Smooth Spectral Muon / results.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  1{
  2  "seed": 1689,
  3  "finite_difference": {
  4    "analytic_directional_derivative": -0.14163368516537783,
  5    "rows": [
  6      {
  7        "h": 0.01,
  8        "numeric": -0.1416329871558819,
  9        "abs_error": 6.980094959252447e-07
 10      },
 11      {
 12        "h": 0.003,
 13        "numeric": -0.1416336223429108,
 14        "abs_error": 6.282246703404049e-08
 15      },
 16      {
 17        "h": 0.001,
 18        "numeric": -0.14163367818476047,
 19        "abs_error": 6.98061736192912e-09
 20      },
 21      {
 22        "h": 0.0003,
 23        "numeric": -0.14163368453612435,
 24        "abs_error": 6.292534826535245e-10
 25      },
 26      {
 27        "h": 0.0001,
 28        "numeric": -0.14163368509567675,
 29        "abs_error": 6.970107824244565e-11
 30      },
 31      {
 32        "h": 3e-05,
 33        "numeric": -0.14163368519485667,
 34        "abs_error": 2.9478836038876466e-11
 35      },
 36      {
 37        "h": 1e-05,
 38        "numeric": -0.14163368513564478,
 39        "abs_error": 2.973304935594001e-11
 40      }
 41    ],
 42    "best_abs_error": 2.9478836038876466e-11
 43  },
 44  "response_sweep": {
 45    "prediction": "singular response transitions at sigma/sqrt(epsilon)=1",
 46    "ratios_sigma_over_sqrt_eps": [
 47      0.0001,
 48      0.00031622776601683794,
 49      0.001,
 50      0.0031622776601683794,
 51      0.01,
 52      0.03162277660168379,
 53      0.1,
 54      0.31622776601683794,
 55      1.0,
 56      3.1622776601683795,
 57      10.0,
 58      31.622776601683793,
 59      100.0,
 60      316.22776601683796,
 61      1000.0,
 62      3162.2776601683795,
 63      10000.0
 64    ],
 65    "responses": [
 66      9.999999950000001e-05,
 67      0.0003162277502054508,
 68      0.000999999500000375,
 69      0.003162261848898663,
 70      0.009999500037496877,
 71      0.0316069770620507,
 72      0.09950371902099893,
 73      0.30151134457776363,
 74      0.7071067811865475,
 75      0.9534625892455922,
 76      0.9950371902099892,
 77      0.9995003746877731,
 78      0.9999500037496876,
 79      0.9999950000374996,
 80      0.999999500000375,
 81      0.9999999500000037,
 82      0.9999999949999999
 83    ],
 84    "predicted_transition_ratio": 1.0,
 85    "observed_nearest_transition_ratio": 1.0,
 86    "response_at_transition_predicted": 0.7071067811865475,
 87    "response_at_transition_observed": 0.7071067811865475,
 88    "small_t_prediction_r_over_t": 0.999999995,
 89    "large_t_prediction_2t2_times_1_minus_r": 1.0000000161269895,
 90    "max_response_observed": 0.9999999949999999,
 91    "epsilon_grid_fixed_sigma_1": [
 92      1e-06,
 93      1e-05,
 94      0.0001,
 95      0.001,
 96      0.01,
 97      0.1,
 98      1.0,
 99      10.0,
100      100.0,
101      1000.0
102    ],
103    "fixed_sigma_predicted": [
104      0.999999500000375,
105      0.9999950000374997,
106      0.9999500037496877,
107      0.9995003746877732,
108      0.9950371902099893,
109      0.9534625892455922,
110      0.7071067811865475,
111      0.30151134457776363,
112      0.09950371902099892,
113      0.0316069770620507
114    ],
115    "fixed_sigma_observed": [
116      0.999999500000375,
117      0.9999950000374997,
118      0.9999500037496877,
119      0.9995003746877732,
120      0.9950371902099893,
121      0.9534625892455922,
122      0.7071067811865475,
123      0.30151134457776363,
124      0.09950371902099892,
125      0.0316069770620507
126    ],
127    "matrix_spectral_formula_max_error": 0.0
128  },
129  "toy_comparison": {
130    "sgd": {
131      "final_loss": 2.8355612452382438e-05,
132      "best_loss": 3.7963962515738206e-06,
133      "update_norm_mean": 0.6642604379633125,
134      "update_norm_std": 0.29678896034672586,
135      "update_norm_cv": 0.4467960808508528,
136      "loss_first": 0.04034999807865995,
137      "loss_curve": [
138        0.04034999807865995,
139        0.018971995832908362,
140        0.005565708537925589,
141        0.00013346439081310643,
142        0.00011024596918112161,
143        4.962311936298114e-05,
144        2.759519221167905e-05,
145        2.9369196115233888e-05,
146        2.712347766058562e-05,
147        2.3167069941137548e-05
148      ],
149      "min_momentum_singular": 0.015688562780269057
150    },
151    "exact": {
152      "final_loss": 0.00018187237486999214,
