Finite-Horizon Walk Reciprocity Control / 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": 0.006,
10 "weight_decay": 0.0001
11 },
12 "sweep": [
13 {
14 "cfg": {
15 "lr": 0.001,
16 "weight_decay": 0.0
17 },
18 "mean": 0.004298779065720737
19 },
20 {
21 "cfg": {
22 "lr": 0.001,
23 "weight_decay": 0.0001
24 },
25 "mean": 0.0042995543335564435
26 },
27 {
28 "cfg": {
29 "lr": 0.003,
30 "weight_decay": 0.0
31 },
32 "mean": 0.0034615940821822733
33 },
34 {
35 "cfg": {
36 "lr": 0.003,
37 "weight_decay": 0.0001
38 },
39 "mean": 0.003432174737099558
40 },
41 {
42 "cfg": {
43 "lr": 0.006,
44 "weight_decay": 0.0
45 },
46 "mean": 0.0018282315868418664
47 },
48 {
49 "cfg": {
50 "lr": 0.006,
51 "weight_decay": 0.0001
52 },
53 "mean": 0.0017760894843377173
54 }
55 ],
56 "full": {
57 "mean": 0.0020704464986920357,
58 "std": 0.00039353273372166414,
59 "per_seed": [
60 0.0019586661364883184,
61 0.0013749259524047375,
62 0.00156879099085927,
63 0.002201974857598543,
64 0.0023778362665325403,
65 0.002474326640367508,
66 0.0020717033185064793,
67 0.0025353478267788887
68 ],
69 "n": 8
70 }
71 },
72 "idea": {
73 "best_config": {
74 "lr": 0.006,
75 "weight_decay": 0.0001,
76 "lambda": 0.8
77 },
78 "sweep": [
79 {
80 "config": {
81 "lr": 0.006,
82 "weight_decay": 0.0001,
83 "lambda": 0.05
84 },
85 "mean_metric": 0.0022101872455095872,
86 "per_seed": [
87 {
88 "seed": 0,
89 "metric": 0.0020552808418869972,
90 "final_train_loss": 0.001104112546890974,
91 "walk_energy": 0.19124190509319305,
92 "lambda": 0.05,
93 "lr": 0.006,
94 "weight_decay": 0.0001,
95 "epochs": 12,
96 "device": "cuda"
97 },
98 {
99 "seed": 1,
100 "metric": 0.0012910687364637852,
101 "final_train_loss": 0.0032270468026399613,
102 "walk_energy": 0.11217004805803299,
103 "lambda": 0.05,
104 "lr": 0.006,
105 "weight_decay": 0.0001,
106 "epochs": 12,
107 "device": "cuda"
108 },
109 {
110 "seed": 2,
111 "metric": 0.0014974906807765365,
112 "final_train_loss": 0.00170422742376104,
113 "walk_energy": 0.1552148163318634,
114 "lambda": 0.05,
115 "lr": 0.006,
116 "weight_decay": 0.0001,
117 "epochs": 12,
118 "device": "cuda"
119 },
120 {
121 "seed": 3,
122 "metric": 0.003510888433083892,
123 "final_train_loss": 0.0010945704020559788,
124 "walk_energy": 0.12788455188274384,
125 "lambda": 0.05,
126 "lr": 0.006,
127 "weight_decay": 0.0001,
128 "epochs": 12,
129 "device": "cuda"
130 },
131 {
132 "seed": 4,
133 "metric": 0.00242305058054626,
134 "final_train_loss": 0.0013659923383966088,
135 "walk_energy": 0.1590692102909088,
136 "lambda": 0.05,
137 "lr": 0.006,
138 "weight_decay": 0.0001,
139 "epochs": 12,
140 "device": "cuda"
141 },
142 {
143 "seed": 5,
144 "metric": 0.002272827783599496,
145 "final_train_loss": 0.0018004881963133812,
146 "walk_energy": 0.12237098067998886,
147 "lambda": 0.05,
148 "lr": 0.006,
149 "weight_decay": 0.0001,
150 "epochs": 12,
151 "device": "cuda"
152 },
153 {
154 "seed": 6,
155 "metric": 0.0021115862764418125,
156 "final_train_loss": 0.0014017952233552932,
157 "walk_energy": 0.13168057799339294,
