Submetry-Lifted Relational Alignment / bench_report.json
Mechanism confirmed, baseline not beaten
1{
2 "bench_version": 1,
3 "track": "relational_graph_classification",
4 "model": "graphnet",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "best_cfg": {
9 "lr": 0.01
10 },
11 "sweep": [
12 {
13 "cfg": {
14 "lr": 0.003
15 },
16 "mean": 0.012500000302679837
17 },
18 {
19 "cfg": {
20 "lr": 0.01
21 },
22 "mean": 0.009375000256113708
23 },
24 {
25 "cfg": {
26 "lr": 0.03
27 },
28 "mean": 0.009375000256113708
29 }
30 ],
31 "full": {
32 "mean": 0.010156250325962901,
33 "std": 0.007614683392424964,
34 "per_seed": [
35 0.012500000186264515,
36 0.0062500000931322575,
37 0.0,
38 0.01875000074505806,
39 0.0062500000931322575,
40 0.01875000074505806,
41 0.0,
42 0.01875000074505806
43 ],
44 "n": 8,
45 "signature_variance": 4.222490175236095e-17
46 }
47 },
48 "idea": {
49 "per_seed": [
50 0.012500000186264515,
51 0.0062500000931322575,
52 0.0,
53 0.012500000186264515,
54 0.0062500000931322575,
55 0.01875000074505806,
56 0.0,
57 0.01875000074505806
58 ],
59 "mean": 0.009375000256113708,
60 "signature_variance": 1.260543858543509e-17,
61 "lr": 0.01
62 },
63 "comparison": {
64 "delta_mean": -0.0007812500698491931,
65 "idea_wins": 1,
66 "n_pairs": 8,
67 "per_seed_diffs": [
68 0.0,
69 0.0,
70 0.0,
71 -0.006250000558793545,
72 0.0,
73 0.0,
74 0.0,
75 0.0
76 ],
77 "p_value": 1.0,
78 "mde": 0.0018476564151933418,
79 "mde_rel_pct": 18.19230873494808,
80 "verdict": "no significant win",
81 "system_worked": false
82 },
83 "mechanism_signature": {
84 "prediction": "alignment should reduce permutation prediction variance",
85 "predicted_baseline_variance_not_available": false,
86 "observed_baseline_variance": 4.222490175236095e-17,
87 "observed_idea_variance": 1.260543858543509e-17,
88 "confirmed": true,
89 "sinkhorn_iterations": 35,
90 "alignment_epsilon": "0.1*median relational row cost"
91 },
92 "idea_sweep": [
93 {
94 "per_seed": [
95 0.01875000074505806,
96 0.0062500000931322575,
97 0.012500000186264515,
98 0.012500000186264515,
99 0.0062500000931322575,
100 0.01875000074505806,
101 0.0,
102 0.01875000074505806
103 ],
104 "mean": 0.011718750349245965,
105 "signature_variance": 9.958932715930935e-17,
106 "lr": 0.003
107 },
108 {
109 "per_seed": [
110 0.012500000186264515,
111 0.0062500000931322575,
112 0.0,
113 0.012500000186264515,
114 0.0062500000931322575,
115 0.01875000074505806,
116 0.0,
117 0.01875000074505806
118 ],
119 "mean": 0.009375000256113708,
120 "signature_variance": 1.260543858543509e-17,
121 "lr": 0.01
122 },
123 {
124 "per_seed": [
125 0.01875000074505806,
126 0.0062500000931322575,
127 0.0,
128 0.012500000186264515,
129 0.0,
130 0.01875000074505806,
131 0.0062500000931322575,
132 0.012500000186264515
133 ],
134 "mean": 0.009375000256113708,
135 "signature_variance": 1.946650362764039e-17,
136 "lr": 0.03
137 }
138 ],
139 "custom_track": {
140 "name": "relational_graph_classification",
141 "file": "graph_track.py",
142 "domain": "graph-nn"
143 }
144}