Characteristic-Region Gain Controller / 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": 0.006,
 10      "clip": 1.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.0015,
 16          "clip": null
 17        },
 18        "mean": 0.005481100233737379
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.0015,
 23          "clip": 1.0
 24        },
 25        "mean": 0.004250517929904163
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "clip": null
 31        },
 32        "mean": 0.003975238680141047
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "clip": 1.0
 38        },
 39        "mean": 0.0035102504189126194
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.006,
 44          "clip": null
 45        },
 46        "mean": 0.002146739774616435
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.006,
 51          "clip": 1.0
 52        },
 53        "mean": 0.0013746844342676923
 54      }
 55    ],
 56    "full": {
 57      "mean": 0.0014901440081303008,
 58      "std": 0.00046532711612512625,
 59      "per_seed": [
 60        0.0009367614984512329,
 61        0.0008422294049523771,
 62        0.0017605363391339779,
 63        0.001959210494533181,
 64        0.0009294281480833888,
 65        0.0016490562120452523,
 66        0.0019374506082385778,
 67        0.0019064793596044183
 68      ],
 69      "n": 8
 70    },
 71    "idea_grid": [
 72      {
 73        "cfg": {
 74          "lr": 0.0015,
 75          "delta": 0.03
 76        },
 77        "mean": 0.001962410024134442
 78      },
 79      {
 80        "cfg": {
 81          "lr": 0.003,
 82          "delta": 0.05
 83        },
 84        "mean": 0.0012551909894682467
 85      },
 86      {
 87        "cfg": {
 88          "lr": 0.006,
 89          "delta": 0.1
 90        },
 91        "mean": 0.0011763981310650706
 92      }
 93    ],
 94    "idea_best_cfg": {
 95      "lr": 0.006,
 96      "delta": 0.1
 97    }
 98  },
 99  "idea": {
100    "mean": 0.001070475198503118,
101    "std": 0.00048340645360771137,
102    "per_seed": [
103      0.0018825745210051537,
104      0.0005507630994543433,
105      0.001416606828570366,
106      0.0008556480752304196,
107      0.0014776132302358747,
108      0.0004715493996627629,
109      0.0012880853610113263,
110      0.0006209610728546977
111    ],
112    "n": 8
113  },
114  "comparison": {
115    "delta_mean": -0.0004196688096271828,
116    "idea_wins": 6,
117    "n_pairs": 8,
118    "per_seed_diffs": [
119      0.0009458130225539207,
120      -0.00029146630549803376,
121      -0.000343929510563612,
122      -0.0011035624193027616,
123      0.0005481850821524858,
124      -0.0011775068123824894,
125      -0.0006493652472272515,
126      -0.0012855182867497206
127    ],
128    "p_value": 0.1834,
129    "mde": 0.0006828097406463084,
130    "mde_rel_pct": 45.821728431672646,
131    "verdict": "no significant win",
132    "system_worked": false
133  },
134  "mechanism_signature": {
135    "prediction": "feedback spectral radius should be held below 1-delta and correspond to local hidden sensitivity",
136    "measurement": {
137      "predicted_feedback_radius": 0.4187633916735649,
138      "observed_local_hidden_jacobian_gain": 0.715471138488652,
139      "relative_gap": 0.7085331526253469,
140      "n_models": 4,
141      "confirmed": false
142    }
143  },
144  "protocol": {
145    "epochs": 12,
146    "batch": 128,
147    "paired_seeds": [
148      0,
149      1,
150      2,
151      3,
152      4,
153      5,
154      6,
155      7
156    ],
157    "baseline_grid": [
158      {
159        "lr": 0.0015,
160        "clip": null
161      },
162      {
163        "lr": 0.0015,
164        "clip": 1.0
165      },
166      {
167        "lr": 0.003,
168        "clip": null
169      },
170      {
171        "lr": 0.003,
172        "clip": 1.0
173      },
174      {
175        "lr": 0.006,
176        "clip": null
177      },
178      {
179        "lr": 0.006,
180        "clip": 1.0
181      }
182    ],
183    "idea_grid": [
184      {
185        "lr": 0.0015,
186        "delta": 0.03
187      },
188      {
189        "lr": 0.003,
190        "delta": 0.05
191      },
192      {
193        "lr": 0.006,
194        "delta": 0.1
195      }
196    ]
197  }
198}