Dimension-Free Brenier Transport Layer / bench_report.json

Mechanism confirmed, baseline not beaten

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": {
  9      "lr": 0.01,
 10      "mean_metric": 1.0635505304890103e-05
 11    },
 12    "sweep": {
 13      "0.001": {
 14        "lr": 0.001,
 15        "mean_metric": 0.0003965005817008205,
 16        "per_seed": [
 17          0.0003344195138197392,
 18          0.00048574263928458095,
 19          0.00034076091833412647,
 20          0.00042507925536483526
 21        ]
 22      },
 23      "0.003": {
 24        "lr": 0.003,
 25        "mean_metric": 3.096502723565209e-05,
 26        "per_seed": [
 27          3.235309850424528e-05,
 28          3.7017063732491806e-05,
 29          2.8209653464728035e-05,
 30          2.6280293241143227e-05
 31        ]
 32      },
 33      "0.01": {
 34        "lr": 0.01,
 35        "mean_metric": 1.0635505304890103e-05,
 36        "per_seed": [
 37          8.224938937928528e-06,
 38          1.7855567421065643e-05,
 39          7.3662031354615465e-06,
 40          9.095311725104693e-06
 41        ]
 42      }
 43    },
 44    "full": {
 45      "per_seed": [
 46        8.224938937928528e-06,
 47        1.7855567421065643e-05,
 48        7.3662031354615465e-06,
 49        9.095311725104693e-06,
 50        1.3408420272753574e-05,
 51        1.0250218110741116e-05,
 52        9.51075980992755e-06,
 53        7.620315045642201e-06
 54      ],
 55      "details": [
 56        {
 57          "metric": 8.224938937928528e-06,
 58          "jacobian": 0.4901633560657501,
 59          "final_train_loss": 1.2194402166642248e-05
 60        },
 61        {
 62          "metric": 1.7855567421065643e-05,
 63          "jacobian": 0.48987317085266113,
 64          "final_train_loss": 1.6715545440092684e-05
 65        },
 66        {
 67          "metric": 7.3662031354615465e-06,
 68          "jacobian": 0.5534982085227966,
 69          "final_train_loss": 1.0150326805160147e-05
 70        },
 71        {
 72          "metric": 9.095311725104693e-06,
 73          "jacobian": 0.5387848615646362,
 74          "final_train_loss": 1.4617473374528345e-05
 75        },
 76        {
 77          "metric": 1.3408420272753574e-05,
 78          "jacobian": 0.5020041465759277,
 79          "final_train_loss": 1.4221523815649562e-05
 80        },
 81        {
 82          "metric": 1.0250218110741116e-05,
 83          "jacobian": 0.49134495854377747,
 84          "final_train_loss": 1.2912527137814323e-05
 85        },
 86        {
 87          "metric": 9.51075980992755e-06,
 88          "jacobian": 0.521506667137146,
 89          "final_train_loss": 9.340690652607008e-06
 90        },
 91        {
 92          "metric": 7.620315045642201e-06,
 93          "jacobian": 0.5639464855194092,
 94          "final_train_loss": 7.38364438075223e-06
 95        }
 96      ],
 97      "config": {
 98        "lr": 0.01,
 99        "epochs": 12
100      }
101    }
102  },
103  "idea": {
104    "per_seed": [
105      8.224938937928528e-06,
106      1.7855567421065643e-05,
107      7.3662031354615465e-06,
108      9.095311725104693e-06,
109      1.3408420272753574e-05,
110      1.0250218110741116e-05,
111      9.51075980992755e-06,
112      7.620315045642201e-06
113    ],
114    "details": [
115      {
116        "metric": 8.224938937928528e-06,
117        "jacobian": 0.4901633560657501,
118        "train_probe_peak": 0.48664137721061707,
119        "cap": 2.0334276480858615,
120        "cap_events": 0
121      },
122      {
123        "metric": 1.7855567421065643e-05,
124        "jacobian": 0.48987317085266113,
125        "train_probe_peak": 0.493243932723999,
126        "cap": 2.0334276480858615,
127        "cap_events": 0
128      },
129      {
130        "metric": 7.3662031354615465e-06,
131        "jacobian": 0.5534982085227966,
132        "train_probe_peak": 0.5459591150283813,
133        "cap": 2.0334276480858615,
134        "cap_events": 0
135      },
136      {
137        "metric": 9.095311725104693e-06,
138        "jacobian": 0.5387848615646362,
139        "train_probe_peak": 0.5404819250106812,
140        "cap": 2.0334276480858615,
141        "cap_events": 0
142      },
143      {
144        "metric": 1.3408420272753574e-05,
145        "jacobian": 0.5020041465759277,
146        "train_probe_peak": 0.502926766872406,
147        "cap": 2.0334276480858615,
148        "cap_events": 0
149      },
150      {
151        "metric": 1.0250218110741116e-05,
152        "jacobian": 0.49134495854377747,
153        "train_probe_peak": 0.4878369867801666,
154        "cap": 2.0334276480858615,
155        "cap_events": 0
156      },
157      {
158        "metric": 9.51075980992755e-06,
159        "jacobian": 0.521506667137146,
160        "train_probe_peak": 0.5264161825180054,
161        "cap": 2.0334276480858615,
162        "cap_events": 0
163      },
164      {
165        "metric": 7.620315045642201e-06,
166        "jacobian": 0.5639464855194092,
167        "train_probe_peak": 0.5624905824661255,
168        "cap": 2.0334276480858615,
169        "cap_events": 0
170      }
171    ],
172    "config": {
173      "lr": 0.01,
174      "cap": 2.0334276480858615,
175      "mean_metric": 1.0635505304890103e-05,
176      "per_seed": [
177        8.224938937928528e-06,
178        1.7855567421065643e-05,
179        7.3662031354615465e-06,
180        9.095311725104693e-06
181      ]
182    }
183  },
184  "comparison": {
185    "delta_mean": 0.0,
186    "idea_wins": 0,
187    "n_pairs": 8,
188    "per_seed_diffs": [
189      0.0,
190      0.0,
191      0.0,
192      0.0,
193      0.0,
194      0.0,
195      0.0,
196      0.0
197    ],
198    "p_value": 1.0,
199    "mde": 0.0,
200    "mde_rel_pct": 0.0,
201    "verdict": "no measurable effect",
202    "system_worked": false
203  },
204  "budget": {
205    "epochs": 12,
206    "batch": 128,
207    "lr_grid": [
208      0.001,
209      0.003,
210      0.01
211    ]
212  },
213  "selection": {
214    "idea_best": {
215      "lr": 0.01,
216      "cap": 2.0334276480858615,
217      "mean_metric": 1.0635505304890103e-05,
218      "per_seed": [
219        8.224938937928528e-06,
220        1.7855567421065643e-05,
221        7.3662031354615465e-06,
222        9.095311725104693e-06
223      ]
224    }
225  },
226  "mechanism_signature": {
227    "prediction": "certificate cap limits Jacobian operator norm",
228    "predicted_max_jacobian": 2.0334276480858615,
229    "observed_idea_max_jacobian": 0.5639464855194092,
230    "observed_baseline_max_jacobian": 0.5639464855194092,
231    "confirmed": true
232  }
233}