Fourier-Tumble Oscillatory Memory / 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_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": 0.001864717691205442
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "weight_decay": 0.0
 24        },
 25        "mean": 0.0010901302302954718
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.01,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 0.0006186682076076977
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.00054517226817552,
 37      "std": 0.00020361842727135583,
 38      "per_seed": [
 39        0.0004930912400595844,
 40        0.0004765945195686072,
 41        0.0004673648509196937,
 42        0.0010376222198829055,
 43        0.000511663849465549,
 44        0.00027233464061282575,
 45        0.0005571767687797546,
 46        0.0005455300561152399
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 0.0008685752491146559,
 53    "std": 0.0006023882637592906,
 54    "per_seed": [
 55      0.000623027968686074,
 56      0.0002831997990142554,
 57      0.0018796202493831515,
 58      0.00019449168758001179,
 59      0.0017858315259218216,
 60      0.0008392879390157759,
 61      0.000899141828995198,
 62      0.00044400099432095885
 63    ],
 64    "n": 8
 65  },
 66  "comparison": {
 67    "delta_mean": 0.0003234029809391359,
 68    "idea_wins": 3,
 69    "n_pairs": 8,
 70    "per_seed_diffs": [
 71      0.00012993672862648964,
 72      -0.0001933947205543518,
 73      0.0014122553984634578,
 74      -0.0008431305323028937,
 75      0.0012741676764562726,
 76      0.0005669532984029502,
 77      0.00034196506021544337,
 78      -0.000101529061794281
 79    ],
 80    "p_value": 0.26825,
 81    "mde": 0.000631684873015127,
 82    "mde_rel_pct": 115.86885648624994,
 83    "verdict": "no significant win",
 84    "system_worked": false
 85  },
 86  "track_match": "stability/control and recurrent memory -> official dynamics track",
 87  "math_sanity": {
 88    "predictions": {
 89      "abs_Pi_le_1": true,
 90      "Jacobian_norm_le_1": true,
 91      "strict_contraction_when_gamma_positive": true
 92    },
 93    "max_abs_Pi": 1.4901161193847656e-08,
 94    "max_jacobian_norm": 0.799992024898529,
 95    "min_gamma": 0.22315354645252228,
 96    "all_abs_Pi_le_1": true,
 97    "all_jacobian_norms_le_1": true
 98  },
 99  "idea_sweep": [
100    {
101      "cfg": {
102        "lr": 0.001,
103        "weight_decay": 0.0
104      },
105      "result": {
106        "mean": 0.0025068143904718454,
107        "std": 0.004455793347706985,
108        "per_seed": [
109          0.0003443159512244165,
110          0.0006758710369467735,
111          0.0004920198116451502,
112          0.00010601181566016749,
113          0.0013106095138937235,
114          0.0029909624718129635,
115          7.743191963527352e-05,
116          0.014057292602956295
117        ],
118        "n": 8
119      }
120    },
121    {
122      "cfg": {
123        "lr": 0.003,
124        "weight_decay": 0.0
125      },
126      "result": {
127        "mean": 0.0011009546731202136,
128        "std": 0.0013963019006052824,
129        "per_seed": [
130          0.0001262359437532723,
131          0.0005790527211502194,
132          0.004503095522522926,
133          2.4947914425865747e-05,
134          0.001412989222444594,
135          0.00057682313490659,
136          5.747442992287688e-05,
137          0.0015270184958353639
138        ],
139        "n": 8
140      }
141    },
142    {
143      "cfg": {
144        "lr": 0.01,
145        "weight_decay": 0.0
146      },
147      "result": {
148        "mean": 0.0008685752491146559,
149        "std": 0.0006023882637592906,
150        "per_seed": [
151          0.000623027968686074,
152          0.0002831997990142554,
153          0.0018796202493831515,
154          0.00019449168758001179,
155          0.0017858315259218216,
156          0.0008392879390157759,
157          0.000899141828995198,
158          0.00044400099432095885
159        ],
160        "n": 8
161      }
162    }
163  ],
164  "selected_idea_cfg": {
165    "lr": 0.01,
166    "weight_decay": 0.0
167  },
168  "mechanism_signature": {
169    "prediction": "trained Fourier transition has ||rho R||_2=rho<=1 and |Pi_1|<=1",
170    "trained_model_samples": 8,
171    "predicted_max_norm": 0.7995531558990479,
172    "observed_max_homogeneous_ratio": 0.7995532751083374,
173    "observed_max_abs_pi": 0.13136059045791626,
174    "within_tolerance": true,
175    "confirmed": true,
176    "per_seed": [
177      {
178        "seed": 0,
179        "metric": 0.000623027968686074,
180        "predicted_norm": 0.7978268265724182,
181        "observed_max_ratio": 0.7978268265724182,
182        "max_abs_pi": 0.06289801001548767,
183        "min_gamma": 0.22586394846439362
184      },
185      {
186        "seed": 1,
187        "metric": 0.0002831997990142554,
188        "predicted_norm": 0.7982399463653564,
189        "observed_max_ratio": 0.7982401251792908,
190        "max_abs_pi": 0.04295755922794342,
191        "min_gamma": 0.2253459393978119
192      },
193      {
194        "seed": 2,
195        "metric": 0.0018796202493831515,
196        "predicted_norm": 0.7995531558990479,
197        "observed_max_ratio": 0.7995532751083374,
198        "max_abs_pi": 0.09743016958236694,
199        "min_gamma": 0.2237023115158081
200      },
201      {
202        "seed": 3,
203        "metric": 0.00019449168758001179,
204        "predicted_norm": 0.7968804836273193,
205        "observed_max_ratio": 0.7968803644180298,
206        "max_abs_pi": 0.07074946910142899,
207        "min_gamma": 0.22705085575580597
208      },
209      {
210        "seed": 4,
211        "metric": 0.0017858315259218216,
212        "predicted_norm": 0.7966040968894958,
213        "observed_max_ratio": 0.7966041564941406,
214        "max_abs_pi": 0.13136059045791626,
215        "min_gamma": 0.22739754617214203
216      },
217      {
218        "seed": 5,
219        "metric": 0.0008392879390157759,
220        "predicted_norm": 0.7936066389083862,
221        "observed_max_ratio": 0.7936067581176758,
222        "max_abs_pi": 0.07851097732782364,
223        "min_gamma": 0.23116742074489594
224      },
225      {
226        "seed": 6,
227        "metric": 0.000899141828995198,
228        "predicted_norm": 0.7972120046615601,
229        "observed_max_ratio": 0.7972121238708496,
230        "max_abs_pi": 0.07305731624364853,
231        "min_gamma": 0.2266346663236618
232      },
233      {
234        "seed": 7,
235        "metric": 0.00044400099432095885,
236        "predicted_norm": 0.7972738146781921,
237        "observed_max_ratio": 0.7972738146781921,
238        "max_abs_pi": 0.07366906106472015,
239        "min_gamma": 0.22655726969242096
240      }
241    ]
242  }
243}