Discriminant-Gated Positive Edge Adaptation / results.json
Failed on benchmark
1{
2 "math_check": {
3 "gap_log_slope": 1.0000000000000073,
4 "discriminant_log_slope": 2.0000000000000155,
5 "condition_times_epsilon": [
6 1.567517728806443,
7 1.5673004720745993,
8 1.566615136383714,
9 1.5644648610783944,
10 1.5578358841395883,
11 1.5386218788264114,
12 1.4964477821762656
13 ],
14 "exact_collision_disc": 0.0,
15 "collision_defective": true,
16 "passed": true
17 },
18 "bench_report": {
19 "bench_version": 1,
20 "track": "dynamics",
21 "model": "directed_rnn_laplacian",
22 "metric_direction": "lower is better",
23 "n_seeds": 8,
24 "baseline": {
25 "best_cfg": {
26 "lr": 0.006
27 },
28 "sweep": [
29 {
30 "cfg": {
31 "lr": 0.001
32 },
33 "mean": 0.13102261628955603
34 },
35 {
36 "cfg": {
37 "lr": 0.003
38 },
39 "mean": 0.01945999194867909
40 },
41 {
42 "cfg": {
43 "lr": 0.006
44 },
45 "mean": 0.009222074411809444
46 }
47 ],
48 "full": {
49 "mean": 0.00807881320361048,
50 "std": 0.002204711370306951,
51 "per_seed": [
52 0.009464219212532043,
53 0.009340533055365086,
54 0.005448193289339542,
55 0.012635352090001106,
56 0.007971327751874924,
57 0.007394182030111551,
58 0.0060087088495492935,
59 0.0063679893501102924
60 ],
61 "n": 8
62 }
63 },
64 "idea": {
65 "mean": 0.009180513268802315,
66 "std": 0.004307928922729323,
67 "per_seed": [
68 0.012163719162344933,
69 0.009676121175289154,
70 0.007755720056593418,
71 0.016999399289488792,
72 0.012369886040687561,
73 0.00709916464984417,
74 0.0028650257736444473,
75 0.004515070002526045
76 ],
77 "n": 8,
78 "best_cfg": {
79 "lr": 0.006,
80 "gmin": 0.01,
81 "beta": 20.0
82 },
83 "trials": [
84 {
85 "cfg": {
86 "lr": 0.001,
87 "gmin": 0.01,
88 "beta": 20.0
89 },
90 "mean": 0.2712850170210004,
91 "std": 0.1494091846161353,
92 "per_seed": [
93 0.5299652218818665,
94 0.2815037667751312,
95 0.13380131125450134,
96 0.3991413414478302,
97 0.18623359501361847,
98 0.05053729563951492,
99 0.19639013707637787,
100 0.3927074670791626
101 ],
102 "n": 8
103 },
104 {
105 "cfg": {
106 "lr": 0.003,
107 "gmin": 0.01,
108 "beta": 20.0
109 },
110 "mean": 0.018376422172877938,
111 "std": 0.008178008506149588,
112 "per_seed": [
113 0.02519739605486393,
114 0.01976214163005352,
115 0.014234600588679314,
116 0.028066840022802353,
117 0.029117483645677567,
118 0.0160621739923954,
119 0.005189564544707537,
120 0.00938117690384388
121 ],
122 "n": 8
123 },
124 {
125 "cfg": {
126 "lr": 0.006,
127 "gmin": 0.01,
128 "beta": 20.0
129 },
130 "mean": 0.009180513268802315,
131 "std": 0.004307928922729323,
132 "per_seed": [
133 0.012163719162344933,
134 0.009676121175289154,
135 0.007755720056593418,
136 0.016999399289488792,
137 0.012369886040687561,
138 0.00709916464984417,
139 0.0028650257736444473,
140 0.004515070002526045
141 ],
142 "n": 8
143 }
144 ]
145 },
146 "comparison": {
147 "delta_mean": 0.0011017000651918352,
148 "idea_wins": 3,
149 "n_pairs": 8,
150 "per_seed_diffs": [
151 0.002699499949812889,
152 0.00033558811992406845,
153 0.0023075267672538757,
154 0.004364047199487686,
155 0.004398558288812637,
156 -0.00029501738026738167,
157 -0.003143683075904846,
158 -0.0018529193475842476
159 ],
