Dynamics-Matched Contractive Reservoir / 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      "epochs": 15,
 11      "weight_decay": 0.0
 12    },
 13    "sweep": [
 14      {
 15        "cfg": {
 16          "lr": 0.001,
 17          "epochs": 15,
 18          "weight_decay": 0.0
 19        },
 20        "mean": 0.0011778276093536988
 21      },
 22      {
 23        "cfg": {
 24          "lr": 0.003,
 25          "epochs": 15,
 26          "weight_decay": 0.0
 27        },
 28        "mean": 0.0007040829586912878
 29      },
 30      {
 31        "cfg": {
 32          "lr": 0.006,
 33          "epochs": 15,
 34          "weight_decay": 0.0
 35        },
 36        "mean": 0.00042879935062956065
 37      }
 38    ],
 39    "full": {
 40      "mean": 0.0003733093799382914,
 41      "std": 0.0001422580194548688,
 42      "per_seed": [
 43        0.00041920592775568366,
 44        0.0005275467410683632,
 45        0.0002469699247740209,
 46        0.0005214748089201748,
 47        0.00021985481726005673,
 48        0.00041659976704977453,
 49        0.00013517960906028748,
 50        0.0004996434436179698
 51      ],
 52      "n": 8
 53    }
 54  },
 55  "idea": {
 56    "per_seed": [
 57      0.001769391936250031,
 58      0.0016720223939046264,
 59      0.001442315406166017,
 60      0.001998485764488578,
 61      0.0017929950263351202,
 62      0.0018884261371567845,
 63      0.00186431675683707,
 64      0.0016658375971019268
 65    ],
 66    "mean": 0.0017617238772800192,
 67    "std": 0.00017026615425158047
 68  },
 69  "comparison": {
 70    "delta_mean": 0.0013884144973417278,
 71    "idea_wins": 0,
 72    "n_pairs": 8,
 73    "per_seed_diffs": [
 74      0.0013501860084943473,
 75      0.0011444756528362632,
 76      0.0011953454813919961,
 77      0.001477010955568403,
 78      0.0015731402090750635,
 79      0.00147182637010701,
 80      0.0017291371477767825,
 81      0.001166194153483957
 82    ],
 83    "p_value": 0.0081,
 84    "mde": 0.00017671141495484378,
 85    "mde_rel_pct": 47.33645186843536,
 86    "verdict": "idea worse (significant)",
 87    "system_worked": false
 88  },
 89  "mechanism_signature": {
 90    "prediction": "contractive recurrent perturbations decay",
 91    "idea_model": "DynamicsMatchedReservoir",
 92    "signature": {
 93      "mean_perturbation_ratio": 0.07229285687208176,
 94      "predicted_rho": 0.5131399999999999,
 95      "observed_ratio": 0.07229285687208176,
 96      "confirmed": false
 97    },
 98    "confirmed": false
 99  },
100  "idea_sweep": [
101    {
102      "cfg": {
103        "lr": 0.001,
104        "epochs": 15,
105        "leak": 0.35,
106        "residual_scale": 0.08
107      },
108      "mean": 0.002627965557621792,
109      "full": {
110        "per_seed": [
111          0.0030362543184310198,
112          0.002667633816599846,
113          0.002069995040073991,
114          0.002907233312726021,
115          0.002530411584302783,
116          0.0025524694938212633,
117          0.0028460959438234568,
118          0.0024136309511959553
119        ],
120        "mean": 0.002627965557621792,
121        "std": 0.00030852911016232825
122      }
123    },
124    {
125      "cfg": {
126        "lr": 0.003,
127        "epochs": 15,
128        "leak": 0.45,
129        "residual_scale": 0.1
130      },
131      "mean": 0.00218639854574576,
132      "full": {
133        "per_seed": [
134          0.0023016654886305332,
135          0.0021684756502509117,
136          0.001754965865984559,
137          0.0023502144031226635,
138          0.0021341126412153244,
139          0.0023281252942979336,
140          0.0023496386129409075,
141          0.0021039904095232487
142        ],
143        "mean": 0.00218639854574576,
144        "std": 0.00020101037575219233
145      }
146    },
147    {
148      "cfg": {
149        "lr": 0.006,
150        "epochs": 15,
151        "leak": 0.55,
152        "residual_scale": 0.14
153      },
154      "mean": 0.0017617238772800192,
155      "full": {
156        "per_seed": [
157          0.001769391936250031,
158          0.0016720223939046264,
159          0.001442315406166017,
160          0.001998485764488578,
161          0.0017929950263351202,
162          0.0018884261371567845,
163          0.00186431675683707,
164          0.0016658375971019268
165        ],
166        "mean": 0.0017617238772800192,
167        "std": 0.00017026615425158047
168      }
169    }
170  ],
171  "protocol": {
172    "paired_seeds": [
173      0,
174      1,
175      2,
176      3,
177      4,
178      5,
179      6,
180      7
181    ],
182    "n_train": 800,
183    "n_test": 300,
184    "epochs": 15,
185    "structural_match": "dynamics/control",
186    "search_space_parity": true
187  }
188}