Evaluate wearer floor changes by event

test_evaluate_floor_state_changes.py

@@ -9,6 +9,7 @@ import numpy as np
 from evaluate_floor_state_changes import (
     NUM_CLASSES,
     NUM_HORIZONS,
+    evaluate_event_horizon,
     evaluate_horizon,
     load_arrays,
     parse_runs,
@@ -21,6 +22,67 @@ def activity(wearer: int, others: int) -> list[int]:
 
 
 class FloorStateMetricTests(unittest.TestCase):
+    def test_event_level_counts_each_boundary_once(self) -> None:
+        target = np.asarray(
+            [
+                activity(0, 1),  # other offset follows
+                activity(0, 0),
+                activity(1, 0),  # wearer responds within 0.4s
+                activity(1, 0),  # other entry follows
+                activity(1, 1),
+                activity(0, 1),  # wearer yields within 0.4s
+                activity(0, 1),
+                activity(0, 1),
+                activity(0, 0),  # short pause, no wearer response
+                activity(0, 1),
+                activity(1, 0),  # another other entry follows
+                activity(1, 1),
+                activity(1, 1),  # wearer continues
+            ],
+            dtype=np.float32,
+        )
+        prediction = np.full(
+            (len(target), NUM_CLASSES, NUM_HORIZONS), 0.5, dtype=np.float32
+        )
+        # Fixed source row 0: response score max(0.2, 0.9).
+        prediction[1, 1, 0] = 0.2
+        prediction[2, 1, 1] = 0.9
+        # Fixed source row 7: no-response score max(0.1, 0.2).
+        prediction[8, 1, 0] = 0.1
+        prediction[9, 1, 1] = 0.2
+        # Fixed source row 3: yield score max(0.2, 0.9).
+        prediction[4, 1, 0] = 0.8
+        prediction[5, 1, 1] = 0.1
+        # Fixed source row 10: continue score max(0.1, 0.2).
+        prediction[11, 1, 0] = 0.9
+        prediction[12, 1, 1] = 0.8
+        valid = np.ones((len(target), NUM_HORIZONS), dtype=bool)
+        session_ids = np.zeros(len(target), dtype=np.int32)
+        session_ids[10:] = 1
+
+        result = evaluate_event_horizon(
+            target,
+            prediction,
+            valid,
+            session_ids,
+            1,
+            [0.8],
+            midturn_pause_max_steps=10,
+        )
+        response = result["other_offset_to_wearer_response_event"]
+        self.assertEqual(response["num_positive"], 1)
+        self.assertEqual(response["num_negative"], 1)
+        self.assertEqual(response["num_short_resume_hard_negative"], 1)
+        self.assertAlmostEqual(response["average_precision"], 1.0)
+        self.assertAlmostEqual(
+            response["operating_points"][0]["negative_false_positive_rate"],
+            0.0,
+        )
+        yield_metric = result["other_onset_to_wearer_yield_event"]
+        self.assertEqual(yield_metric["num_positive"], 1)
+        self.assertEqual(yield_metric["num_negative"], 1)
+        self.assertAlmostEqual(yield_metric["average_precision"], 1.0)
+
     def test_horizon_one_proxies_and_session_boundary(self) -> None:
         # t0->t1: EOT-response; t3->t4: HOLD; t4->t5: early entry;
         # t6->t7: other-entry yield; t8->t9: other-entry continue.