Port-Hamiltonian Neural ODE / bench_report.json
Mechanism confirmed, baseline not beaten
1{
2 "bench_version": 1,
3 "track": "dynamics",
4 "model": "rnn_small",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "best_cfg": {
9 "lr": 0.005,
10 "weight_decay": 0.0
11 },
12 "sweep": [
13 {
14 "cfg": {
15 "lr": 0.001,
16 "weight_decay": 0.0
17 },
18 "mean": 0.7906209975481033
19 },
20 {
21 "cfg": {
22 "lr": 0.003,
23 "weight_decay": 0.0
24 },
25 "mean": 0.6789352595806122
26 },
27 {
28 "cfg": {
29 "lr": 0.005,
30 "weight_decay": 0.0
31 },
32 "mean": 0.5756678730249405
33 }
34 ],
35 "full": {
36 "mean": 0.5096767991781235,
37 "std": 0.09851907442454794,
38 "per_seed": [
39 0.49524950981140137,
40 0.5737435817718506,
41 0.5768014788627625,
42 0.6568769216537476,
43 0.31393858790397644,
44 0.5531363487243652,
45 0.47284379601478577,
46 0.4348241686820984
47 ],
48 "n": 8
49 }
50 },
51 "idea": {
52 "mean": 0.8012284114956856,
53 "std": 0.20420508185800218,
54 "per_seed": [
55 0.7726240158081055,
56 1.0544251203536987,
57 1.041422724723816,
58 0.9839093685150146,
59 0.5845463275909424,
60 0.789264976978302,
61 0.7381134629249573,
62 0.4455212950706482
63 ],
64 "n": 8
65 },
66 "comparison": {
67 "delta_mean": 0.2915516123175621,
68 "idea_wins": 0,
69 "n_pairs": 8,
70 "per_seed_diffs": [
71 0.2773745059967041,
72 0.48068153858184814,
73 0.46462124586105347,
74 0.3270324468612671,
75 0.27060773968696594,
76 0.23612862825393677,
77 0.2652696669101715,
78 0.010697126388549805
79 ],
80 "p_value": 0.0081,
81 "mde": 0.12250781647639007,
82 "mde_rel_pct": 24.03637298655528,
83 "verdict": "idea worse (significant)",
84 "system_worked": false
85 },
86 "mechanism_signature": {
87 "track_match": "stability/control -> dynamics",
88 "math_sanity": {
89 "max_skew": 0.0,
90 "min_R_eig": 0.03000270389020443,
91 "max_identity_residual": 3.725290298461914e-09,
92 "max_dHdt": -0.000846284267026931
93 },
94 "mechanism_signature": {
95 "trained_model_samples": 8,
96 "max_skew_residual": 0.0,
97 "min_R_eigenvalue": 0.030000008642673492,
98 "max_energy_identity_residual": 1.4901161193847656e-08,
99 "max_observed_dHdt": -0.0005119751440361142,
100 "predicted": {
101 "skew_zero": true,
102 "R_eigenvalue_ge_epsilon": true,
103 "dHdt_le_zero": true
104 },
105 "confirmed": true
106 },
107 "idea_sweep": [
108 {
109 "cfg": {
110 "lr": 0.001,
111 "weight_decay": 0.0
112 },
113 "result": {
114 "mean": 0.8644205369055271,
115 "std": 0.20784380092323357,
116 "per_seed": [
117 0.8139765858650208,
118 1.1354857683181763,
119 1.0707145929336548,
120 1.0692869424819946,
121 0.6739808320999146,
122 0.8699936866760254,
123 0.7972943186759949,
124 0.48463156819343567
125 ],
126 "n": 8
127 }
128 },
129 {
130 "cfg": {
131 "lr": 0.003,
132 "weight_decay": 0.0
133 },
134 "result": {
135 "mean": 0.8342878073453903,
136 "std": 0.20629487017615392,
137 "per_seed": [
138 0.792961597442627,
139 1.0980685949325562,
140 1.055668830871582,
141 1.029300332069397,
142 0.6329599618911743,
143 0.8318073749542236,
144 0.7700064182281494,
145 0.4635293483734131
146 ],
147 "n": 8
148 }
149 },
150 {
151 "cfg": {
152 "lr": 0.005,
153 "weight_decay": 0.0
154 },
155 "result": {
156 "mean": 0.8012284114956856,
157 "std": 0.20420508185800218,
158 "per_seed": [
159 0.7726240158081055,
160 1.0544251203536987,
161 1.041422724723816,
162 0.9839093685150146,
163 0.5845463275909424,
164 0.789264976978302,
165 0.7381134629249573,
166 0.4455212950706482
167 ],
168 "n": 8
169 }
170 }
171 ],
172 "selected_cfg": {
173 "lr": 0.005,
174 "weight_decay": 0.0
175 },
176 "parameter_counts": {
177 "baseline": 13313,
178 "idea": 3142
179 }
180 }
181}