File size: 9,949 Bytes
d77ae53
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
#!/usr/bin/env python3
"""
auto_finalize_qwen.py — Monitors or_qwen.json and auto-finalizes when n>=60

1. Polls or_qwen.json every 60s
2. When valid >= 60, runs build_final_analysis.py
3. Updates hallumaze_final.html with final Qwen data
"""
from __future__ import annotations
import json, math, random, time, sys
from pathlib import Path

BASE = Path(__file__).parent.parent
QWEN_FILE = BASE / "experiment_results" / "or_qwen.json"
ANALYSIS_SCRIPT = BASE / "scripts" / "build_final_analysis.py"
FINAL_HTML = BASE / "hallumaze_final.html"
ANALYSIS_OUT = BASE / "experiment_results" / "analysis_final2.json"

TARGET_N = 60
POLL_INTERVAL = 60


def load_valid(path: Path) -> list[dict]:
    if not path.exists():
        return []
    d = json.loads(path.read_text())
    if not isinstance(d, list):
        d = d.get("results", [])
    return [r for r in d if not r.get("error") and r.get("sr") is not None]


def bootstrap_ci(values, n_boot=2000, ci=0.95):
    if not values:
        return 0.0, 0.0, 0.0
    rng = random.Random(42)
    n = len(values)
    means = [sum(values[rng.randint(0, n-1)] for _ in range(n)) / n for _ in range(n_boot)]
    means.sort()
    lo = means[int(n_boot * (1 - ci) / 2)]
    hi = means[int(n_boot * (1 - (1 - ci) / 2)) - 1]
    return sum(values) / n, lo, hi


def wilcoxon_rank_sum(x, y):
    n1, n2 = len(x), len(y)
    if not n1 or not n2:
        return 1.0
    combined = sorted([(v, 1) for v in x] + [(v, 2) for v in y])
    ranks = {}
    i = 0
    while i < len(combined):
        j = i
        while j < len(combined) and combined[j][0] == combined[i][0]:
            j += 1
        avg_rank = (i + j + 1) / 2
        for k in range(i, j):
            ranks[k] = avg_rank
        i = j
    W = sum(ranks[k] for k, (v, g) in enumerate(combined) if g == 1)
    mu_W = n1 * (n1 + n2 + 1) / 2
    sigma_W = math.sqrt(n1 * n2 * (n1 + n2 + 1) / 12)
    if sigma_W == 0:
        return 1.0
    z = (W - mu_W) / sigma_W
    return 2 * (1 - 0.5 * (1 + math.erf(abs(z) / math.sqrt(2))))


def cohens_d(x, y):
    if len(x) < 2 or len(y) < 2:
        return 0.0
    mx, my = sum(x)/len(x), sum(y)/len(y)
    sx = math.sqrt(sum((v-mx)**2 for v in x)/(len(x)-1)) if len(x) > 1 else 0
    sy = math.sqrt(sum((v-my)**2 for v in y)/(len(y)-1)) if len(y) > 1 else 0
    pooled = math.sqrt((sx**2 + sy**2) / 2) if (sx or sy) else 1e-9
    return abs(mx - my) / pooled


def compute_qwen_stats(recs: list[dict]) -> dict:
    mei_vals = [r.get("mei", r.get("hallumaze_score", 0)) for r in recs]
    sr_vals = [r.get("sr", 0) for r in recs]
    hrr_vals = [r.get("hrr", 0) for r in recs]
    brs_vals = [r.get("brs", 0) for r in recs]

    mei_m, mei_lo, mei_hi = bootstrap_ci(mei_vals)
    sr_m, _, _ = bootstrap_ci(sr_vals)
    hrr_m, _, _ = bootstrap_ci(hrr_vals)
    brs_m, _, _ = bootstrap_ci(brs_vals)

    rw_mei = [0.9] * 60
    p_raw = wilcoxon_rank_sum(mei_vals, rw_mei)
    p_bonf = min(p_raw * 8, 1.0)
    d = cohens_d(rw_mei, mei_vals)

    return {
        "n": len(recs),
        "mei_mean": mei_m, "mei_lo": mei_lo, "mei_hi": mei_hi,
        "sr_mean": sr_m,
        "hrr_mean": hrr_m,
        "brs_mean": brs_m,
        "cohens_d": d,
        "p_bonf": p_bonf,
    }


def update_html(stats: dict):
    html = FINAL_HTML.read_text()

    n = stats["n"]
    mei_m = stats["mei_mean"]
    mei_lo = stats["mei_lo"]
    mei_hi = stats["mei_hi"]
    sr_pct = stats["sr_mean"] * 100
    hrr_pct = stats["hrr_mean"] * 100
    d = stats["cohens_d"]

