Calorimetric Training Transition Detector / bench_report.json

Failed on benchmark

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "tabular",
  4  "model": "mlp_tiny",
  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": 14.981590270996094
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 9.223435640335083
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 7.291872143745422
 29      }
 30    ],
 31    "full": {
 32      "mean": 7.431148707866669,
 33      "std": 0.5900729884817257,
 34      "per_seed": [
 35        6.959560871124268,
 36        6.490821361541748,
 37        7.417182445526123,
 38        8.29992389678955,
 39        7.356024742126465,
 40        6.94450569152832,
 41        8.157182693481445,
 42        7.82398796081543
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 7.3019766211509705,
 49    "std": 0.7734523375507841,
 50    "per_seed": [
 51      6.782430648803711,
 52      6.035555839538574,
 53      7.4112749099731445,
 54      8.592246055603027,
 55      7.169427394866943,
 56      6.662042140960693,
 57      8.142901420593262,
 58      7.619934558868408
 59    ],
 60    "n": 8,
 61    "sweep": [
 62      {
 63        "cfg": {
 64          "lr": 0.001
 65        },
 66        "mean": 14.74151885509491
 67      },
 68      {
 69        "cfg": {
 70          "lr": 0.003
 71        },
 72        "mean": 8.378218531608582
 73      },
 74      {
 75        "cfg": {
 76          "lr": 0.006
 77        },
 78        "mean": 7.3019766211509705
 79      }
 80    ]
 81  },
 82  "comparison": {
 83    "delta_mean": -0.12917208671569824,
 84    "idea_wins": 7,
 85    "n_pairs": 8,
 86    "per_seed_diffs": [
 87      -0.17713022232055664,
 88      -0.45526552200317383,
 89      -0.005907535552978516,
 90      0.29232215881347656,
 91      -0.18659734725952148,
 92      -0.28246355056762695,
 93      -0.014281272888183594,
 94      -0.20405340194702148
 95    ],
 96    "p_value": 0.1473,
 97    "mde": 0.18619118804819146,
 98    "mde_rel_pct": 2.5055505597820713,
 99    "verdict": "no significant win",
100    "system_worked": false
101  },
102  "mechanism_signature": {
103    "prediction": "temperature/minibatch-noise pulse produces transient excess update dissipation",
104    "predicted": {
105      "pulse_to_baseline_q_ratio": ">1"
106    },
107    "observed_on_trained_models": {
108      "n_pulses": 128,
109      "mean_pulse_to_baseline_q_ratio": 0.8008347272847326,
110      "mean_triggers": 0.0
111    },
112    "confirmed": false,
113    "toy_sanity": {
114      "stable_below_2": true,
115      "C_monotone": true,
116      "C_values": [
117        44.04943522876032,
118        49.68340090799896,
119        56.88773636795182,
120        66.42488931177628,
121        79.64597854178102,
122        99.19381688466098,
123        131.02741214829146,
124        191.96684213596245,
125        354.29487880387836,
126        1511.1844114828386
127      ]
128    }
129  },
130  "sanity_check": {
131    "stable_below_2": true,
132    "C_monotone": true,
133    "C_values": [
134      44.04943522876032,
135      49.68340090799896,
136      56.88773636795182,
137      66.42488931177628,
138      79.64597854178102,
139      99.19381688466098,
140      131.02741214829146,
141      191.96684213596245,
142      354.29487880387836,
143      1511.1844114828386
144    ]
145  }
146}