Skip to content

evaluate

evaluate

Classes:

Name Description
TimeSeriesMetricParameters

Base for one time-series metric's parameters.

AutocorrelationSimilarityParameters

Control autocorrelation similarity evaluation.

TimeSeriesEvaluationParameters

Metric-specific time-series evaluation configuration.

EvaluationParameters

Configuration for evaluating synthetic data quality and privacy.

TimeSeriesMetricParameters pydantic-model

Bases: Parameters

Base for one time-series metric's parameters.

Preflight validates value_columns for every subclass instance found on TimeSeriesEvaluationParameters, naming the metric by metric_label.

Fields:

Validators:

  • _reject_removed_legacy_fields

value_columns = None pydantic-field

Numeric value columns to evaluate. Defaults to all shared columns inferred as numeric. Changing this selection changes the meaning of the aggregate score.

AutocorrelationSimilarityParameters pydantic-model

Bases: TimeSeriesMetricParameters

Control autocorrelation similarity evaluation.

Timestamp ordering and sequence grouping use the top-level time-series and data settings. The requested lag is capped automatically for short sequences, and undersized or constant training profiles are skipped. A constant synthetic profile paired with varying training data receives zero similarity.

Fields:

Validators:

  • _reject_removed_legacy_fields

max_lag = DEFAULT_AUTOCORRELATION_MAX_LAG pydantic-field

Maximum requested lag; short sequences use a smaller stable lag cap. Scores computed with different lag horizons are not directly comparable.

min_points = DEFAULT_AUTOCORRELATION_MIN_POINTS pydantic-field

Minimum finite observations required in each sequence.

max_groups = 128 pydantic-field

Maximum shared groups to evaluate using a reproducible sample to bound runtime.

TimeSeriesEvaluationParameters pydantic-model

Bases: Parameters

Metric-specific time-series evaluation configuration.

Each metric's parameters are one TimeSeriesMetricParameters field.

Fields:

Validators:

  • _reject_removed_legacy_fields

enabled = None pydantic-field

Enable time-series evaluation metrics. Defaults to None, which follows time_series.is_timeseries. True requires top-level time-series mode.

autocorrelation pydantic-field

Autocorrelation similarity metric parameters.

EvaluationParameters pydantic-model

Bases: Parameters

Configuration for evaluating synthetic data quality and privacy.

This class controls which evaluation metrics are computed and how they are configured. It includes privacy attack evaluations, statistical quality metrics, and downstream machine learning performance assessments.

Fields:

Validators:

  • _reject_removed_legacy_fields

mia_enabled pydantic-field

Enable membership inference attack evaluation for privacy assessment.

mia_excluded_columns = None pydantic-field

Columns to leave out of the membership inference attack. Useful when identifiers were already replaced outside Safe Synthesizer, or are expected to repeat verbatim in the synthetic data.

aia_enabled pydantic-field

Enable attribute inference attack evaluation for privacy assessment.

sqs_report_columns = DEFAULT_SQS_REPORT_COLUMNS pydantic-field

Number of columns to include in statistical quality reports.

sqs_report_rows = DEFAULT_RECORD_COUNT pydantic-field

Number of rows to include in statistical quality reports.

mandatory_columns pydantic-field

Number of mandatory columns that must be used in evaluation.

enabled pydantic-field

Enable or disable evaluation.

quasi_identifier_count pydantic-field

Number of quasi-identifiers to sample for privacy attacks.

pii_replay_enabled pydantic-field

Enable PII Replay detection.

pii_replay_entities pydantic-field

List of entities for PII Replay. If not provided, default entities will be used.

pii_replay_columns pydantic-field

List of columns for PII Replay. If not provided, only entities will be used.

time_series pydantic-field

Time-series-specific evaluation settings.