File size: 2,106 Bytes
016dd6c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
import csv
import math

ALLOWED = {"aligned", "partial", "divergent"}

def parse_prediction(text):
    out = {}
    if not text:
        return out
    parts = [p.strip() for p in text.split(";") if p.strip()]
    for p in parts:
        if "=" in p:
            k, v = p.split("=", 1)
            out[k.strip()] = v.strip()
    return out

def clamp01(x):
    return max(0.0, min(1.0, x))

def score(pred_file, gold_file):
    preds = {}
    with open(pred_file, newline="") as f:
        r = csv.DictReader(f)
        for row in r:
            preds[row["id"]] = parse_prediction(row.get("prediction",""))

    gold = {}
    with open(gold_file, newline="") as f:
        r = csv.DictReader(f)
        for row in r:
            gold[row["id"]] = row

    n = 0
    acc = 0
    mse = 0.0
    div_hit = 0

    for k, g in gold.items():
        if k not in preds:
            continue
        p = preds[k]
        n += 1

        gl = (g.get("alignment_label","") or "").strip().lower()
        pl = (p.get("alignment_label","") or "").strip().lower()
        if pl == gl and gl in ALLOWED:
            acc += 1

        try:
            gs = float(g.get("alignment_score",""))
            ps = float(p.get("alignment_score",""))
            ps = clamp01(ps)
            mse += (gs - ps) ** 2
        except:
            pass

        # simple divergence_points presence check (should not be empty for divergent/partial)
        dp = (p.get("divergence_points","") or "").strip()
        if gl in {"divergent", "partial"} and len(dp) >= 8:
            div_hit += 1
        if gl == "aligned" and (dp.lower() in {"none", ""}):
            div_hit += 1

    label_acc = acc / n if n else 0.0
    rmse = math.sqrt(mse / n) if n else 0.0
    div_score = div_hit / n if n else 0.0

    # weighted total score
    total = 0.55 * label_acc + 0.30 * (1.0 - rmse) + 0.15 * div_score
    total = clamp01(total)

    print("label_accuracy:", round(label_acc, 3))
    print("alignment_score_RMSE:", round(rmse, 3))
    print("divergence_points_quality:", round(div_score, 3))
    print("total_score:", round(total, 3))