Characteristic-Invariant BT Monitor / 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": 0.01,
 10      "weight_decay": 0.0001,
 11      "epochs": 8
 12    },
 13    "sweep": [
 14      {
 15        "cfg": {
 16          "lr": 0.001,
 17          "weight_decay": 0.0,
 18          "epochs": 8
 19        },
 20        "mean": 0.02264283667318523
 21      },
 22      {
 23        "cfg": {
 24          "lr": 0.001,
 25          "weight_decay": 0.0001,
 26          "epochs": 8
 27        },
 28        "mean": 0.022735907114110887
 29      },
 30      {
 31        "cfg": {
 32          "lr": 0.003,
 33          "weight_decay": 0.0,
 34          "epochs": 8
 35        },
 36        "mean": 0.006244247197173536
 37      },
 38      {
 39        "cfg": {
 40          "lr": 0.003,
 41          "weight_decay": 0.0001,
 42          "epochs": 8
 43        },
 44        "mean": 0.00622874335385859
 45      },
 46      {
 47        "cfg": {
 48          "lr": 0.01,
 49          "weight_decay": 0.0,
 50          "epochs": 8
 51        },
 52        "mean": 0.00514151350944303
 53      },
 54      {
 55        "cfg": {
 56          "lr": 0.01,
 57          "weight_decay": 0.0001,
 58          "epochs": 8
 59        },
 60        "mean": 0.005085793818579987
 61      }
 62    ],
 63    "full": {
 64      "mean": 0.004114087234484032,
 65      "std": 0.0026467894721201165,
 66      "per_seed": [
 67        0.0021156652364879847,
 68        0.008485781028866768,
 69        0.0016995229525491595,
 70        0.008042206056416035,
 71        0.0027983859181404114,
 72        0.001188579830341041,
 73        0.003477072110399604,
 74        0.005105484742671251
 75      ],
 76      "n": 8
 77    }
 78  },
 79  "idea": {
 80    "mean": 0.004023682136903517,
 81    "std": 0.0026034026575959125,
 82    "per_seed": [
 83      0.002100873738527298,
 84      0.008452225476503372,
 85      0.0018759106751531363,
 86      0.007780176121741533,
 87      0.002418885938823223,
 88      0.0011175131658092141,
 89      0.0034842072054743767,
 90      0.004959664773195982
 91    ],
 92    "n": 8
 93  },
 94  "comparison": {
 95    "delta_mean": -9.040509758051485e-05,
 96    "idea_wins": 6,
 97    "n_pairs": 8,
 98    "per_seed_diffs": [
 99      -1.4791497960686684e-05,
100      -3.355555236339569e-05,
101      0.00017638772260397673,
102      -0.0002620299346745014,
103      -0.00037949997931718826,
104      -7.106666453182697e-05,
105      7.135095074772835e-06,
106      -0.00014581996947526932
107    ],
108    "p_value": 0.1825,
109    "mde": 0.00014348105739960172,
110    "mde_rel_pct": 3.4875550570963134,
111    "verdict": "no significant win",
112    "system_worked": false
113  },
114  "mechanism_signature": {
115    "idea_config": {
116      "lr": 0.01,
117      "weight_decay": 0.0001,
118      "epochs": 8,
119      "rho": 0.05
120    },
121    "idea_sweep": [
122      {
123        "cfg": {
124          "lr": 0.01,
125          "weight_decay": 0.0001,
126          "epochs": 8,
127          "rho": 0.001
128        },
129        "mean": 0.004106936205062084
130      },
131      {
132        "cfg": {
133          "lr": 0.01,
134          "weight_decay": 0.0001,
135          "epochs": 8,
136          "rho": 0.01
137        },
138        "mean": 0.004095953670912422
139      },
140      {
141        "cfg": {
142          "lr": 0.01,
143          "weight_decay": 0.0001,
144          "epochs": 8,
145          "rho": 0.05
146        },
147        "mean": 0.004023682136903517
148      }
149    ],
150    "prediction": "BT monitor drives delta,tau toward zero and the trained local dynamics toward a near-zero continuous eigenvalue",
151    "trained_mean_abs_delta": 0.08188444655388594,
152    "trained_mean_abs_tau": 0.573352038860321,
153    "observed_mean_discrete_radius": 0.736884132027626,
154    "observed_mean_continuous_max_real": -0.26311645470559597,
155    "confirmed": false
156  },
157  "custom_track": null
158}