Exponential-Map Stochastic Residual Layer / bench_report.json

✓✓ Beats tuned baseline

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "sphere_one_step_dynamics",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01,
 10      "h": 0.12
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "h": 0.06
 17        },
 18        "mean": 0.00018889913917519152
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "h": 0.12
 24        },
 25        "mean": 8.064255553108524e-05
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.001,
 30          "h": 0.24
 31        },
 32        "mean": 3.386509808933624e-05
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "h": 0.06
 38        },
 39        "mean": 1.4137278640191653e-05
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.003,
 44          "h": 0.12
 45        },
 46        "mean": 5.718101306229073e-06
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.003,
 51          "h": 0.24
 52        },
 53        "mean": 3.6321278003015323e-06
 54      },
 55      {
 56        "cfg": {
 57          "lr": 0.01,
 58          "h": 0.06
 59        },
 60        "mean": 2.5334072688565357e-06
 61      },
 62      {
 63        "cfg": {
 64          "lr": 0.01,
 65          "h": 0.12
 66        },
 67        "mean": 1.9908102331100963e-06
 68      },
 69      {
 70        "cfg": {
 71          "lr": 0.01,
 72          "h": 0.24
 73        },
 74        "mean": 2.2787774582866405e-06
 75      }
 76    ],
 77    "full": {
 78      "mean": 1.9605001710942815e-06,
 79      "std": 4.787010942122163e-07,
 80      "per_seed": [
 81        1.5833142015253543e-06,
 82        1.5794613545949687e-06,
 83        2.8728609322570264e-06,
 84        1.9276044440630358e-06,
 85        2.232864062534645e-06,
 86        2.1574342099484056e-06,
 87        2.126187155226944e-06,
 88        1.2042750086038723e-06
 89      ],
 90      "n": 8
 91    }
 92  },
 93  "idea": {
 94    "mean": 1.8447870697713142e-07,
 95    "std": 3.449141553482167e-08,
 96    "per_seed": [
 97      1.9253596406088036e-07,
 98      2.091002073711934e-07,
 99      2.2312131875423802e-07,
100      1.0322276722263268e-07,
101      1.9205077705919393e-07,
102      1.9699885456248012e-07,
103      1.9496692971188168e-07,
104      1.638328370745512e-07
105    ],
106    "n": 8
107  },
108  "comparison": {
109    "delta_mean": -1.7760214641171501e-06,
110    "idea_wins": 8,
111    "n_pairs": 8,
112    "per_seed_diffs": [
113      -1.390778237464474e-06,
114      -1.3703611472237753e-06,
115      -2.6497396135027884e-06,
116      -1.824381676840403e-06,
117      -2.0408132854754513e-06,
118      -1.9604353553859255e-06,
119      -1.9312202255150623e-06,
120      -1.040442171529321e-06
121    ],
122    "p_value": 0.0081,
123    "mde": 4.178357032172272e-07,
124    "mde_rel_pct": 21.31270934722776,
125    "verdict": "idea better (significant)",
126    "system_worked": true
127  },
128  "mechanism_signature": {
129    "trained_model_seed": 0,
130    "baseline_mean_abs_norm_error": 0.0008932742057368159,
131    "idea_mean_abs_norm_error": 1.7583369427143225e-08,
132    "predicted_idea_lower_constraint_error": true,
133    "confirmed": true
134  },
135  "custom_track": {
136    "name": "sphere_one_step_dynamics",
137    "file": "sphere_track.py",
138    "domain": "geometric_dynamics"
139  }
140}