Initial-Only Weight Decay with Tail Averaging / results.json

Audited (legacy)

Raw ⬇ ZIP
  1{
  2  "config": {
  3    "runs": 8,
  4    "gamma": 0.08,
  5    "lambda": 0.35,
  6    "m": 300,
  7    "T": 900,
  8    "batch": 32
  9  },
 10  "math_check": {
 11    "max_E_P2_minus_(1-gamma-lambda)A": -0.0009189718750894915,
 12    "max_(1-gamma-lambda)A_minus_scalar_bound": -0.0009375000000000355,
 13    "gamma_lambda": 0.25,
 14    "A_min": 0.725,
 15    "A_max": 0.74875,
 16    "passed": true
 17  },
 18  "summary": {
 19    "constant": {
 20      "final_test_mse_mean": 0.3203642681774104,
 21      "final_test_mse_std": 0.071445710306155,
 22      "tail_test_mse_mean": 0.3125744971991499,
 23      "tail_test_mse_std": 0.07139515447500065,
 24      "final_train_loss_mean": 0.15636681107505424,
 25      "max_update_ratio_mean": 0.10918760426791724
 26    },
 27    "initial": {
 28      "final_test_mse_mean": 0.0638978313147808,
 29      "final_test_mse_std": 0.0014369235048593272,
 30      "tail_test_mse_mean": 0.06369500187211916,
 31      "tail_test_mse_std": 0.0015074108328651016,
 32      "final_train_loss_mean": 0.031352790754280525,
 33      "max_update_ratio_mean": 0.10918760426791724
 34    },
 35    "none": {
 36      "final_test_mse_mean": 0.06373018867063518,
 37      "final_test_mse_std": 0.0013421314232348817,
 38      "tail_test_mse_mean": 0.0632831670600167,
 39      "tail_test_mse_std": 0.0013682426611847597,
 40      "final_train_loss_mean": 0.0312815944542155,
 41      "max_update_ratio_mean": 0.11112518680829603
 42    }
 43  },
 44  "raw": {
 45    "constant": [
 46      {
 47        "final_test_mse": 0.21331728641313216,
 48        "tail_test_mse": 0.2036002119241642,
 49        "final_train_loss": 0.1036035373723015,
 50        "max_update_ratio": 0.06499270761096203,
 51        "loss_trace": [
 52          0.09660208658154522,
 53          0.09913308304537978,
 54          0.09783132778109116,
 55          0.09602668356046042,
 56          0.09918072115516395,
 57          0.0960653519884296,
 58          0.0985837769559367,
 59          0.09757753741342245,
 60          0.1036035373723015
 61        ]
 62      },
 63      {
 64        "final_test_mse": 0.3336762274372533,
 65        "tail_test_mse": 0.3007667696015438,
 66        "final_train_loss": 0.16835525287881534,
 67        "max_update_ratio": 0.0965236182580031,
 68        "loss_trace": [
 69          0.14809438256306276,
 70          0.13818714998040907,
 71          0.1451342682279891,
 72          0.1522444636454781,
 73          0.1634730511346396,
 74          0.16086683699355697,
 75          0.16065222043098087,
 76          0.1513527999014487,
 77          0.16835525287881534
 78        ]
 79      },
 80      {
 81        "final_test_mse": 0.4118998875700791,
 82        "tail_test_mse": 0.4007535906631638,
 83        "final_train_loss": 0.2041949722213358,
 84        "max_update_ratio": 0.14430660665559253,
 85        "loss_trace": [
 86          0.18711951757953166,
 87          0.2062805549191513,
 88          0.19846467011958058,
 89          0.1888386444343984,
 90          0.19793456586673097,
 91          0.19955399019019116,
 92          0.21034249228064364,
 93          0.19306826541308295,
 94          0.2041949722213358
 95        ]
 96      },
 97      {
 98        "final_test_mse": 0.37388073157837576,
 99        "tail_test_mse": 0.35940212733192584,
100        "final_train_loss": 0.17025399883689188,
101        "max_update_ratio": 0.11492408402621879,
102        "loss_trace": [
103          0.19037835491255756,
104          0.16522536337192992,
105          0.1692678368442206,
106          0.18104866797708324,
107          0.18830999159545558,
108          0.1700615823983183,
109          0.1652802597017328,
110          0.1638268494213526,
111          0.17025399883689188
112        ]
113      },
114      {
115        "final_test_mse": 0.3159508289147718,
116        "tail_test_mse": 0.33265916277090796,
117        "final_train_loss": 0.15427306953203185,
118        "max_update_ratio": 0.12427139243681841,
119        "loss_trace": [
120          0.16694751716968168,
121          0.15884929663452926,
122          0.15522839138970188,
123          0.16603710996094348,
124          0.1673888391300696,
125          0.1583384592063742,
126          0.1605998974043406,
127          0.158740309307857,
128          0.15427306953203185
129        ]
130      },
131      {
132        "final_test_mse": 0.21650134727051537,
133        "tail_test_mse": 0.21134039656991946,
134        "final_train_loss": 0.10927616595019976,
135        "max_update_ratio": 0.09133585629500679,
136        "loss_trace": [
137          0.11381388847599726,
138          0.10770694334432898,
139          0.10723794334639294,
140          0.10853061158912658,
141          0.10670395724736192,
142          0.10445179276578088,
143          0.1097677679295758,
144          0.10233805062102014,
145          0.10927616595019976
146        ]
147      },
148      {
149        "final_test_mse": 0.3633906515046213,
150        "tail_test_mse": 0.36611774010447645,