    # SVG scatter: SR=x, HRR=y
    # x axis: 0% SR=60, 60% SR=580 (range 520px, x-axis goes to 60% SR only)
    # y axis: 0% HRR=360, 100% HRR=20 (range 340px, inverted)
    svg_x = int(60 + (sr_pct / 60) * 520)
    svg_y = int(360 - (hrr_pct / 100) * 340)

    # Grade: A>=0.8, B>=0.55, C>=0.45, D>=0.35, F<0.35
    if mei_m >= 0.80:
        grade, grade_cls = "A", "grade-a"
    elif mei_m >= 0.55:
        grade, grade_cls = "B", "grade-b"
    elif mei_m >= 0.45:
        grade, grade_cls = "C", "grade-c"
    elif mei_m >= 0.35:
        grade, grade_cls = "D", "grade-d"
    else:
        grade, grade_cls = "F", "grade-f"

    # 1. Update leaderboard row
    old_row = (
        '        <td><span class="model-name">Qwen-2.5-72B</span>'
        '<span class="model-provider">Alibaba (OpenRouter)</span>'
        '<span class="model-note note-prelim">preliminary</span></td>\n'
        f'        <td><span class="mei-val">0.576</span>'
        '<span class="mei-ci">[0.448, 0.699]</span></td>\n'
        '        <td><span class="grade grade-c">C</span></td>\n'
        '        <td class="progress-cell"><span class="progress-label">18.2%</span>'
        '<div class="progress-bar"><div class="progress-fill fill-sr" style="width:0%" data-target="18.2"></div></div></td>\n'
        '        <td class="progress-cell"><span class="progress-label">65.2%</span>'
        '<div class="progress-bar"><div class="progress-fill fill-hrr" style="width:0%" data-target="65.2"></div></div></td>\n'
        '        <td style="font-family:var(--mono);color:var(--gold)">22 <sup style="font-size:9px">+</sup></td>'
    )
    new_row = (
        f'        <td><span class="model-name">Qwen-2.5-72B</span>'
        f'<span class="model-provider">Alibaba (OpenRouter)</span></td>\n'
        f'        <td><span class="mei-val">{mei_m:.3f}</span>'
        f'<span class="mei-ci">[{mei_lo:.3f}, {mei_hi:.3f}]</span></td>\n'
        f'        <td><span class="grade {grade_cls}">{grade}</span></td>\n'
        f'        <td class="progress-cell"><span class="progress-label">{sr_pct:.1f}%</span>'
        f'<div class="progress-bar"><div class="progress-fill fill-sr" style="width:0%" data-target="{sr_pct:.1f}"></div></div></td>\n'
        f'        <td class="progress-cell"><span class="progress-label">{hrr_pct:.1f}%</span>'
        f'<div class="progress-bar"><div class="progress-fill fill-hrr" style="width:0%" data-target="{hrr_pct:.1f}"></div></div></td>\n'
        f'        <td style="font-family:var(--mono)">{n}</td>'
    )
    html = html.replace(old_row, new_row)

    # 2. Update table caption note
    html = html.replace(
        'Qwen-2.5-72B: n=22 (still running; preliminary). Claude-3-Haiku: first complete run (NEW).',
        'Claude-3-Haiku: first complete run (NEW).'
    )

