Implementation:Evidentlyai Evidently Legacy Target Drift Preset

Knowledge Sources	Evidentlyai_Evidently
Domains	ML Monitoring, Data Drift
Last Updated	2026-02-14 12:00 GMT

Overview

TargetDriftPreset is a metric preset that assembles a set of drift detection and correlation metrics for target and prediction columns, adapting its behavior based on task type (regression vs classification).

Description

The TargetDriftPreset class extends MetricPreset and dynamically generates a list of metrics based on the data definition (target column, prediction columns, and task type). The generated metrics include:

• ColumnDriftMetric -- applied to target and prediction columns (including prediction probability columns for classification) to detect statistical drift
• ColumnValuePlot -- added for the target column only when the task is regression
• ColumnCorrelationsMetric -- added for both target and prediction columns to show correlations
• TargetByFeaturesTable -- added when target or prediction columns are present

The preset supports extensive customization of statistical tests and thresholds:

• A global stattest and stattest_threshold
• Type-specific overrides: cat_stattest, num_stattest, text_stattest (and corresponding thresholds)
• Per-column overrides: per_column_stattest and per_column_stattest_threshold

Statistical test resolution is handled by resolve_stattest_threshold, which determines the appropriate test and threshold for each column based on its type (categorical or numerical, determined by the task type).

The class is registered with the type alias "evidently:metric_preset:TargetDriftPreset".

Usage

Use this preset when you need to monitor target and prediction drift between a reference and current dataset. It is particularly useful for model monitoring pipelines where you want to detect distribution shifts in the target variable and model predictions over time.

Code Reference

Source Location

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

Signature

class TargetDriftPreset(MetricPreset):
    class Config:
        type_alias = "evidently:metric_preset:TargetDriftPreset"

    columns: Optional[List[str]]
    stattest: Optional[PossibleStatTestType]
    cat_stattest: Optional[PossibleStatTestType]
    num_stattest: Optional[PossibleStatTestType]
    text_stattest: Optional[PossibleStatTestType]
    per_column_stattest: Optional[Dict[str, PossibleStatTestType]]
    stattest_threshold: Optional[float]
    cat_stattest_threshold: Optional[float]
    num_stattest_threshold: Optional[float]
    text_stattest_threshold: Optional[float]
    per_column_stattest_threshold: Optional[Dict[str, float]]

    def __init__(
        self,
        columns: Optional[List[str]] = None,
        stattest: Optional[PossibleStatTestType] = None,
        cat_stattest: Optional[PossibleStatTestType] = None,
        num_stattest: Optional[PossibleStatTestType] = None,
        text_stattest: Optional[PossibleStatTestType] = None,
        per_column_stattest: Optional[Dict[str, PossibleStatTestType]] = None,
        stattest_threshold: Optional[float] = None,
        cat_stattest_threshold: Optional[float] = None,
        num_stattest_threshold: Optional[float] = None,
        text_stattest_threshold: Optional[float] = None,
        per_column_stattest_threshold: Optional[Dict[str, float]] = None,
    ):
        ...

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

Import

from evidently.legacy.metric_preset.target_drift import TargetDriftPreset

I/O Contract

Inputs

Name	Type	Required	Description
columns	Optional[List[str]]	No	List of column names (currently stored but not directly used in metric generation).
stattest	Optional[PossibleStatTestType]	No	Default statistical test to use for drift detection.
cat_stattest	Optional[PossibleStatTestType]	No	Statistical test override for categorical columns.
num_stattest	Optional[PossibleStatTestType]	No	Statistical test override for numerical columns.
text_stattest	Optional[PossibleStatTestType]	No	Statistical test override for text columns.
per_column_stattest	Optional[Dict[str, PossibleStatTestType]]	No	Per-column statistical test overrides.
stattest_threshold	Optional[float]	No	Default threshold for the statistical test.
cat_stattest_threshold	Optional[float]	No	Threshold override for categorical columns.
num_stattest_threshold	Optional[float]	No	Threshold override for numerical columns.
text_stattest_threshold	Optional[float]	No	Threshold override for text columns.
per_column_stattest_threshold	Optional[Dict[str, float]]	No	Per-column threshold overrides.

Outputs

Name	Type	Description
return	List[AnyMetric]	A dynamically generated list of metrics including ColumnDriftMetric, ColumnValuePlot (regression only), ColumnCorrelationsMetric, and TargetByFeaturesTable based on the data definition.

Usage Examples

from evidently.legacy.metric_preset.target_drift import TargetDriftPreset
from evidently.legacy.report import Report

# Basic target drift analysis with default settings
report = Report(metrics=[TargetDriftPreset()])
report.run(reference_data=ref_df, current_data=curr_df)

# Target drift with custom statistical test and threshold
report = Report(metrics=[
    TargetDriftPreset(
        stattest="ks",
        stattest_threshold=0.05,
        cat_stattest="chi2",
        cat_stattest_threshold=0.01,
    )
])
report.run(reference_data=ref_df, current_data=curr_df)

# Target drift with per-column test overrides
report = Report(metrics=[
    TargetDriftPreset(
        per_column_stattest={"target": "wasserstein"},
        per_column_stattest_threshold={"target": 0.1},
    )
])
report.run(reference_data=ref_df, current_data=curr_df)

Related Pages

• Environment:Evidentlyai_Evidently_Python_Core_Environment