153      "best_loss": 0.0001235159431871079,
154      "update_norm_mean": 2.8284271247461907,
155      "update_norm_std": 8.500568152673366e-16,
156      "update_norm_cv": 3.005404692345807e-16,
157      "loss_first": 0.03880606406517094,
158      "loss_curve": [
159        0.03880606406517094,
160        0.004898389769818062,
161        0.000510353220816358,
162        0.0002474814146679972,
163        0.0002766249694004397,
164        0.00020730950813550102,
165        0.00016127188648561307,
166        0.00020316015023368785,
167        0.00014701428715598988,
168        0.00016976999600274185
169      ],
170      "min_momentum_singular": 0.010968158387157818
171    },
172    "smooth": {
173      "final_loss": 0.00016259570651245642,
174      "best_loss": 0.00012357335294179447,
175      "update_norm_mean": 2.79794567182597,
176      "update_norm_std": 0.053903336108416,
177      "update_norm_cv": 0.01926532621815306,
178      "loss_first": 0.038808014792410056,
179      "loss_curve": [
180        0.038808014792410056,
181        0.004967632102564926,
182        0.0005360218026843526,
183        0.0002558978687632341,
184        0.0002706488501778976,
185        0.00018918430224761884,
186        0.00017027405554747235,
187        0.00018922800627345067,
188        0.00014579205264547506,
189        0.00018336382455353108
190      ],
191      "min_momentum_singular": 0.008469241662925672
192    }
193  },
194  "epsilon_ablation": {
195    "1e-05": {
196      "final_loss": 0.0001767168613448352,
197      "best_loss": 0.00013135737329040946,
198      "update_norm_mean": 2.8277751212553195,
199      "update_norm_std": 0.002005691478721801,
200      "update_norm_cv": 0.0007092825252068542,
201      "loss_first": 0.038806083637211296,
202      "loss_curve": [
203        0.038806083637211296,
204        0.004903377859070076,
205        0.0005191519840394792,
206        0.0002441090553350558,
207        0.00027259920767325776,
208        0.0001999924830932556,
209        0.00017071707777003412,
210        0.0001625462112968725,
211        0.000179855242917784,
212        0.00015411317515638573
213      ],
214      "min_momentum_singular": 0.01147536117033023
215    },
216    "0.001": {
217      "final_loss": 0.00016259570651245642,
218      "best_loss": 0.00012357335294179447,
219      "update_norm_mean": 2.79794567182597,
220      "update_norm_std": 0.053903336108416,
221      "update_norm_cv": 0.01926532621815306,
222      "loss_first": 0.038808014792410056,
223      "loss_curve": [
224        0.038808014792410056,
225        0.004967632102564926,
226        0.0005360218026843526,
227        0.0002558978687632341,
228        0.0002706488501778976,
229        0.00018918430224761884,
230        0.00017027405554747235,
231        0.00018922800627345067,
232        0.00014579205264547506,
233        0.00018336382455353108
234      ],
235      "min_momentum_singular": 0.008469241662925672
236    },
237    "0.1": {
238      "final_loss": 0.0001446746911068479,
239      "best_loss": 0.0001149052776128946,
240      "update_norm_mean": 2.4077276845977913,
241      "update_norm_std": 0.20196409120491812,
242      "update_norm_cv": 0.08388161688582825,
243      "loss_first": 0.03896081751583297,
244      "loss_curve": [
245        0.03896081751583297,
246        0.00477489390293047,
247        0.000675243531337686,
248        0.0002468074182051285,
249        0.00026568389997491887,
250        0.00020115738121748522,
251        0.0001636006046114946,
252        0.00013298645904230209,
253        0.00013854358478854737,
254        0.00016357838653318266
255      ],
256      "min_momentum_singular": 0.011686803628782494
257    },
258    "1.0": {
259      "final_loss": 0.00010893248726784571,
260      "best_loss": 7.824853826985934e-05,
261      "update_norm_mean": 1.7349755997879706,
262      "update_norm_std": 0.15130256760976338,
263      "update_norm_cv": 0.08720731728340542,
264      "loss_first": 0.03955777056698747,
265      "loss_curve": [
266        0.03955777056698747,
267        0.008052650685282986,
268        0.0005614260869127422,
269        0.00024422607435581036,
270        0.00018234544423404095,
271        0.0001261603464356092,
272        0.00014145998633750993,
273        0.0001226991436438825,
274        0.00011995393292355408,
275        9.801926935681749e-05
276      ],
277      "min_momentum_singular": 0.011106851943835004
278    }
279  }
280}