Create scorer.py

scorer.py

@@ -0,0 +1,45 @@
+from dataclasses import dataclass
+from typing import Dict, Any, List
+import json
+
+@dataclass
+class ScoreResult:
+    score: float
+    details: Dict[str, Any]
+
+BANDS = {"low", "medium", "high"}
+
+def score(sample: Dict[str, Any], prediction: str) -> ScoreResult:
+    try:
+        pred = json.loads(prediction)
+        cs = float(pred.get("coherence_score", -1))
+        band = str(pred.get("risk_band", "")).strip().lower()
+    except Exception:
+        return ScoreResult(0.0, {"error":"parse_fail","id":sample.get("id")})
+
+    true_cs_raw = sample.get("coherence_score", "")
+    true_band_raw = sample.get("stochastic_risk_band", "")
+
+    try:
+        true_cs = float(true_cs_raw) if true_cs_raw not in ("", None) else None
+    except Exception:
+        true_cs = None
+
+    true_band = str(true_band_raw).strip().lower() if true_band_raw not in ("", None) else ""
+
+    # format-only if no ground truth
+    if true_cs is None or true_band == "":
+        ok = (0.0 <= cs <= 1.0) and (band in BANDS)
+        return ScoreResult(1.0 if ok else 0.0, {"mode":"format_only","id":sample.get("id")})
+
+    cs_err = abs(true_cs - cs)
+    cs_score = max(0.0, 1.0 - cs_err)
+    band_score = 1.0 if band == true_band else 0.0
+
+    total = 0.65 * cs_score + 0.35 * band_score
+    return ScoreResult(total, {"id":sample.get("id"),"cs":cs,"true_cs":true_cs,"band":band,"true_band":true_band})
+
+def aggregate(results: List[ScoreResult]) -> Dict[str, Any]:
+    if not results:
+        return {"mean":0.0,"n":0}
+    return {"mean": sum(r.score for r in results)/len(results), "n": len(results)}