Active-Set CG Router / bench_report.json
Beats tuned baseline
1{
2 "bench_version": 1,
3 "track": "tabular",
4 "model": "mlp_tiny",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "best_cfg": {
9 "lr": 0.01,
10 "weight_decay": 0.0001,
11 "epochs": 12,
12 "batch": 64,
13 "lam": 0.001
14 },
15 "sweep": [
16 {
17 "cfg": {
18 "lr": 0.001,
19 "weight_decay": 0.0,
20 "epochs": 12,
21 "batch": 64,
22 "lam": 0.001
23 },
24 "mean": 18.613102436065674
25 },
26 {
27 "cfg": {
28 "lr": 0.001,
29 "weight_decay": 0.0001,
30 "epochs": 12,
31 "batch": 64,
32 "lam": 0.001
33 },
34 "mean": 18.691447734832764
35 },
36 {
37 "cfg": {
38 "lr": 0.003,
39 "weight_decay": 0.0,
40 "epochs": 12,
41 "batch": 64,
42 "lam": 0.001
43 },
44 "mean": 13.314521551132202
45 },
46 {
47 "cfg": {
48 "lr": 0.003,
49 "weight_decay": 0.0001,
50 "epochs": 12,
51 "batch": 64,
52 "lam": 0.001
53 },
54 "mean": 13.184970617294312
55 },
56 {
57 "cfg": {
58 "lr": 0.01,
59 "weight_decay": 0.0,
60 "epochs": 12,
61 "batch": 64,
62 "lam": 0.001
63 },
64 "mean": 7.832738399505615
65 },
66 {
67 "cfg": {
68 "lr": 0.01,
69 "weight_decay": 0.0001,
70 "epochs": 12,
71 "batch": 64,
72 "lam": 0.001
73 },
74 "mean": 7.646669268608093
75 }
76 ],
77 "full": {
78 "mean": 7.848381221294403,
79 "std": 0.797417166889868,
80 "per_seed": [
81 7.573554515838623,
82 8.29439926147461,
83 7.347502708435059,
84 7.371220588684082,
85 6.545068740844727,
86 7.915830612182617,
87 9.391494750976562,
88 8.347978591918945
89 ],
90 "n": 8
91 },
92 "idea_grid": [
93 {
94 "cfg": {
95 "lr": 0.01,
96 "weight_decay": 0.0001,
97 "epochs": 12,
98 "batch": 64,
99 "lam": 0.0003
100 },
101 "mean": 6.944762110710144
102 },
103 {
104 "cfg": {
105 "lr": 0.01,
106 "weight_decay": 0.0001,
107 "epochs": 12,
108 "batch": 64,
109 "lam": 0.001
110 },
111 "mean": 6.944755673408508
112 },
113 {
114 "cfg": {
115 "lr": 0.01,
116 "weight_decay": 0.0001,
117 "epochs": 12,
118 "batch": 64,
119 "lam": 0.003
120 },
121 "mean": 6.944760799407959
122 }
123 ]
124 },
125 "idea": {
126 "mean": 7.08390074968338,
127 "std": 0.49960844624863177,
128 "per_seed": [
129 6.786007404327393,
130 7.105201244354248,
131 7.026352882385254,
132 6.861461162567139,
133 6.146136283874512,
134 7.232509613037109,
135 7.602020263671875,
136 7.911517143249512
137 ],
138 "n": 8
139 },
140 "comparison": {
141 "delta_mean": -0.764480471611023,
142 "idea_wins": 8,
143 "n_pairs": 8,
144 "per_seed_diffs": [
145 -0.7875471115112305,
146 -1.1891980171203613,
147 -0.3211498260498047,
148 -0.5097594261169434,
149 -0.39893245697021484,
150 -0.6833209991455078,
151 -1.7894744873046875,
152 -0.4364614486694336
153 ],
154 "p_value": 0.0081,
155 "mde": 0.416972102836185,
156 "mde_rel_pct": 5.312842114560478,
157 "verdict": "idea better (significant)",
158 "system_worked": true
159 },
160 "mechanism_signature": {
161 "track_match": "tabular optimizer/training-procedure structure",
162 "prediction": "simplex router remains feasible and active-set identifies sparse support",
163 "predicted": {
164 "simplex_error": 0.0,
165 "min_x": 0.0,
166 "sparse_support": true
167 },
168 "observed": {
169 "mean_simplex_error": 8.326672684688674e-17,
170 "min_x": 0.0,
171 "mean_support": 2.875,
172 "mean_pivots": 13.625
173 },
174 "confirmed": true
175 },
176 "idea_cfg": {
177 "lr": 0.01,
178 "weight_decay": 0.0001,
179 "epochs": 12,
180 "batch": 64,
181 "lam": 0.001
182 }
183}