Fractional Memory State-Space Layer / bench_report.json

Failed on benchmark

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "dynamics",
  4  "model": "rnn_small_fractional",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.0004224302538204938
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 3.919086066161981e-05
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 1.4552034826920135e-05
 29      }
 30    ],
 31    "full": {
 32      "mean": 1.477801822602487e-05,
 33      "std": 5.904456722576056e-06,
 34      "per_seed": [
 35        2.641624269017484e-05,
 36        8.83757365954807e-06,
 37        8.367352165805642e-06,
 38        1.4586970792151988e-05,
 39        1.0999674486811273e-05,
 40        2.1751329768449068e-05,
 41        1.276554939977359e-05,
 42        1.4499452845484484e-05
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.00025887351512210444,
 49    "std": 5.818967880653584e-05,
 50    "per_seed": [
 51      0.0002734987938310951,
 52      0.00015879524289630353,
 53      0.00034040064201690257,
 54      0.00028941663913428783,
 55      0.000296872021863237,
 56      0.00026502739638090134,
 57      0.0002749341365415603,
 58      0.0001720432483125478
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.00024409549689607957,
 64    "idea_wins": 0,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.00024708255114092026,
 68      0.00014995766923675546,
 69      0.0003320332898510969,
 70      0.00027482966834213585,
 71      0.00028587234737642575,
 72      0.00024327606661245227,
 73      0.0002621685871417867,
 74      0.00015754379546706332
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 5.2037973707415915e-05,
 78    "mde_rel_pct": 352.13093468631877,
 79    "verdict": "idea worse (significant)",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "math_sanity": [
 84      {
 85        "dt": 0.1,
 86        "predicted_rho": 0.9992190550963239,
 87        "observed_rho": 0.9992190550963239,
 88        "stable": true
 89      },
 90      {
 91        "dt": 1.0,
 92        "predicted_rho": 0.9922179382602435,
 93        "observed_rho": 0.9922179382602435,
 94        "stable": true
 95      },
 96      {
 97        "dt": 4.0,
 98        "predicted_rho": 0.9692332344763441,
 99        "observed_rho": 0.9692332344763441,
100        "stable": true
101      },
102      {
103        "predicted_log_slope": -0.5,
104        "observed_log_slope": -0.8731882906575884,
105        "abs_error": 0.3731882906575884,
106        "positive_weights": true
107      }
108    ],
109    "trained_model_behavior": {
110      "prediction": "trained exponential-bank memory impulse log-slope p-1",
111      "p": 0.5,
112      "predicted": -0.5,
113      "observed": -0.8731883349875018,
114      "abs_error": 0.3731883349875018,
115      "trained_positive_weights": true,
116      "measurement": "trained model lam/w state response",
117      "confirmed": false
118    },
119    "track_choice": "dynamics: actuated pendulum stability/control is structurally matched",
120    "idea_sweep": [
121      {
122        "cfg": {
123          "lr": 0.001,
124          "p": 0.5
125        },
126        "mean": 0.00031832019703870174
127      },
128      {
129        "cfg": {
130          "lr": 0.003,
131          "p": 0.5
132        },
133        "mean": 0.00025887351512210444
134      },
135      {
136        "cfg": {
137          "lr": 0.01,
138          "p": 0.5
139        },
140        "mean": 0.0007522592495661229
141      }
142    ]
143  },
144  "stage2_protocol": {
145    "epochs": 12,
146    "batch": 128,
147    "baseline_grid": [
148      {
149        "lr": 0.001
150      },
151      {
152        "lr": 0.003
153      },
154      {
155        "lr": 0.01
156      }
157    ],
158    "idea_grid": [
159      {
160        "lr": 0.001,
161        "p": 0.5
162      },
163      {
164        "lr": 0.003,
165        "p": 0.5
166      },
167      {
168        "lr": 0.01,
169        "p": 0.5
170      }
171    ],
172    "paired_seeds": [
173      0,
174      1,
175      2,
176      3,
177      4,
178      5,
179      6,
180      7
181    ]
182  }
183}