Calibrated Compact-Support Anomaly Score / bench_report.json

Failed on benchmark

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "geometry_support_regression",
  4  "model": "matched_mlp_embedding_regressor",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.07304414547979832
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.059416051022708416
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 0.04957881290465593
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.050347565207630396,
 33      "std": 0.007310327369941531,
 34      "per_seed": [
 35        0.052693966776132584,
 36        0.044472310692071915,
 37        0.04606407880783081,
 38        0.055084895342588425,
 39        0.06478562951087952,
 40        0.03895844891667366,
 41        0.04829391837120056,
 42        0.052427273243665695
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.07299377396702766,
 49    "std": 0.005527255691171378,
 50    "per_seed": [
 51      0.07619384676218033,
 52      0.0728721097111702,
 53      0.07278692722320557,
 54      0.07364533096551895,
 55      0.07819892466068268,
 56      0.06470585614442825,
 57      0.06447294354438782,
 58      0.08107425272464752
 59    ],
 60    "n": 8,
 61    "per_seed_signatures": [
 62      {
 63        "mean_test_radius2": 2.279507637023926,
 64        "test_inside_support_fraction": 0.9799999594688416
 65      },
 66      {
 67        "mean_test_radius2": 2.5773558616638184,
 68        "test_inside_support_fraction": 0.9774999618530273
 69      },
 70      {
 71        "mean_test_radius2": 2.476716995239258,
 72        "test_inside_support_fraction": 0.9975000023841858
 73      },
 74      {
 75        "mean_test_radius2": 3.006222724914551,
 76        "test_inside_support_fraction": 0.9874999523162842
 77      },
 78      {
 79        "mean_test_radius2": 2.760526657104492,
 80        "test_inside_support_fraction": 0.9749999642372131
 81      },
 82      {
 83        "mean_test_radius2": 2.750863552093506,
 84        "test_inside_support_fraction": 0.9874999523162842
 85      },
 86      {
 87        "mean_test_radius2": 2.6677849292755127,
 88        "test_inside_support_fraction": 0.9749999642372131
 89      },
 90      {
 91        "mean_test_radius2": 2.4258546829223633,
 92        "test_inside_support_fraction": 1.0
 93      }
 94    ]
 95  },
 96  "comparison": {
 97    "delta_mean": 0.02264620875939727,
 98    "idea_wins": 0,
 99    "n_pairs": 8,
100    "per_seed_diffs": [
101      0.023499879986047745,
102      0.028399799019098282,
103      0.026722848415374756,
104      0.018560435622930527,
105      0.013413295149803162,
106      0.025747407227754593,
107      0.016179025173187256,
108      0.028646979480981827
109    ],
110    "p_value": 0.0081,
111    "mde": 0.004894412730734753,
112    "mde_rel_pct": 9.721250095313414,
113    "verdict": "idea worse (significant)",
114    "system_worked": false
115  },
116  "mechanism_signature": {
117    "prediction": "calibrated compact density predicts nominal E[r2]=d=5",
118    "predicted_mean_radius2": 5.0,
119    "observed_mean_radius2_on_trained_models": 2.6181041300296783,
120    "absolute_error": 2.3818958699703217,
121    "mean_inside_support_fraction": 0.9849999696016312,
122    "confirmed": false
123  }
124}