158 "lambda": 0.05,
159 "lr": 0.006,
160 "weight_decay": 0.0001,
161 "epochs": 12,
162 "device": "cuda"
163 },
164 {
165 "seed": 7,
166 "metric": 0.002519304631277919,
167 "final_train_loss": 0.0020884747803211214,
168 "walk_energy": 0.11234524846076965,
169 "lambda": 0.05,
170 "lr": 0.006,
171 "weight_decay": 0.0001,
172 "epochs": 12,
173 "device": "cuda"
174 }
175 ]
176 },
177 {
178 "config": {
179 "lr": 0.006,
180 "weight_decay": 0.0001,
181 "lambda": 0.2
182 },
183 "mean_metric": 0.0016457567835459486,
184 "per_seed": [
185 {
186 "seed": 0,
187 "metric": 0.0013085808604955673,
188 "final_train_loss": 0.0015945398341864348,
189 "walk_energy": 0.12736238539218903,
190 "lambda": 0.2,
191 "lr": 0.006,
192 "weight_decay": 0.0001,
193 "epochs": 12,
194 "device": "cuda"
195 },
196 {
197 "seed": 1,
198 "metric": 0.0016004062490537763,
199 "final_train_loss": 0.003343571275472641,
200 "walk_energy": 0.1243424117565155,
201 "lambda": 0.2,
202 "lr": 0.006,
203 "weight_decay": 0.0001,
204 "epochs": 12,
205 "device": "cuda"
206 },
207 {
208 "seed": 2,
209 "metric": 0.0011136672692373395,
210 "final_train_loss": 0.0020982898748479785,
211 "walk_energy": 0.1328844130039215,
212 "lambda": 0.2,
213 "lr": 0.006,
214 "weight_decay": 0.0001,
215 "epochs": 12,
216 "device": "cuda"
217 },
218 {
219 "seed": 3,
220 "metric": 0.00136173190549016,
221 "final_train_loss": 0.003071792107075453,
222 "walk_energy": 0.12374606728553772,
223 "lambda": 0.2,
224 "lr": 0.006,
225 "weight_decay": 0.0001,
226 "epochs": 12,
227 "device": "cuda"
228 },
229 {
230 "seed": 4,
231 "metric": 0.0020485215354710817,
232 "final_train_loss": 0.00123758090659976,
233 "walk_energy": 0.1430458128452301,
234 "lambda": 0.2,
235 "lr": 0.006,
236 "weight_decay": 0.0001,
237 "epochs": 12,
238 "device": "cuda"
239 },
240 {
241 "seed": 5,
242 "metric": 0.0015574118588119745,
243 "final_train_loss": 0.001721083912998438,
244 "walk_energy": 0.12923158705234528,
245 "lambda": 0.2,
246 "lr": 0.006,
247 "weight_decay": 0.0001,
248 "epochs": 12,
249 "device": "cuda"
250 },
251 {
252 "seed": 6,
253 "metric": 0.0018434509402140975,
254 "final_train_loss": 0.0012715212209150194,
255 "walk_energy": 0.12898437678813934,
256 "lambda": 0.2,
257 "lr": 0.006,
258 "weight_decay": 0.0001,
259 "epochs": 12,
260 "device": "cuda"
261 },
262 {
263 "seed": 7,
264 "metric": 0.0023322836495935917,
265 "final_train_loss": 0.0020547736436128615,
266 "walk_energy": 0.12020274996757507,
267 "lambda": 0.2,
268 "lr": 0.006,
269 "weight_decay": 0.0001,
270 "epochs": 12,
271 "device": "cuda"
272 }
273 ]
274 },
275 {
276 "config": {
277 "lr": 0.006,
278 "weight_decay": 0.0001,
279 "lambda": 0.8
280 },
281 "mean_metric": 0.0015542887995252386,
282 "per_seed": [
283 {
284 "seed": 0,
285 "metric": 0.0011760086053982377,
286 "final_train_loss": 0.0023461854923516514,
287 "walk_energy": 0.08652391284704208,
288 "lambda": 0.8,
289 "lr": 0.006,
290 "weight_decay": 0.0001,
291 "epochs": 12,
292 "device": "cuda"
293 },
294 {
295 "seed": 1,