151        "final_train_loss": 0.16672841665292176,
152        "max_update_ratio": 0.1189318931753644,
153        "loss_trace": [
154          0.17102765078086116,
155          0.17287814350797365,
156          0.16707115518766164,
157          0.15375190204834296,
158          0.16392554189749958,
159          0.15750587972672578,
160          0.1645870269209436,
161          0.15870054778818793,
162          0.16672841665292176
163        ]
164      },
165      {
166        "final_test_mse": 0.3342971847305346,
167        "tail_test_mse": 0.325955978627098,
168        "final_train_loss": 0.17424907515593577,
169        "max_update_ratio": 0.11821467568537189,
170        "loss_trace": [
171          0.1670257196884393,
172          0.17742467556531785,
173          0.17064711026491683,
174          0.1671730825563226,
175          0.1576267234719567,
176          0.17188930711541753,
177          0.17228754733423943,
178          0.17551962923167336,
179          0.17424907515593577
180        ]
181      }
182    ],
183    "initial": [
184      {
185        "final_test_mse": 0.06492002161680639,
186        "tail_test_mse": 0.06520083308607512,
187        "final_train_loss": 0.031200969560428452,
188        "max_update_ratio": 0.06499270761096203,
189        "loss_trace": [
190          0.09660208658154522,
191          0.09913308304537978,
192          0.09783132778109116,
193          0.036500407818686256,
194          0.03255487307711314,
195          0.0315038416251689,
196          0.03140892355079365,
197          0.031174608431663857,
198          0.031200969560428452
199        ]
200      },
201      {
202        "final_test_mse": 0.06344514018813517,
203        "tail_test_mse": 0.06286174835883732,
204        "final_train_loss": 0.031282136143950964,
205        "max_update_ratio": 0.0965236182580031,
206        "loss_trace": [
207          0.14809438256306276,
208          0.13818714998040907,
209          0.1451342682279891,
210          0.03391484686398738,
211          0.0320223370081068,
212          0.03136270728163026,
213          0.03130317868352726,
214          0.0311905641648367,
215          0.031282136143950964
216        ]
217      },
218      {
219        "final_test_mse": 0.06571035575984654,
220        "tail_test_mse": 0.06511946264624632,
221        "final_train_loss": 0.03200987902839998,
222        "max_update_ratio": 0.14430660665559253,
223        "loss_trace": [
224          0.18711951757953166,
225          0.2062805549191513,
226          0.19846467011958058,
227          0.04167717645077452,
228          0.03467092928737746,
229          0.032882226707053694,
230          0.032079726261500796,
231          0.03168363411556142,
232          0.03200987902839998
233        ]
234      },
235      {
236        "final_test_mse": 0.06552666763368384,
237        "tail_test_mse": 0.0651870328072081,
238        "final_train_loss": 0.03067828744404345,
239        "max_update_ratio": 0.11492408402621879,
240        "loss_trace": [
241          0.19037835491255756,
242          0.16522536337192992,
243          0.1692678368442206,
244          0.03529816673326417,
245          0.03209175831463503,
246          0.03126981883740531,
247          0.030625041356300602,
248          0.030743664376722844,
249          0.03067828744404345
250        ]
251      },
252      {
253        "final_test_mse": 0.0625480708133807,
254        "tail_test_mse": 0.06220476492124024,
255        "final_train_loss": 0.03142619618777982,
256        "max_update_ratio": 0.12427139243681841,
257        "loss_trace": [
258          0.16694751716968168,
259          0.15884929663452926,
260          0.15522839138970188,
261          0.035241499240020194,
262          0.032314712219419416,
263          0.03178441100373226,
264          0.03150388698247101,
265          0.03167112858263505,
266          0.03142619618777982
267        ]
268      },
269      {
270        "final_test_mse": 0.06254444221197832,
271        "tail_test_mse": 0.06213366836455627,
272        "final_train_loss": 0.030805195133915313,
273        "max_update_ratio": 0.09133585629500679,
274        "loss_trace": [
275          0.11381388847599726,
276          0.10770694334432898,
277          0.10723794334639294,
278          0.03311656889763006,
279          0.03138335464258121,
280          0.031053855321264832,
281          0.030797475956533437,
282          0.030894287520638258,
283          0.030805195133915313
284        ]
285      },
286      {
287        "final_test_mse": 0.06441796945835455,
288        "tail_test_mse": 0.06482015106732768,
289        "final_train_loss": 0.03161579930348645,
290        "max_update_ratio": 0.1189318931753644,
291        "loss_trace": [
292          0.17102765078086116,
293          0.17287814350797365,
294          0.16707115518766164,
295          0.038045702706975784,
296          0.033786422243261366,
297          0.03220516075440658,
298          0.03201198650584709,
299          0.03177018474381001,
300          0.03161579930348645
301        ]
302      },
303      {
304        "final_test_mse": 0.062069982836060966,
