Projective-Gap Regularization for Random Jacobian Cocycles / bench_report.json

Failed on benchmark

Raw ⬇ ZIP
  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": 0.002
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.36154668405652046
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.002
 21        },
 22        "mean": 0.04037633468396962
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.003
 27        },
 28        "mean": 0.05718149617314339
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.04037633468396962,
 33      "std": 0.019611854198954906,
 34      "per_seed": [
 35        0.014195265248417854,
 36        0.05633978545665741,
 37        0.01657027006149292,
 38        0.06460500508546829,
 39        0.021465152502059937,
 40        0.04150689020752907,
 41        0.04312236234545708,
 42        0.06520594656467438
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "per_seed": [
 49      0.01551117654889822,
 50      0.05140165239572525,
 51      0.008671858347952366,
 52      0.06345316767692566,
 53      0.025820285081863403,
 54      0.03666491433978081,
 55      0.037566497921943665,
 56      0.061046112328767776
 57    ],
 58    "mean": 0.03751695808023214,
 59    "std": 0.020258781501441284,
 60    "config": {
 61      "lr": 0.002,
 62      "kappa": 0.05,
 63      "lambda_star": -0.02,
 64      "gamma_star": 0.05,
 65      "epochs": 4
 66    },
 67    "sweep": [
 68      {
 69        "lr": 0.001,
 70        "kappa": 0.05,
 71        "mean": 0.35876577347517014,
 72        "std": 0.07349462586459757
 73      },
 74      {
 75        "lr": 0.001,
 76        "kappa": 0.2,
 77        "mean": 0.3647887259721756,
 78        "std": 0.07371815195644271
 79      },
 80      {
 81        "lr": 0.001,
 82        "kappa": 0.8,
 83        "mean": 0.3905522972345352,
 84        "std": 0.0744036324602034
 85      },
 86      {
 87        "lr": 0.002,
 88        "kappa": 0.05,
 89        "mean": 0.03751695808023214,
 90        "std": 0.020258781501441284
 91      },
 92      {
 93        "lr": 0.002,
 94        "kappa": 0.2,
 95        "mean": 0.041668576071970165,
 96        "std": 0.02693072627574791
 97      },
 98      {
 99        "lr": 0.002,
100        "kappa": 0.8,
101        "mean": 0.06623788620345294,
102        "std": 0.04677960203658622
103      },
104      {
105        "lr": 0.003,
106        "kappa": 0.05,
107        "mean": 0.06425974145531654,
108        "std": 0.016254177274281544
109      },
110      {
111        "lr": 0.003,
112        "kappa": 0.2,
113        "mean": 0.06531570013612509,
114        "std": 0.013804677930261229
115      },
116      {
117        "lr": 0.003,
118        "kappa": 0.8,
119        "mean": 0.060623346362262964,
120        "std": 0.014286012866245238
121      }
122    ]
123  },
124  "comparison": {
125    "delta_mean": -0.0028593766037374735,
126    "idea_wins": 6,
127    "n_pairs": 8,
128    "per_seed_diffs": [
129      0.0013159113004803658,
130      -0.004938133060932159,
131      -0.007898411713540554,
132      -0.001151837408542633,
133      0.004355132579803467,
134      -0.0048419758677482605,
135      -0.0055558644235134125,
136      -0.004159834235906601
137    ],
138    "p_value": 0.10075,
139    "mde": 0.0033891705100824727,
140    "mde_rel_pct": 8.393952885049904,
141    "verdict": "no significant win",
142    "system_worked": false
143  },
144  "mechanism_signature": {
145    "track_match": "dynamics stability/control",
146    "prediction": {
147      "observed_log_angle_slope": -0.035032859627017375,
148      "predicted_minus_gap": -0.08474337961524725,
149      "observed_gap": 0.08474337961524725,
150      "relative_error": 0.5866006313876798,
151      "confirmed": false
152    },
153    "selection": {
154      "kappa_candidates": [
155        0.05,
156        0.2,
157        0.8
158      ],
159      "best_config": {
160        "lr": 0.002,
161        "kappa": 0.05
162      }
163    }
164  }
165}