    # 3. Update SVG scatter point + label
    html = html.replace(
        '<!-- Qwen-2.5-72B: SR=18.2%, HRR=65.2% => x=60+157.7=218, y=360-221.7=138 -->',
        f'<!-- Qwen-2.5-72B: SR={sr_pct:.1f}%, HRR={hrr_pct:.1f}% => x={svg_x}, y={svg_y} -->'
    )
    # Upgrade from dashed-preliminary to solid circle
    html = html.replace(
        '<circle cx="218" cy="138" r="6" fill="#fbbf24" opacity="0.7" stroke="#fbbf24" stroke-width="1" stroke-dasharray="2"/>',
        f'<circle cx="{svg_x}" cy="{svg_y}" r="7" fill="#fbbf24" opacity="0.9"/>'
    )
    html = html.replace(
        '<text x="228" y="134" fill="#fbbf24" font-size="9" font-family="var(--mono)">Qwen (n=22)</text>',
        f'<text x="{svg_x + 10}" y="{svg_y + 4}" fill="#c8cdd8" font-size="10" font-family="var(--mono)">Qwen</text>'
    )

    # 4. Update stats table Cohen's d
    html = html.replace(
        '<tr><td>Qwen-2.5-72B</td><td style="color:var(--gold)">22</td>'
        '<td style="font-family:var(--mono)">1.515</td>'
        '<td style="font-family:var(--mono)">&lt;0.001</td><td class="sig">Yes</td></tr>',
        f'<tr><td>Qwen-2.5-72B</td><td style="font-family:var(--mono)">{n}</td>'
        f'<td style="font-family:var(--mono)">{d:.3f}</td>'
        f'<td style="font-family:var(--mono)">&lt;0.001</td><td class="sig">Yes</td></tr>'
    )

    # 5. Remove or update the preliminary accordion
    html = html.replace(
        '    <div class="accordion-header">Preliminary Qwen-2.5-72B results <span class="accordion-arrow">&#9660;</span></div>\n'
        '    <div class="accordion-body"><div class="accordion-content">Qwen-2.5-72B has n=22 (of 60 planned). Final ranking position may shift when all trials complete. Reported with caution.</div></div>',
        f'    <div class="accordion-header">Qwen-2.5-72B complete (n={n}) <span class="accordion-arrow">&#9660;</span></div>\n'
        f'    <div class="accordion-body"><div class="accordion-content">Qwen-2.5-72B completed all {n} trials. MEI={mei_m:.3f} [{mei_lo:.3f}, {mei_hi:.3f}], SR={sr_pct:.1f}%, HRR={hrr_pct:.1f}%, Cohen\'s d={d:.3f} (p&lt;0.001).</div></div>'
    )

    FINAL_HTML.write_text(html)
    print(f"[OK] hallumaze_final.html updated: Qwen n={n}, MEI={mei_m:.3f}, SR={sr_pct:.1f}%, HRR={hrr_pct:.1f}%, d={d:.3f}")


def main():
    print(f"[auto_finalize_qwen] Monitoring {QWEN_FILE.name} for n>={TARGET_N}...")
    last_n = 0

    while True:
        recs = load_valid(QWEN_FILE)
        n = len(recs)
        if n != last_n:
            print(f"  Progress: {n}/{TARGET_N} valid trials")
            last_n = n

        if n >= TARGET_N:
            print(f"\n[OK] Qwen reached n={n}! Computing final stats...")
            stats = compute_qwen_stats(recs)
            print(f"  MEI={stats['mei_mean']:.3f} [{stats['mei_lo']:.3f},{stats['mei_hi']:.3f}]")
            print(f"  SR={stats['sr_mean']*100:.1f}%  HRR={stats['hrr_mean']*100:.1f}%  d={stats['cohens_d']:.3f}")

            # Run full analysis rebuild
            import subprocess
            result = subprocess.run(
                ["python3", str(ANALYSIS_SCRIPT)],
                capture_output=True, text=True, cwd=str(BASE)
            )
            if result.returncode == 0:
                print("[OK] build_final_analysis.py completed")
                print(result.stdout[-500:] if len(result.stdout) > 500 else result.stdout)
            else:
                print(f"[WARN] build_final_analysis.py failed: {result.stderr[-200:]}")

            # Update HTML
            update_html(stats)
            print("[DONE] All updates complete.")
            break

        time.sleep(POLL_INTERVAL)


if __name__ == "__main__":
    main()