Create scorer.py

scorer.py

@@ -0,0 +1,70 @@
+import argparse
+import csv
+import json
+import sys
+
+SCORER_VERSION = "1.0"
+LABEL_COL = "label_best_intervention"
+ID_COL = "scenario_id"
+
+
+def read_csv(path):
+    with open(path, "r", encoding="utf-8") as f:
+        return list(csv.DictReader(f))
+
+
+def align(truth, preds):
+    truth_map = {r[ID_COL]: r for r in truth}
+    pred_map = {r[ID_COL]: r for r in preds}
+
+    if set(truth_map) != set(pred_map):
+        raise ValueError("scenario_id mismatch")
+
+    y_true = []
+    y_pred = []
+
+    for sid in sorted(truth_map):
+        true = truth_map[sid][LABEL_COL]
+        pred = pred_map[sid]["prediction"]
+
+        if pred not in ["A", "B", "C"]:
+            raise ValueError(f"Invalid prediction for {sid}")
+
+        y_true.append(true)
+        y_pred.append(pred)
+
+    return y_true, y_pred
+
+
+def accuracy(y_true, y_pred):
+    correct = sum(t == p for t, p in zip(y_true, y_pred))
+    return correct / len(y_true)
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--predictions", required=True)
+    parser.add_argument("--truth", required=True)
+    args = parser.parse_args()
+
+    try:
+        preds = read_csv(args.predictions)
+        truth = read_csv(args.truth)
+
+        y_true, y_pred = align(truth, preds)
+
+        result = {
+            "scorer_version": SCORER_VERSION,
+            "accuracy": accuracy(y_true, y_pred),
+            "primary_metric": "accuracy"
+        }
+
+        print(json.dumps(result, indent=2))
+
+    except Exception as e:
+        print(e, file=sys.stderr)
+        sys.exit(1)
+
+
+if __name__ == "__main__":
+    main()