Coordinate Path-Integral Joint Gibbs Policy / bench_report.json
Mechanism confirmed, baseline not beaten
1{
2 "bench_version": 1,
3 "track": "constrained_two_agent_dynamics",
4 "model": "custom_gru_cell",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "best_cfg": {
9 "lr": 0.006,
10 "epochs": 12,
11 "hidden": 32,
12 "wd": 0.0
13 },
14 "sweep": [
15 {
16 "cfg": {
17 "lr": 0.001,
18 "epochs": 12,
19 "hidden": 32,
20 "wd": 0.0
21 },
22 "mean": 0.0322869922965765
23 },
24 {
25 "cfg": {
26 "lr": 0.003,
27 "epochs": 12,
28 "hidden": 32,
29 "wd": 0.0
30 },
31 "mean": 0.012963124550879002
32 },
33 {
34 "cfg": {
35 "lr": 0.006,
36 "epochs": 12,
37 "hidden": 32,
38 "wd": 0.0
39 },
40 "mean": 0.0070766935823485255
41 }
42 ],
43 "full": {
44 "mean": 0.006266987067647278,
45 "std": 0.0012038076406143412,
46 "per_seed": [
47 0.0054755136370658875,
48 0.008236748166382313,
49 0.007714319974184036,
50 0.006880192551761866,
51 0.005664798431098461,
52 0.006498826202005148,
53 0.004625890403985977,
54 0.005039607174694538
55 ],
56 "n": 8
57 }
58 },
59 "idea": {
60 "mean": 0.024705890100449324,
61 "std": 0.008283505340451159,
62 "per_seed": [
63 0.01789645291864872,
64 0.03382120281457901,
65 0.017041007056832314,
66 0.02942308597266674,
67 0.03449169918894768,
68 0.03302520141005516,
69 0.012732679024338722,
70 0.019215792417526245
71 ],
72 "n": 8
73 },
74 "comparison": {
75 "delta_mean": 0.018438903032802045,
76 "idea_wins": 0,
77 "n_pairs": 8,
78 "per_seed_diffs": [
79 0.012420939281582832,
80 0.025584454648196697,
81 0.009326687082648277,
82 0.022542893420904875,
83 0.028826900757849216,
84 0.026526375208050013,
85 0.008106788620352745,
86 0.014176185242831707
87 ],
88 "p_value": 0.0081,
89 "mde": 0.0069627855763603015,
90 "mde_rel_pct": 111.10260004053649,
91 "verdict": "idea worse (significant)",
92 "system_worked": false
93 },
94 "mechanism_signature": {
95 "custom_track": {
96 "name": "constrained_two_agent_dynamics",
97 "file": "bench/custom_tracks/constrained_two_agent_dynamics.py",
98 "domain": "dynamics"
99 },
100 "mechanism_signature": {
101 "predicted_rms": 2.836034980191471e-07,
102 "observed_rms": 2.836034980191471e-07,
103 "confirmed": true,
104 "note": "prediction is zero path disagreement only when conservative; this learned scalar field was measured directly"
105 },
106 "idea_trials": [
107 {
108 "cfg": {
109 "lr": 0.001,
110 "epochs": 12,
111 "hidden": 32,
112 "wd": 0.0,
113 "lam": 0.01,
114 "K": 3
115 },
116 "result": {
117 "mean": 0.45704858377575874,
118 "std": 0.10490055843604634,
119 "per_seed": [
120 0.3964577615261078,
121 0.5977270007133484,
122 0.37981122732162476,
123 0.45022860169410706,
124 0.5956597328186035,
125 0.5277622938156128,
126 0.27294501662254333,
127 0.4357970356941223
128 ],
129 "n": 8
130 }
131 },
132 {
133 "cfg": {
134 "lr": 0.001,
135 "epochs": 12,
136 "hidden": 32,
137 "wd": 0.0,
138 "lam": 0.05,
139 "K": 5
140 },
141 "result": {
142 "mean": 0.457201711833477,
143 "std": 0.1048935001658233,
144 "per_seed": [
145 0.3966236412525177,
146 0.5978873372077942,
147 0.3799804449081421,
148 0.45041248202323914,
149 0.5957075357437134,
150 0.5278149843215942,
151 0.27297601103782654,
152 0.4362112581729889