160 "p_value": 0.30765,
161 "mde": 0.002339736229369135,
162 "mde_rel_pct": 28.96138542135731,
163 "verdict": "no significant win",
164 "system_worked": false
165 },
166 "mechanism_signature": {
167 "prediction": "smaller normalized spectral gap predicts larger eigenvector conditioning",
168 "trained_model_observations": [
169 {
170 "seed": 0,
171 "baseline": {
172 "test_mse": 0.009464219212532043,
173 "gap": 0.0,
174 "condition": 12.532785415649414,
175 "gap_times_condition": 0.0
176 },
177 "idea": {
178 "test_mse": 0.012163719162344933,
179 "gap": 0.0,
180 "condition": 7.53489351272583,
181 "gap_times_condition": 0.0
182 }
183 },
184 {
185 "seed": 1,
186 "baseline": {
187 "test_mse": 0.009340533055365086,
188 "gap": 0.0,
189 "condition": 87.05731201171875,
190 "gap_times_condition": 0.0
191 },
192 "idea": {
193 "test_mse": 0.009676121175289154,
194 "gap": 0.0,
195 "condition": 30.088836669921875,
196 "gap_times_condition": 0.0
197 }
198 },
199 {
200 "seed": 2,
201 "baseline": {
202 "test_mse": 0.005448193289339542,
203 "gap": 0.0,
204 "condition": 13.377286911010742,
205 "gap_times_condition": 0.0
206 },
207 "idea": {
208 "test_mse": 0.007755720056593418,
209 "gap": 0.0,
210 "condition": 4.62471342086792,
211 "gap_times_condition": 0.0
212 }
213 },
214 {
215 "seed": 3,
216 "baseline": {
217 "test_mse": 0.012635352090001106,
218 "gap": 0.0,
219 "condition": 12.30354118347168,
220 "gap_times_condition": 0.0
221 },
222 "idea": {
223 "test_mse": 0.016999399289488792,
224 "gap": 0.0,
225 "condition": 8.969597816467285,
226 "gap_times_condition": 0.0
227 }
228 },
229 {
230 "seed": 4,
231 "baseline": {
232 "test_mse": 0.007971327751874924,
233 "gap": 0.0,
234 "condition": 18.872774124145508,
235 "gap_times_condition": 0.0
236 },
237 "idea": {
238 "test_mse": 0.012369886040687561,
239 "gap": 0.0,
240 "condition": 26.306346893310547,
241 "gap_times_condition": 0.0
242 }
243 },
244 {
245 "seed": 5,
246 "baseline": {
247 "test_mse": 0.007394182030111551,
248 "gap": 0.0,
249 "condition": 34.713871002197266,
250 "gap_times_condition": 0.0
251 },
252 "idea": {
253 "test_mse": 0.00709916464984417,
254 "gap": 0.0,
255 "condition": 17.471481323242188,
256 "gap_times_condition": 0.0
257 }
258 },
259 {
260 "seed": 6,
261 "baseline": {
262 "test_mse": 0.0060087088495492935,
263 "gap": 0.0,
264 "condition": 14.241284370422363,
265 "gap_times_condition": 0.0
266 },
267 "idea": {
268 "test_mse": 0.0028650257736444473,
269 "gap": 0.0,
270 "condition": 8.450821876525879,
271 "gap_times_condition": 0.0
272 }
273 },
274 {
275 "seed": 7,
276 "baseline": {
277 "test_mse": 0.0063679893501102924,
278 "gap": 0.0,
279 "condition": 27.74420166015625,
280 "gap_times_condition": 0.0
281 },
282 "idea": {
283 "test_mse": 0.004515070002526045,
284 "gap": 0.0,
285 "condition": 10.37414264678955,
286 "gap_times_condition": 0.0
287 }
288 }
289 ],
290 "baseline_mean_gap": 0.0,
291 "idea_mean_gap": 0.0,
292 "baseline_mean_condition": 27.605382084846497,
293 "idea_mean_condition": 14.227604269981384,
294 "confirmed": false
295 }
296 },
297 "device": "cuda",
298 "settings": {
299 "epochs": 12,
300 "n_train": 800,
301 "n_test": 300
302 }
303}