Prioritize recall in failure-driven sampling
GeoIntel release gates / Compile, test, contracts and builds (push) Canceled after 0s
GeoIntel release gates / Python and npm vulnerability policy (push) Canceled after 0s
GeoIntel release gates / GIS image, SBOM and container scan (push) Canceled after 0s

This commit is contained in:
Jens
2026-07-29 23:10:19 +02:00
parent be5cbb855a
commit b8e40ba7a8
2 changed files with 44 additions and 2 deletions
@@ -163,11 +163,39 @@ def test_sampling_targets_failed_calibration_contexts_without_using_protected_ti
)
assert paths.count(str(Path("/tmp/industry-pos.png").resolve())) == 5
assert paths.count(str(Path("/tmp/industry-neg.png").resolve())) == 6
assert paths.count(str(Path("/tmp/industry-neg.png").resolve())) == 1
assert paths.count(str(Path("/tmp/suburban-pos.png").resolve())) == 3
assert not any("protected" in path for path in paths)
assert metadata["weak_recall_contexts"] == ["flanders:industrial"]
assert metadata["weak_precision_contexts"] == ["flanders:industrial"]
assert metadata["recall_dominant_regions"] == ["flanders"]
def test_recall_dominance_does_not_suppress_negatives_when_background_gate_failed() -> None:
manifest = {"samples": [
{"sample_slug": "positive", "split": "train", "region": "flanders", "context": "industrial"},
{"sample_slug": "negative", "split": "train", "region": "flanders", "context": "industrial-hard-negative"},
]}
summary = {"tiles": [
{"sample_slug": "positive", "split": "train", "label_count": 1, "image_path": "/tmp/positive.png"},
{"sample_slug": "negative", "split": "train", "label_count": 0, "image_path": "/tmp/negative.png"},
]}
assessment = {
"status": "continue_training_loop",
"gates": {"min_region_f1": .45, "min_region_precision": .5, "min_region_recall": .4,
"max_pure_empty_false_positives": 0},
"calibration": {"regions": {
"flanders": {"f1": .25, "precision": .4, "recall": .2},
}},
"background": {"pure_empty_false_positives": 1},
}
paths, metadata = MODULE.build_sampling(
summary=summary, manifest=manifest, assessment=assessment, max_region_share=1.0,
)
assert paths.count(str(Path("/tmp/negative.png").resolve())) == 4
assert metadata["recall_dominant_regions"] == []
def test_region_cap_drops_only_repeats_and_preserves_every_unique_tile() -> None: