Permutation-Symmetric Quadratic Module / bench_report.json

Failed on benchmark

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "tabular",
  4  "model": "custom_branch_mlp",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01,
 10      "weight_decay": 0.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 12.950589418411255
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "weight_decay": 0.0
 24        },
 25        "mean": 6.888427972793579
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.01,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 6.682573318481445
 33      }
 34    ],
 35    "full": {
 36      "mean": 6.839571237564087,
 37      "std": 0.3468641081837492,
 38      "per_seed": [
 39        6.566773414611816,
 40        6.700428485870361,
 41        6.708205223083496,
 42        6.754886150360107,
 43        6.259535312652588,
 44        7.132112979888916,
 45        7.337502479553223,
 46        7.2571258544921875
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "best_cfg": {
 53      "lr": 0.01,
 54      "weight_decay": 0.0
 55    },
 56    "sweep": [
 57      {
 58        "cfg": {
 59          "lr": 0.001,
 60          "weight_decay": 0.0
 61        },
 62        "mean": 13.624429821968079
 63      },
 64      {
 65        "cfg": {
 66          "lr": 0.003,
 67          "weight_decay": 0.0
 68        },
 69        "mean": 7.040245294570923
 70      },
 71      {
 72        "cfg": {
 73          "lr": 0.01,
 74          "weight_decay": 0.0
 75        },
 76        "mean": 6.819535434246063
 77      }
 78    ],
 79    "per_seed": [
 80      6.49015998840332,
 81      6.686732769012451,
 82      6.5792107582092285,
 83      6.80715799331665,
 84      6.233836650848389,
 85      7.175884246826172,
 86      7.312864780426025,
 87      7.2704362869262695
 88    ],
 89    "mean": 6.819535434246063,
 90    "std": 0.3709357840355057,
 91    "n": 8
 92  },
 93  "comparison": {
 94    "delta_mean": -0.02003580331802368,
 95    "idea_wins": 5,
 96    "n_pairs": 8,
 97    "per_seed_diffs": [
 98      -0.0766134262084961,
 99      -0.013695716857910156,
100      -0.12899446487426758,
101      0.05227184295654297,
102      -0.02569866180419922,
103      0.04377126693725586,
104      -0.024637699127197266,
105      0.013310432434082031
106    ],
107    "p_value": 0.4109,
108    "mde": 0.05054426158733883,
109    "mde_rel_pct": 0.7389975165364338,
110    "verdict": "no measurable effect",
111    "system_worked": false
112  },
113  "mechanism_signature": {
114    "math_check": {
115      "epsilon_values": [
116        0.001,
117        0.002130941015366798,
118        0.004540909610972475,
119        0.009676410537094535,
120        0.020619860095022202,
121        0.043939705607607904,
122        0.09363292088239417,
123        0.19952623149688797
124      ],
125      "remainder_values": [
126        1.9010061570122444e-08,
127        1.8393117422730257e-07,
128        1.7794226864628757e-06,
129        1.7210834407346843e-05,
130        0.00016639443047186702,
131        0.001607205940536596,
132        0.01551344151929597,
133        0.15093468018953293
134      ],
135      "observed_remainder_slope": 2.9993339395602745,
136      "predicted_remainder_slope": 3.0,
137      "gram_invariance_max_error": 5.1514348342607263e-14,
138      "prediction_pass": true
139    },
140    "trained_model": {
141      "source": "trained baseline model on tabular test inputs",
142      "epsilon_values": [
143        0.0031622776601683794,
144        0.006072021956909884,
145        0.011659144011798317,
146        0.0223872113856834,
147        0.042986623470822764,
148        0.08254041852680181,
149        0.15848931924611134
150      ],
151      "observed_remainder_slope": 3.0196060096593453,
152      "predicted_remainder_slope": 3.0,
153      "permutation_max_error": 1.9073486328125e-06,
154      "confirmed": true
155    },
156    "custom_track": null
157  },
158  "protocol_notes": {
159    "paired_seeds": [
160      0,
161      1,
162      2,
163      3,
164      4,
165      5,
166      6,
167      7
168    ],
169    "n_train": 800,
170    "n_test": 400,
171    "architecture": "same width-64 exchangeable branch MLP; only phi vs quadratic Taylor block differs",
172    "baseline_grid": [
173      {
174        "lr": 0.001,
175        "weight_decay": 0.0
176      },
177      {
178        "lr": 0.003,
179        "weight_decay": 0.0
180      },
181      {
182        "lr": 0.01,
183        "weight_decay": 0.0
184      }
185    ],
186    "idea_grid": [
187      {
188        "lr": 0.001,
189        "weight_decay": 0.0
190      },
191      {
192        "lr": 0.003,
193        "weight_decay": 0.0
194      },
195      {
196        "lr": 0.01,
197        "weight_decay": 0.0
198      }
199    ]
200  }
201}