Evaluation

Evaluate predicted results and inspect scoring metadata

Evaluation

This example demonstrates aligner.evaluate(...), used to classify predicted alignments with evaluation metadata.

What you learn

1. How to run evaluation on a set of relevant_distances.
2. How to retrieve EVALUATED_PREDICTIONS as GeoJSON.
3. How to interpret key output fields such as:
   • evaluation status
   • prediction score
   • selected relevant distance

Why evaluation is useful

Evaluation helps when you need automated quality interpretation for batch processing, reporting, or downstream decision logic.

Notes

• The example uses FullReferenceStrategy.PREFER_FULL_REFERENCE.
• GRB data is loaded at runtime, so network access is required.

import numpy as np

from brdr.aligner import Aligner
from brdr.be.grb.enums import GRBType
from brdr.be.grb.loader import GRBActualLoader
from brdr.constants import (
    EVALUATION_FIELD_NAME,
    RELEVANT_DISTANCE_FIELD_NAME,
    PREDICTION_SCORE,
)
from brdr.enums import AlignerResultType, FullReferenceStrategy
from brdr.geometry_utils import geom_from_wkt
from brdr.loader import DictLoader

if __name__ == "__main__":
    """Example: evaluate predictions and print score/distance summary."""

    aligner = Aligner(crs="EPSG:31370")
    relevant_distances = np.arange(0, 400, 10, dtype=int) / 100

    input_geometry_wkt = "POLYGON ((174099.53572337731020525 179375.43192003236617893, 174116.23475504826637916 179372.81056040961993858, 174109.5357249012158718 179324.16977629828033969, 174094.68135370552772656 179324.75230065890355036, 174099.53572337731020525 179375.43192003236617893))"
    aligner.load_thematic_data(
        DictLoader(data_dict={"id_1": geom_from_wkt(input_geometry_wkt)})
    )

    aligner.load_reference_data(
        GRBActualLoader(grb_type=GRBType.ADP, partition=1000, aligner=aligner)
    )

    aligner_result = aligner.evaluate(
        relevant_distances=relevant_distances,
        thematic_ids=None,
        metadata_field=None,
        full_reference_strategy=FullReferenceStrategy.PREFER_FULL_REFERENCE,
    )

    geojson_results = aligner_result.get_results_as_geojson(
        result_type=AlignerResultType.EVALUATED_PREDICTIONS,
        add_metadata=True,
        add_original_attributes=True,
        aligner=aligner,
    )

    for feature in geojson_results["result"]["features"]:
        print(
            feature["properties"][aligner.thematic_data.id_fieldname]
            + ": "
            + feature["properties"][EVALUATION_FIELD_NAME]
            + " - score "
            + str(feature["properties"][PREDICTION_SCORE])
            + " - distance "
            + str(feature["properties"][RELEVANT_DISTANCE_FIELD_NAME])
        )

id_1: prediction_unique_full - score 100 - distance 1.4

Figure 1