Incidence-Matrix Structured Action Head / bench_report.json

✓✓ Beats tuned baseline

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "incidence_actions",
  4  "model": "custom_incidence_head",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006,
 10      "epochs": 12
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "epochs": 12
 17        },
 18        "mean": 0.7750000208616257
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "epochs": 12
 24        },
 25        "mean": 0.6015625
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.006,
 30          "epochs": 12
 31        },
 32        "mean": 0.3687500059604645
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.3562500048428774,
 37      "std": 0.06817944972613313,
 38      "per_seed": [
 39        0.32500001788139343,
 40        0.41875001788139343,
 41        0.3812499940395355,
 42        0.3499999940395355,
 43        0.3062500059604645,
 44        0.22500000894069672,
 45        0.4625000059604645,
 46        0.3812499940395355
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 0.12421874981373549,
 53    "std": 0.02854504035137365,
 54    "per_seed": [
 55      0.09375,
 56      0.15000000596046448,
 57      0.14374999701976776,
 58      0.11874999850988388,
 59      0.15625,
 60      0.06875000149011612,
 61      0.14374999701976776,
 62      0.11874999850988388
 63    ],
 64    "n": 8,
 65    "best_cfg": {
 66      "lr": 0.006,
 67      "epochs": 12,
 68      "alpha": 0.0
 69    },
 70    "sweep": [
 71      {
 72        "cfg": {
 73          "lr": 0.006,
 74          "epochs": 12,
 75          "alpha": 0.0
 76        },
 77        "mean": 0.12656250037252903
 78      },
 79      {
 80        "cfg": {
 81          "lr": 0.006,
 82          "epochs": 12,
 83          "alpha": 0.5
 84        },
 85        "mean": 0.27812499552965164
 86      },
 87      {
 88        "cfg": {
 89          "lr": 0.006,
 90          "epochs": 12,
 91          "alpha": 1.0
 92        },
 93        "mean": 0.690625011920929
 94      }
 95    ]
 96  },
 97  "comparison": {
 98    "delta_mean": -0.2320312550291419,
 99    "idea_wins": 8,
100    "n_pairs": 8,
101    "per_seed_diffs": [
102      -0.23125001788139343,
103      -0.26875001192092896,
104      -0.23749999701976776,
105      -0.23124999552965164,
106      -0.15000000596046448,
107      -0.1562500074505806,
108      -0.3187500089406967,
109      -0.26249999552965164
110    ],
111    "p_value": 0.0081,
112    "mde": 0.04719409486255682,
113    "mde_rel_pct": 13.247465044490758,
114    "verdict": "idea better (significant)",
115    "system_worked": true
116  },
117  "mechanism_signature": {
118    "custom_track": {
119      "name": "incidence_actions",
120      "file": "incidence_actions.py",
121      "domain": "structured_action_classification"
122    },
123    "baseline_sweep_union_lrs": [
124      0.001,
125      0.003,
126      0.006
127    ],
128    "idea_sweep": [
129      {
130        "cfg": {
131          "lr": 0.006,
132          "epochs": 12,
133          "alpha": 0.0
134        },
135        "mean": 0.12656250037252903
136      },
137      {
138        "cfg": {
139          "lr": 0.006,
140          "epochs": 12,
141          "alpha": 0.5
142        },
143        "mean": 0.27812499552965164
144      },
145      {
146        "cfg": {
147          "lr": 0.006,
148          "epochs": 12,
149          "alpha": 1.0
150        },
151        "mean": 0.690625011920929
152      }
153    ],
154    "mechanism_signature": {
155      "prediction": "trained incidence logits equal A times trained atomic scores (alpha-normalized)",
156      "n_observed_logits": 1636,
157      "predicted_max_abs_error": 0.0,
158      "observed_mean_abs_error": 0.0,
159      "confirmed": true
160    }
161  }
162}