Conservative Parallel-Edge Decomposition / results.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  1{
  2  "seed": 2870,
  3  "core_math": {
  4    "incidence_shape": [
  5      7,
  6      9
  7    ],
  8    "incidence_column_sum_max_abs": 0.0,
  9    "channel_sweep": [
 10      {
 11        "K": 1,
 12        "abs_total_internal_residual": 1.1102230246251565e-16
 13      },
 14      {
 15        "K": 2,
 16        "abs_total_internal_residual": 0.0
 17      },
 18      {
 19        "K": 4,
 20        "abs_total_internal_residual": 1.3322676295501878e-15
 21      },
 22      {
 23        "K": 8,
 24        "abs_total_internal_residual": 0.0
 25      },
 26      {
 27        "K": 16,
 28        "abs_total_internal_residual": 4.440892098500626e-16
 29      },
 30      {
 31        "K": 32,
 32        "abs_total_internal_residual": 7.105427357601002e-15
 33      }
 34    ],
 35    "leakage_sweep": [
 36      {
 37        "lambda": 0.0,
 38        "observed_abs_residual": 5.551115123125783e-16,
 39        "predicted_abs_residual": 0.0
 40      },
 41      {
 42        "lambda": 0.01,
 43        "observed_abs_residual": 0.009643521566509716,
 44        "predicted_abs_residual": 0.009643521566510067
 45      },
 46      {
 47        "lambda": 0.05,
 48        "observed_abs_residual": 0.04821760783255069,
 49        "predicted_abs_residual": 0.04821760783255033
 50      },
 51      {
 52        "lambda": 0.1,
 53        "observed_abs_residual": 0.09643521566510027,
 54        "predicted_abs_residual": 0.09643521566510066
 55      },
 56      {
 57        "lambda": 0.25,
 58        "observed_abs_residual": 0.24108803916275057,
 59        "predicted_abs_residual": 0.24108803916275165
 60      },
 61      {
 62        "lambda": 0.5,
 63        "observed_abs_residual": 0.4821760783255027,
 64        "predicted_abs_residual": 0.4821760783255033
 65      },
 66      {
 67        "lambda": 1.0,
 68        "observed_abs_residual": 0.964352156651006,
 69        "predicted_abs_residual": 0.9643521566510066
 70      },
 71      {
 72        "lambda": 2.0,
 73        "observed_abs_residual": 1.9287043133020125,
 74        "predicted_abs_residual": 1.9287043133020132
 75      }
 76    ],
 77    "leakage_fitted_slope": 0.9643521566510062,
 78    "leakage_predicted_slope": 0.9643521566510066,
 79    "rollout_sweep": [
 80      {
 81        "steps": 1,
 82        "conservative_abs_mass_drift": 1.942890293094024e-16,
 83        "unconstrained_abs_mass_drift": 0.0070000000000000895,
 84        "predicted_unconstrained_drift": 0.007
 85      },
 86      {
 87        "steps": 2,
 88        "conservative_abs_mass_drift": 3.608224830031759e-16,
 89        "unconstrained_abs_mass_drift": 0.014000000000000151,
 90        "predicted_unconstrained_drift": 0.014
 91      },
 92      {
 93        "steps": 5,
 94        "conservative_abs_mass_drift": 8.049116928532385e-16,
 95        "unconstrained_abs_mass_drift": 0.03500000000000017,
 96        "predicted_unconstrained_drift": 0.035
 97      },
 98      {
 99        "steps": 10,
100        "conservative_abs_mass_drift": 1.8041124150158794e-15,
101        "unconstrained_abs_mass_drift": 0.07000000000000009,
102        "predicted_unconstrained_drift": 0.07
103      },
104      {
105        "steps": 20,
106        "conservative_abs_mass_drift": 2.4424906541753444e-15,
107        "unconstrained_abs_mass_drift": 0.14,
108        "predicted_unconstrained_drift": 0.14
109      },
110      {
111        "steps": 40,
112        "conservative_abs_mass_drift": 3.3306690738754696e-16,
113        "unconstrained_abs_mass_drift": 0.28000000000000036,
114        "predicted_unconstrained_drift": 0.28
115      },
116      {
117        "steps": 80,
118        "conservative_abs_mass_drift": 1.709743457922741e-14,
119        "unconstrained_abs_mass_drift": 0.5600000000000103,
120        "predicted_unconstrained_drift": 0.56
121      }
122    ]
123  },
124  "toy_model": {
125    "device": "cuda",
126    "results": {
127      "conservative": {
128        "test_q_rmse": 0.16049449145793915,
129        "mean_abs_mass_residual": 8.096297676729591e-08,
130        "parameters": 162
131      },
132      "unconstrained": {
133        "test_q_rmse": 0.1806080937385559,
134        "mean_abs_mass_residual": 0.06358527392148972,
135        "parameters": 170
136      }
137    }
138  }
139}