296 "metric": 0.002169822808355093,
297 "final_train_loss": 0.0021757491677999495,
298 "walk_energy": 0.10662304610013962,
299 "lambda": 0.8,
300 "lr": 0.006,
301 "weight_decay": 0.0001,
302 "epochs": 12,
303 "device": "cuda"
304 },
305 {
306 "seed": 2,
307 "metric": 0.0013154273619875312,
308 "final_train_loss": 0.0021398004796355963,
309 "walk_energy": 0.09533055126667023,
310 "lambda": 0.8,
311 "lr": 0.006,
312 "weight_decay": 0.0001,
313 "epochs": 12,
314 "device": "cuda"
315 },
316 {
317 "seed": 3,
318 "metric": 0.0012873380910605192,
319 "final_train_loss": 0.0009754107752814889,
320 "walk_energy": 0.08922960609197617,
321 "lambda": 0.8,
322 "lr": 0.006,
323 "weight_decay": 0.0001,
324 "epochs": 12,
325 "device": "cuda"
326 },
327 {
328 "seed": 4,
329 "metric": 0.0010369628435000777,
330 "final_train_loss": 0.002171484816353768,
331 "walk_energy": 0.09222258627414703,
332 "lambda": 0.8,
333 "lr": 0.006,
334 "weight_decay": 0.0001,
335 "epochs": 12,
336 "device": "cuda"
337 },
338 {
339 "seed": 5,
340 "metric": 0.0011381041258573532,
341 "final_train_loss": 0.002860049856826663,
342 "walk_energy": 0.1054113581776619,
343 "lambda": 0.8,
344 "lr": 0.006,
345 "weight_decay": 0.0001,
346 "epochs": 12,
347 "device": "cuda"
348 },
349 {
350 "seed": 6,
351 "metric": 0.002115012379363179,
352 "final_train_loss": 0.0033814193774014713,
353 "walk_energy": 0.09434174001216888,
354 "lambda": 0.8,
355 "lr": 0.006,
356 "weight_decay": 0.0001,
357 "epochs": 12,
358 "device": "cuda"
359 },
360 {
361 "seed": 7,
362 "metric": 0.0021956341806799173,
363 "final_train_loss": 0.001880955481901765,
364 "walk_energy": 0.1060996726155281,
365 "lambda": 0.8,
366 "lr": 0.006,
367 "weight_decay": 0.0001,
368 "epochs": 12,
369 "device": "cuda"
370 }
371 ]
372 }
373 ],
374 "per_seed": [
375 0.0011760086053982377,
376 0.002169822808355093,
377 0.0013154273619875312,
378 0.0012873380910605192,
379 0.0010369628435000777,
380 0.0011381041258573532,
381 0.002115012379363179,
382 0.0021956341806799173
383 ]
384 },
385 "comparison": {
386 "delta_mean": -0.0005161576991667971,
387 "idea_wins": 6,
388 "n_pairs": 8,
389 "per_seed_diffs": [
390 -0.0007826575310900807,
391 0.0007948968559503555,
392 -0.0002533636288717389,
393 -0.000914636766538024,
394 -0.0013408734230324626,
395 -0.0013362225145101547,
396 4.3309060856699944e-05,
397 -0.00033971364609897137
398 ],
399 "p_value": 0.09615,
400 "mde": 0.0006091006866650638,
401 "mde_rel_pct": 29.418808312595925,
402 "verdict": "no significant win",
403 "system_worked": false
404 },
405 "mechanism_signature": {
406 "prediction": "walk regularization lowers finite-horizon forward/backward energy",
407 "baseline_walk_energy_mean": 0.14293695613741875,
408 "idea_walk_energy_mean": 0.09697280917316675,
409 "relative_change_pct": -32.15693702058573,
410 "confirmed": true
411 },
412 "calibration": {
413 "g": 0.8,
414 "L": 4,
415 "phi_star": 0.40000000000000013,
416 "c_median": 0.4009627178311347,
417 "target": 0.16038508713245392
418 }
419}