Symplectic Hamiltonian Optimizer / results.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  1{
  2  "harmonic_check": {
  3    "step": 0.5,
  4    "steps": 2000,
  5    "leapfrog_energy_start": 0.5,
  6    "leapfrog_energy_end": 0.46990697814208626,
  7    "leapfrog_energy_abs_drift": 0.030093021857913738,
  8    "leapfrog_energy_range": 0.031249974365196387,
  9    "euler_energy_start": 0.5,
 10    "euler_energy_end": 2.6429321103222574e+193,
 11    "euler_energy_growth_factor": 5.285864220644515e+193,
 12    "euler_energy_range": 2.6429321103222574e+193,
 13    "claim_observed": false
 14  },
 15  "device": "cuda",
 16  "sgd": {
 17    "steps": 240,
 18    "gradient_evals": 240,
 19    "final_train_loss": 0.01566408760845661,
 20    "val_loss": 0.14021605253219604,
 21    "val_accuracy": 0.9622222185134888,
 22    "mean_grad_norm": 0.14898149490666887,
 23    "time_sec": 1.406975507736206,
 24    "trace": [
 25      [
 26        40,
 27        40,
 28        0.16501370072364807
 29      ],
 30      [
 31        80,
 32        80,
 33        0.028226403519511223
 34      ],
 35      [
 36        120,
 37        120,
 38        0.03268488496541977
 39      ],
 40      [
 41        160,
 42        160,
 43        0.02408800832927227
 44      ],
 45      [
 46        200,
 47        200,
 48        0.010942094959318638
 49      ],
 50      [
 51        240,
 52        240,
 53        0.01566408760845661
 54      ]
 55    ]
 56  },
 57  "adamw": {
 58    "steps": 240,
 59    "gradient_evals": 240,
 60    "final_train_loss": 0.041272617876529694,
 61    "val_loss": 0.13408653438091278,
 62    "val_accuracy": 0.9644444584846497,
 63    "mean_grad_norm": 0.22746927465001743,
 64    "time_sec": 1.6944279670715332,
 65    "trace": [
 66      [
 67        40,
 68        40,
 69        0.4444103240966797
 70      ],
 71      [
 72        80,
 73        80,
 74        0.1208081766963005
 75      ],
 76      [
 77        120,
 78        120,
 79        0.09874539077281952
 80      ],
 81      [
 82        160,
 83        160,
 84        0.07196119427680969
 85      ],
 86      [
 87        200,
 88        200,
 89        0.028583494946360588
 90      ],
 91      [
 92        240,
 93        240,
 94        0.041272617876529694
 95      ]
 96    ]
 97  },
 98  "leapfrog": {
 99    "steps": 120,
100    "gradient_evals": 240,
101    "final_train_loss": 0.10090326517820358,
102    "val_loss": 0.2772459387779236,
103    "val_accuracy": 0.948888897895813,
104    "mean_grad_norm": 0.4930637594312429,
105    "time_sec": 1.007786750793457,
106    "trace": [
107      [
108        20,
109        40,
110        1.5030821561813354
111      ],
112      [
113        40,
114        80,
115        0.6404362320899963
116      ],
117      [
118        60,
119        120,
120        0.2691977322101593
121      ],
122      [
123        80,
124        160,
125        0.0722397118806839
126      ],
127      [
128        100,
129        200,
130        0.16021402180194855
131      ],
132      [
133        120,
134        240,
135        0.10090326517820358
136      ]
137    ]
138  }
139}