Steady-State First-Passage Sensitivity Regularizer / 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": 0.005,
 10      "weight_decay": 0.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 0.5572434142231941
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "weight_decay": 0.0
 24        },
 25        "mean": 0.18839699774980545
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.005,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 0.04846041998825967
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.044931408134289086,
 37      "std": 0.03835479942009323,
 38      "per_seed": [
 39        0.11073682457208633,
 40        0.013142957352101803,
 41        0.052351340651512146,
 42        0.01761055737733841,
 43        0.10532612353563309,
 44        0.013849301263689995,
 45        0.029122132807970047,
 46        0.017312027513980865
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 0.048401523381471634,
 53    "std": 0.0418776346225875,
 54    "per_seed": [
 55      0.018566235899925232,
 56      0.14753688871860504,
 57      0.05397643893957138,
 58      0.04880054295063019,
 59      0.061454158276319504,
 60      0.008690654300153255,
 61      0.037352874875068665,
 62      0.010834393091499805
 63    ],
 64    "n": 8
 65  },
 66  "comparison": {
 67    "delta_mean": 0.003470115247182548,
 68    "idea_wins": 4,
 69    "n_pairs": 8,
 70    "per_seed_diffs": [
 71      -0.0921705886721611,
 72      0.13439393136650324,
 73      0.0016250982880592346,
 74      0.03118998557329178,
 75      -0.04387196525931358,
 76      -0.005158646963536739,
 77      0.008230742067098618,
 78      -0.00647763442248106
 79    ],
 80    "p_value": 0.89505,
 81    "mde": 0.054278585169126606,
 82    "mde_rel_pct": 120.80321410560086,
 83    "verdict": "no significant win",
 84    "system_worked": false
 85  },
 86  "mechanism_signature": {
 87    "predicted_vs_observed": {
 88      "predicted_aux_response": -0.0,
 89      "observed_short_fpt_response": -0.25157928466796875,
 90      "abs_error": 0.25157928466796875,
 91      "confirmed": false,
 92      "event_rate": 0.27937498688697815
 93    },
 94    "selected_idea_cfg": {
 95      "lr": 0.005,
 96      "weight_decay": 0.0
 97    },
 98    "regularizer": "soft regenerative FPT over short hidden-state prefixes"
 99  },
100  "idea_grid": [
101    {
102      "cfg": {
103        "lr": 0.001,
104        "weight_decay": 0.0
105      },
106      "full": {
107        "mean": 0.607760451734066,
108        "std": 0.06258788186798508,
109        "per_seed": [
110          0.6598727703094482,
111          0.7312052845954895,
112          0.6191193461418152,
113          0.622126042842865,
114          0.5374733209609985,
115          0.584874153137207,
116          0.5832061171531677,
117          0.5242065787315369
118        ],
119        "n": 8
120      }
121    },
122    {
123      "cfg": {
124        "lr": 0.003,
125        "weight_decay": 0.0
126      },
127      "full": {
128        "mean": 0.24948854185640812,
129        "std": 0.08010761069435396,
130        "per_seed": [
131          0.28027984499931335,
132          0.4332640767097473,
133          0.19582977890968323,
134          0.19673879444599152,
135          0.1759151667356491,
136          0.22718526422977448,
137          0.2931903898715973,
138          0.19350501894950867
139        ],
140        "n": 8
141      }
142    },
143    {
144      "cfg": {
145        "lr": 0.005,
146        "weight_decay": 0.0
147      },
148      "full": {
149        "mean": 0.048401523381471634,
150        "std": 0.0418776346225875,
151        "per_seed": [
152          0.018566235899925232,
153          0.14753688871860504,
154          0.05397643893957138,
155          0.04880054295063019,
156          0.061454158276319504,
157          0.008690654300153255,
158          0.037352874875068665,
159          0.010834393091499805
160        ],
161        "n": 8
162      }
163    }
164  ],
165  "budget": {
166    "epochs": 4,
167    "n_train": 200,
168    "n_test": 200
169  }
170}