scorer.py

import csv
import json
import re

def normalize(s):
    return re.sub(r"\s+"," ",(s or "").strip().lower())

def token_set(s):
    s = normalize(s)
    s = re.sub(r"[^a-z0-9\s]"," ",s)
    return set([t for t in s.split(" ") if t])

def jaccard(a,b):
    ta, tb = token_set(a), token_set(b)
    if not ta or not tb:
        return 0.0
    return len(ta & tb)/len(ta | tb)

def load_refs(path):
    refs={}
    with open(path,encoding="utf-8") as f:
        r=csv.DictReader(f)
        for row in r:
            refs[row["id"]] = row["gold_necessary_action"]
    return refs

def score(pred_path,test_csv):
    refs=load_refs(test_csv)
    n=0
    correct=0
    sim_total=0

    with open(pred_path,encoding="utf-8") as f:
        for line in f:
            if not line.strip():
                continue
            obj=json.loads(line)
            ex_id=obj.get("id")
            pred=obj.get("prediction","")
            if ex_id not in refs:
                continue
            n+=1
            gold=refs[ex_id]
            if normalize(pred)==normalize(gold):
                correct+=1
            sim_total+=jaccard(pred,gold)

    if n==0:
        return {"final_score":0}

    acc=correct/n
    sim=sim_total/n
    final=0.6*acc+0.4*sim

    return {
        "final_score":final,
        "exact_accuracy":acc,
        "similarity":sim,
        "n":n
    }

if __name__=="__main__":
    import argparse
    p=argparse.ArgumentParser()
    p.add_argument("--predictions")
    p.add_argument("--test_csv")
    args=p.parse_args()
    print(score(args.predictions,args.test_csv))