305        "tail_test_mse": 0.06203235372546228,
306        "final_train_loss": 0.03180386323223979,
307        "max_update_ratio": 0.11821467568537189,
308        "loss_trace": [
309          0.1670257196884393,
310          0.17742467556531785,
311          0.17064711026491683,
312          0.03794612469542157,
313          0.033739606534917345,
314          0.032699117597285436,
315          0.032368507449067975,
316          0.03241676851174073,
317          0.03180386323223979
318        ]
319      }
320    ],
321    "none": [
322      {
323        "final_test_mse": 0.06476665054800247,
324        "tail_test_mse": 0.06480021038821769,
325        "final_train_loss": 0.031147235906757862,
326        "max_update_ratio": 0.06568032398690693,
327        "loss_trace": [
328          0.04059513527630325,
329          0.033120904749436936,
330          0.03212091620039727,
331          0.03145724569989188,
332          0.03133375040014127,
333          0.031075550414622975,
334          0.03119433047020284,
335          0.0310885314156861,
336          0.031147235906757862
337        ]
338      },
339      {
340        "final_test_mse": 0.0634673688140642,
341        "tail_test_mse": 0.06277368076026113,
342        "final_train_loss": 0.03126276320510808,
343        "max_update_ratio": 0.0965236182580031,
344        "loss_trace": [
345          0.03744952041244664,
346          0.03257417930972933,
347          0.0315796171623284,
348          0.031327703305756044,
349          0.03126696393278839,
350          0.031117266134124594,
351          0.031174746374250252,
352          0.031147815990513646,
353          0.03126276320510808
354        ]
355      },
356      {
357        "final_test_mse": 0.06513689900289948,
358        "tail_test_mse": 0.06404720804012122,
359        "final_train_loss": 0.03174032140035571,
360        "max_update_ratio": 0.1456721704731453,
361        "loss_trace": [
362          0.04566273470985355,
363          0.03547186150984154,
364          0.03311067367363531,
365          0.03224013351234214,
366          0.03163440884557926,
367          0.03141763021941014,
368          0.03145244264651024,
369          0.031316720190829705,
370          0.03174032140035571
371        ]
372      },
373      {
374        "final_test_mse": 0.06539207050196598,
375        "tail_test_mse": 0.06487953349087666,
376        "final_train_loss": 0.030619988579689266,
377        "max_update_ratio": 0.11599251758465144,
378        "loss_trace": [
379          0.038094082180988614,
380          0.032179410462715094,
381          0.031065473442462424,
382          0.0307754119540056,
383          0.030816071304505623,
384          0.030734877353926153,
385          0.03042905733717449,
386          0.030639891339573476,
387          0.030619988579689266
388        ]
389      },
390      {
391        "final_test_mse": 0.062420358897009164,
392        "tail_test_mse": 0.06187282864980745,
393        "final_train_loss": 0.03140021261133678,
394        "max_update_ratio": 0.12868041890587384,
395        "loss_trace": [
396          0.03660380033124209,
397          0.03252721005390503,
398          0.03189075060330656,
399          0.0316534740313171,
400          0.03141734222350263,
401          0.03142589056026065,
402          0.03138274665130269,
403          0.031609799083981906,
404          0.03140021261133678
405        ]
406      },
407      {
408        "final_test_mse": 0.06251747077640823,
409        "tail_test_mse": 0.06200593504540686,
410        "final_train_loss": 0.03079597016249418,
411        "max_update_ratio": 0.0948062440527868,
412        "loss_trace": [
413          0.03472450739264996,
414          0.03178663345558015,
415          0.031061482586863656,
416          0.030967235558944414,
417          0.030901346864452176,
418          0.030911047641199587,
419          0.030724667002626063,
420          0.030860230973488872,
421          0.03079597016249418
422        ]
423      },
424      {
425        "final_test_mse": 0.06422535182844527,
426        "tail_test_mse": 0.06427706724765629,
427        "final_train_loss": 0.0315678106147631,
428        "max_update_ratio": 0.1191788989657806,
429        "loss_trace": [
430          0.04244847776330363,
431          0.034370330838248996,
432          0.03270160746426715,
433          0.03208047216870872,
434          0.0320743531885073,
435          0.03159902510937825,
436          0.03176352809269138,
437          0.03169293118460079,
438          0.0315678106147631
439        ]
440      },
441      {
442        "final_test_mse": 0.06191533899628657,
443        "tail_test_mse": 0.06160887285778629,
444        "final_train_loss": 0.03171845315321902,
445        "max_update_ratio": 0.12246730223922014,
446        "loss_trace": [
447          0.04265877400789941,
448          0.03519239713304322,
449          0.033025147512847554,
450          0.032131059340709374,
451          0.03195997250511952,
452          0.03200583712739417,
453          0.0319855697782013,
454          0.032237837076235856,
455          0.03171845315321902
456        ]
457      }
458    ]
459  }
460}