153 ],
154 "n": 8
155 }
156 },
157 {
158 "cfg": {
159 "lr": 0.001,
160 "epochs": 12,
161 "hidden": 32,
162 "wd": 0.0,
163 "lam": 0.2,
164 "K": 7
165 },
166 "result": {
167 "mean": 0.45794062316417694,
168 "std": 0.1048035662894998,
169 "per_seed": [
170 0.3975609242916107,
171 0.5985288619995117,
172 0.3805820643901825,
173 0.4510311186313629,
174 0.5958411693572998,
175 0.5279558300971985,
176 0.2731569707393646,
177 0.43886804580688477
178 ],
179 "n": 8
180 }
181 },
182 {
183 "cfg": {
184 "lr": 0.003,
185 "epochs": 12,
186 "hidden": 32,
187 "wd": 0.0,
188 "lam": 0.01,
189 "K": 3
190 },
191 "result": {
192 "mean": 0.10495583922602236,
193 "std": 0.06396028545593932,
194 "per_seed": [
195 0.05165071412920952,
196 0.2026720494031906,
197 0.04937386140227318,
198 0.07717829197645187,
199 0.18805915117263794,
200 0.15454889833927155,
201 0.02122720517218113,
202 0.0949365422129631
203 ],
204 "n": 8
205 }
206 },
207 {
208 "cfg": {
209 "lr": 0.003,
210 "epochs": 12,
211 "hidden": 32,
212 "wd": 0.0,
213 "lam": 0.05,
214 "K": 5
215 },
216 "result": {
217 "mean": 0.10538749303668737,
218 "std": 0.06388145594067134,
219 "per_seed": [
220 0.05256393551826477,
221 0.2028532475233078,
222 0.050049472600221634,
223 0.07778554409742355,
224 0.18824990093708038,
225 0.1546095907688141,
226 0.020773518830537796,
227 0.09621473401784897
228 ],
229 "n": 8
230 }
231 },
232 {
233 "cfg": {
234 "lr": 0.003,
235 "epochs": 12,
236 "hidden": 32,
237 "wd": 0.0,
238 "lam": 0.2,
239 "K": 7
240 },
241 "result": {
242 "mean": 0.10764682595618069,
243 "std": 0.06333663433666077,
244 "per_seed": [
245 0.05678858980536461,
246 0.20373065769672394,
247 0.053114570677280426,
248 0.08071713894605637,
249 0.189082533121109,
250 0.1552448719739914,
251 0.020189819857478142,
252 0.10230642557144165
253 ],
254 "n": 8
255 }
256 },
257 {
258 "cfg": {
259 "lr": 0.006,
260 "epochs": 12,
261 "hidden": 32,
262 "wd": 0.0,
263 "lam": 0.01,
264 "K": 3
265 },
266 "result": {
267 "mean": 0.05755363428033888,
268 "std": 0.01782041693525763,
269 "per_seed": [
270 0.050933968275785446,
271 0.06813938170671463,
272 0.034426964819431305,
273 0.07328452914953232,
274 0.07746168226003647,
275 0.07605248689651489,
276 0.028837481513619423,
277 0.051292579621076584
278 ],
279 "n": 8
280 }
281 },
282 {
283 "cfg": {
284 "lr": 0.006,
285 "epochs": 12,
286 "hidden": 32,
287 "wd": 0.0,
288 "lam": 0.05,
289 "K": 5
290 },
291 "result": {
292 "mean": 0.04705770337022841,
293 "std": 0.016346008984626046,
294 "per_seed": [
295 0.03983130306005478,
296 0.05912895128130913,
297 0.026205483824014664,
298 0.0593021921813488,
299 0.0660790354013443,
300 0.0639730915427208,
301 0.021240977570414543,
302 0.04070059210062027
303 ],
304 "n": 8
305 }
306 },
307 {
308 "cfg": {
309 "lr": 0.006,
310 "epochs": 12,
311 "hidden": 32,
312 "wd": 0.0,
313 "lam": 0.2,
314 "K": 7
315 },
316 "result": {
317 "mean": 0.024705890100449324,
318 "std": 0.008283505340451159,
319 "per_seed": [
320 0.01789645291864872,
321 0.03382120281457901,
322 0.017041007056832314,
323 0.02942308597266674,
324 0.03449169918894768,
325 0.03302520141005516,
326 0.012732679024338722,
327 0.019215792417526245
328 ],
329 "n": 8
330 }
331 }
332 ]
333 }
334}