scorer.py

import csv
import json
import re
from typing import Dict, Tuple

ALLOWED_ERROR_TYPES = {
    "no_error",
    "low_confidence_region_overtrust",
    "interdomain_orientation_overclaim",
    "disorder_as_structure",
    "loop_position_overtrust",
    "complex_negation_from_monomer",
    "interface_overconfidence",
    "cofactor_absence_misread",
    "low_complexity_as_domain",
}

def _norm(s: str) -> str:
    return re.sub(r"\s+", " ", (s or "").strip().lower())

def _token_set(s: str) -> set:
    s = _norm(s)
    s = re.sub(r"[^a-z0-9\s]", " ", s)
    return {t for t in s.split(" ") if t}

def _jaccard(a: str, b: str) -> float:
    ta = _token_set(a)
    tb = _token_set(b)
    if not ta and not tb:
        return 1.0
    if not ta or not tb:
        return 0.0
    return len(ta & tb) / len(ta | tb)

def _extract_json(text: str) -> Dict:
    if text is None:
        return {}
    text = text.strip()
    try:
        return json.loads(text)
    except Exception:
        pass
    m = re.search(r"\{.*\}", text, flags=re.DOTALL)
    if not m:
        return {}
    try:
        return json.loads(m.group(0))
    except Exception:
        return {}

def load_refs(test_csv_path: str) -> Dict[str, Tuple[str, str, str]]:
    refs = {}
    with open(test_csv_path, newline="", encoding="utf-8") as f:
        r = csv.DictReader(f)
        for row in r:
            refs[row["id"]] = (
                _norm(row["gold_misinterpretation"]),
                row["gold_error_type"].strip(),
                row["gold_correction"].strip(),
            )
    return refs

def score_predictions(predictions_path: str, test_csv_path: str) -> Dict[str, float]:
    refs = load_refs(test_csv_path)

    n = 0
    mis_hits = 0
    type_hits = 0
    corr_sim_sum = 0.0
    format_hits = 0

    with open(predictions_path, encoding="utf-8") as f:
        preds = [json.loads(line) for line in f if line.strip()]

    for item in preds:
        ex_id = item.get("id", "")
        raw = item.get("prediction", "")
        if ex_id not in refs:
            continue

        n += 1
        parsed = _extract_json(raw)

        pred_mis = _norm(parsed.get("misinterpretation", ""))
        pred_type = (parsed.get("error_type") or "").strip()
        pred_corr = (parsed.get("correction") or "").strip()

        gold_mis, gold_type, gold_corr = refs[ex_id]

        mis_hits += 1 if pred_mis == gold_mis else 0
        type_hits += 1 if pred_type == gold_type else 0
        corr_sim_sum += _jaccard(pred_corr, gold_corr)

        has_keys = pred_mis in {"yes", "no"} and pred_type != "" and pred_corr != ""
        type_allowed = pred_type in ALLOWED_ERROR_TYPES
        consistency = (pred_mis == "no" and pred_type == "no_error") or (pred_mis == "yes" and pred_type != "no_error")
        format_hits += 1 if (has_keys and type_allowed and consistency) else 0

    if n == 0:
        return {
            "final_score": 0.0,
            "misinterpretation_accuracy": 0.0,
            "error_type_accuracy": 0.0,
            "correction_similarity": 0.0,
            "format_pass_rate": 0.0,
            "n_scored": 0.0,
        }

    mis_acc = mis_hits / n
    type_acc = type_hits / n
    corr_sim = corr_sim_sum / n
    fmt = format_hits / n

    final = 0.4 * mis_acc + 0.3 * type_acc + 0.2 * corr_sim + 0.1 * fmt

    return {
        "final_score": float(final),
        "misinterpretation_accuracy": float(mis_acc),
        "error_type_accuracy": float(type_acc),
        "correction_similarity": float(corr_sim),
        "format_pass_rate": float(fmt),
        "n_scored": float(n),
    }

if __name__ == "__main__":
    import argparse
    p = argparse.ArgumentParser()
    p.add_argument("--predictions", required=True)
    p.add_argument("--test_csv", required=True)
    args = p.parse_args()
    print(json.dumps(score_predictions(args.predictions, args.test_csv), indent=2))