fs.review()

When to Use It

Use in Python scripts, data ingestion pipelines, or notebooks to evaluate a dataset's readiness for ML modeling in a single call. It consolidates schema checks, data quality audits, and target leakage diagnostics into a single structured result.

Parameters

• source: Dataset | str | DataFrame. The input dataset path (CSV, Parquet, Excel) or in-memory DataFrame (pandas, Polars).
• previous: object | None (default None). Prior snapshot for diff-aware reviews. When provided, the DiffReviewer compares the current dataset against it and attaches the DatasetDiffResult to result.diff.
• target_column: str | None (default None). Name of the target column. Highly recommended to enable target leakage checks.
• enabled_reviewers: Sequence[str] | None (default None). Optional list of specific reviewer IDs to execute.
• enabled_categories: Sequence[ReviewCategory] | None (default None). Optional list of reviewer categories to execute (schema, quality, leakage, diff, feature_quality, custom).
• reviewer_config: Mapping[str, Mapping[str, Any]] | None (default None). Parameter overrides for specific reviewers (e.g. customized thresholds).
• max_correlation_columns: int (default 100). Cap limit for correlation matrix computation during profiling.
• max_frequency_table_size: int (default 1000). Frequency table storage cap.

Return Value

Returns a frozen ReviewResult dataclass containing:

• engine_version: str representing the Review Engine result schema version (currently "0.4.0").
• dataset_summary: DatasetSummary with row and column count descriptors.
• generated_at: UTC timestamp.
• sections: Sorted sequence of ReviewSection objects representing the active reviewers' sections (sorted from critical to passed).
• recommendations: Flat, ranked, cross-section list of Recommendation objects generated by the centralized Recommendation Engine.
• overall_summary: Concise plain-text roll-up.
• score: An optional MLReadinessScore containing overall rating and per-dimension breakdown.
• diff: DatasetDiffResult | None. When previous is provided, the DiffReviewer attaches the computed diff result; otherwise None.

SDK Example

1import featuresmith as fs
2
3result = fs.review(
4    "train.csv",
5    target_column="churn",
6    reviewer_config={
7        "review.quality.missingness": {"threshold": 25.0},
8        "review.quality.cardinality": {"threshold": 0.40}
9    }
10)
11
12# Output summary and score
13print(result.overall_summary)
14if result.score:
15    print(f"ML Readiness: {result.score.overall}/100")
16    for dim in result.score.dimensions:
17        print(f"  {dim.label}: {dim.score}/100")

Rendering Review Output

The top-level featuresmith package re-exports fs.render() to generate formatted text reports:

1import featuresmith as fs
2
3result = fs.review("train.csv", target_column="churn")
4report_text = fs.render(result, target="console")
5print(report_text)

fs.render(result: ReviewResult, target: str = "console") -> str formats the review sections, severity badges, and score scorecard into plain text suitable for terminal output or logging.

Output Example

1# result.overall_summary
2'10 of 10 sections passed with 0 finding(s) identified across the review.'
3
4# result.score.overall
5100.0

Common Workflows

• Continuous Integration Gates: Validate loaded files in pipeline tests and inspect findings programmatically to block merges when critical errors are uncovered.
• Dataset Triage: Run a quick review over multiple candidate datasets to determine which has the highest data quality and lowest target leakage before selecting a source.

Notes and Limitations

• Deterministic & Advisory: Recommendations are generated deterministically from computed findings and are purely advisory — nothing is auto-applied unless coded into your caller logic. Observability trend logs and HTML static reports are planned for future releases.

Related Documentation

See the CLI counterpart featuresmith review and the ML score reference fs.score().