geointel/scripts/run_golden_qa_benchmark.py

from __future__ import annotations

import argparse
import json
import sys
from pathlib import Path
from uuid import uuid4


ROOT = Path(__file__).resolve().parents[1]
BACKEND_ROOT = ROOT / "backend"
if str(BACKEND_ROOT) not in sys.path:
    sys.path.insert(0, str(BACKEND_ROOT))

from app.models import Area, Dataset, Metric, QualityCheck  # noqa: E402
from app.services.qa_service import QaService  # noqa: E402
from app.services.quality_service import QualityService  # noqa: E402


class BenchmarkSession:
    def __init__(self, datasets: list[Dataset]) -> None:
        self.datasets = {dataset.id: dataset for dataset in datasets}
        self.added: list[object] = []
        self.commits = 0
        self.refreshes: list[object] = []

    def get(self, model, item_id):
        if model.__name__ == "Dataset":
            return self.datasets.get(item_id)
        if model.__name__ == "Area":
            return None
        return None

    def add(self, item) -> None:
        self.added.append(item)

    def commit(self) -> None:
        self.commits += 1

    def refresh(self, item) -> None:
        self.refreshes.append(item)


def _load_expected() -> dict:
    return json.loads((ROOT / "fixtures" / "golden" / "expected_qa_metrics.json").read_text(encoding="utf-8"))


def _dataset(dataset_id, project_id, name: str, path: Path, *, role: str) -> Dataset:
    return Dataset(
        id=dataset_id,
        project_id=project_id,
        name=name,
        dataset_type="vector",
        source="golden_fixture",
        dataset_role=role,
        source_name="fixture",
        reference_layer_name="buildings" if role == "reference" else None,
        storage_path=str(path),
        crs="EPSG:4326",
        metadata_json={"crs_assumed": False},
        status="ready",
    )


def _assert_close(label: str, actual: float | int | None, expected: float | int, tolerance: float) -> None:
    if actual is None:
        raise AssertionError(f"{label} is None, expected {expected}")
    if abs(float(actual) - float(expected)) > tolerance:
        raise AssertionError(f"{label} drifted: actual={actual}, expected={expected}, tolerance={tolerance}")


def run_benchmark() -> dict:
    expected = _load_expected()
    project_id = uuid4()
    candidate_dataset_id = uuid4()
    reference_dataset_id = uuid4()
    candidate_path = ROOT / expected["candidate_fixture"]
    reference_path = ROOT / expected["reference_fixture"]
    tolerance = float(expected["tolerance"])

    session = BenchmarkSession(
        [
            _dataset(candidate_dataset_id, project_id, "golden_predicted_buildings.geojson", candidate_path, role="source"),
            _dataset(reference_dataset_id, project_id, "golden_reference_buildings.geojson", reference_path, role="reference"),
        ]
    )

    result = QaService.compare_candidate_with_reference(
        db=session,
        project_id=project_id,
        candidate_dataset_id=candidate_dataset_id,
        reference_dataset_id=reference_dataset_id,
        iou_threshold=float(expected["iou_threshold"]),
    )

    metrics = {
        "precision": result.precision,
        "recall": result.recall,
        "f1": result.f1_score,
        "mean_iou": result.mean_iou,
        "false_positive_count": result.false_positives,
        "false_negative_count": result.false_negatives,
    }

    _assert_close("candidate_feature_count", result.candidate_feature_count, expected["candidate_feature_count"], 0)
    _assert_close("reference_feature_count", result.reference_feature_count, expected["reference_feature_count"], 0)
    _assert_close("matches", result.matches, expected["matches"], 0)
    _assert_close("false_positive_count", result.false_positives, expected["false_positive_count"], 0)
    _assert_close("false_negative_count", result.false_negatives, expected["false_negative_count"], 0)
    _assert_close("precision", result.precision, expected["precision"], tolerance)
    _assert_close("recall", result.recall, expected["recall"], tolerance)
    _assert_close("f1", result.f1_score, expected["f1"], tolerance)
    _assert_close("mean_iou", result.mean_iou, expected["mean_iou"], tolerance)

    quality_check = QualityService.persist_quality_check(
        db=session,
        project_id=project_id,
        reference_dataset_id=reference_dataset_id,
        check_type="golden_candidate_vs_reference",
        status=result.status,
        score=result.f1_score,
        parameters={"iou_threshold": result.iou_threshold, "benchmark_id": expected["benchmark_id"]},
        findings={
            "matches": result.matches,
            "false_positives": result.false_positives,
            "false_negatives": result.false_negatives,
        },
        candidate_dataset_id=candidate_dataset_id,
        metrics=metrics,
    )
    persisted_metrics = [item for item in session.added if isinstance(item, Metric)]

    return {
        "status": "passed",
        "benchmark_id": expected["benchmark_id"],
        "metrics": metrics,
        "quality_check_id": str(quality_check.id),
        "persistence": {
            "quality_check_count": len([item for item in session.added if isinstance(item, QualityCheck)]),
            "metric_count": len(persisted_metrics),
            "metric_keys": [metric.metric_key for metric in persisted_metrics],
            "commit_count": session.commits,
        },
        "fixtures": {
            "candidate": expected["candidate_fixture"],
            "reference": expected["reference_fixture"],
        },
    }


def main() -> int:
    parser = argparse.ArgumentParser(description="Run GeoIntel golden QA/QC benchmark.")
    parser.add_argument("--json", action="store_true", help="Print machine-readable JSON only.")
    args = parser.parse_args()

    try:
        payload = run_benchmark()
    except Exception as exc:
        if args.json:
            print(json.dumps({"status": "failed", "error": str(exc)}, indent=2))
        else:
            print(f"Golden QA/QC benchmark failed: {exc}", file=sys.stderr)
        return 1

    if args.json:
        print(json.dumps(payload, indent=2, sort_keys=True))
    else:
        print("GeoIntel golden QA/QC benchmark passed")
        print(json.dumps(payload, indent=2, sort_keys=True))
    return 0


if __name__ == "__main__":
    raise SystemExit(main())