Fix F1 for disjoint track identity sets

src/pyrecest/utils/track_evaluation/__init__.py

@@ -3,6 +3,8 @@
 
 import numpy as np
 
+from ._identity_f1 import patch_identity_f1
+
 d = runpy.run_path(
     str(Path(__file__).resolve().parents[1] / "track_evaluation.py"), run_name=__name__
 )
@@ -20,6 +22,7 @@ def f(v):
 
 
 d["_optional_int_candidate"] = f
+patch_identity_f1(d)
 for n in d["__all__"]:
     globals()[n] = d[n]
 __all__ = d["__all__"]

src/pyrecest/utils/track_evaluation/_identity_f1.py

@@ -0,0 +1,39 @@
+"""Local identity-set F1 fixes for track-evaluation exports."""
+
+from __future__ import annotations
+
+from typing import Any
+
+
+def patch_identity_f1(namespace: dict[str, Any]) -> None:
+    """Install an identity-set scorer with a zero-valued disjoint F1."""
+
+    safe_ratio = namespace["_safe_ratio"]
+    zero_ratio = namespace["_zero_ratio"]
+
+    def score_identity_sets(
+        predicted: set[Any],
+        reference: set[Any],
+        *,
+        prefix: str,
+        predicted_total_name: str,
+        reference_total_name: str,
+    ) -> dict[str, float | int]:
+        true_positives = len(predicted & reference)
+        false_positives = len(predicted - reference)
+        false_negatives = len(reference - predicted)
+        precision = safe_ratio(true_positives, true_positives + false_positives)
+        recall = safe_ratio(true_positives, true_positives + false_negatives)
+        f1 = zero_ratio(2.0 * precision * recall, precision + recall)
+        return {
+            f"{prefix}_true_positives": true_positives,
+            f"{prefix}_false_positives": false_positives,
+            f"{prefix}_false_negatives": false_negatives,
+            f"{prefix}_precision": precision,
+            f"{prefix}_recall": recall,
+            f"{prefix}_f1": f1,
+            predicted_total_name: len(predicted),
+            reference_total_name: len(reference),
+        }
+
+    namespace["_score_identity_sets"] = score_identity_sets

tests/test_track_evaluation_identity_f1.py

@@ -0,0 +1,31 @@
+from pyrecest.utils.track_evaluation import score_complete_tracks, score_track_links
+
+
+def test_disjoint_track_identity_sets_have_zero_f1():
+    complete_scores = score_complete_tracks([[0, 1]], [[2, 3]])
+
+    assert complete_scores["complete_track_true_positives"] == 0
+    assert complete_scores["complete_track_false_positives"] == 1
+    assert complete_scores["complete_track_false_negatives"] == 1
+    assert complete_scores["complete_track_precision"] == 0.0
+    assert complete_scores["complete_track_recall"] == 0.0
+    assert complete_scores["complete_track_f1"] == 0.0
+
+    link_scores = score_track_links([[0, 1]], [[2, 3]])
+
+    assert link_scores["track_link_true_positives"] == 0
+    assert link_scores["track_link_false_positives"] == 1
+    assert link_scores["track_link_false_negatives"] == 1
+    assert link_scores["track_link_precision"] == 0.0
+    assert link_scores["track_link_recall"] == 0.0
+    assert link_scores["track_link_f1"] == 0.0
+
+
+def test_empty_track_identity_sets_keep_perfect_precision_recall_and_f1():
+    scores = score_complete_tracks([[None, None]], [[None, None]])
+
+    assert scores["complete_tracks"] == 0
+    assert scores["reference_complete_tracks"] == 0
+    assert scores["complete_track_precision"] == 1.0
+    assert scores["complete_track_recall"] == 1.0
+    assert scores["complete_track_f1"] == 1.0