Evidently report or test suite

Usage Examples

from evidently.legacy.metric_preset.metric_preset import MetricPreset, AnyMetric
from evidently.legacy.utils.data_preprocessing import DataDefinition
from typing import Any, Dict, List, Optional

# Creating a custom metric preset by subclassing MetricPreset
class MyCustomPreset(MetricPreset):
    def generate_metrics(
        self, data_definition: DataDefinition,
        additional_data: Optional[Dict[str, Any]]
    ) -> List[AnyMetric]:
        # Return a list of metric instances
        return [
            # ... metric instances here
        ]

Related Pages

Page Connections

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

Principle

Implementation

Heuristic

Environment