{ "seed": 1071, "predictions": { "P1_pushforward_exact": "aggregation error is 0 up to floating point, and synonym splitting leaves state mass unchanged", "P2_power_distortion": "raw log-odds scale with gamma; fitted calibration slope a should satisfy gamma*a ~= 1", "P3_calibration_effect": "calibrated TV-to-exact-posterior should be near 0 for every gamma, while raw TV is 0 at gamma=1 and grows away from 1" }, "sanity": { "pushforward_max_abs_error": 0.0, "synonym_split_max_abs_error": 1.1102230246251565e-16, "response_mass": 1.0, "pushed_mass": 1.0 }, "sweep": [ { "gamma": 0.5, "predicted_a": 2.0, "fitted_a_mean": 1.8311318383591928, "gamma_times_fitted_a": 0.9155659191795964, "raw": { "nll": 0.8572267721071046, "brier": 0.5050365069825522, "tv_to_exact_q": 0.1369419752489948, "ece": 0.13228225310990974, "set90_coverage": 0.9986, "mean_set_size": 2.9038 }, "calibrated": { "nll": 0.8008695190016372, "brier": 0.4724195908791922, "tv_to_exact_q": 0.020023923130417648, "ece": 0.019039719148512373, "set90_coverage": 0.9754, "mean_set_size": 2.4738 }, "optimizer_success": false }, { "gamma": 0.75, "predicted_a": 1.3333333333333333, "fitted_a_mean": 1.3110876536290874, "gamma_times_fitted_a": 0.9833157402218156, "raw": { "nll": 0.805427923492955, "brier": 0.4741506019639819, "tv_to_exact_q": 0.061074856125397876, "ece": 0.06562909937882927, "set90_coverage": 0.9872, "mean_set_size": 2.6366 }, "calibrated": { "nll": 0.7915430152128272, "brier": 0.46653125348977226, "tv_to_exact_q": 0.005265335286719836, "ece": 0.009886458166121609, "set90_coverage": 0.9702, "mean_set_size": 2.4068 }, "optimizer_success": true }, { "gamma": 1.0, "predicted_a": 1.0, "fitted_a_mean": 0.9953532162760084, "gamma_times_fitted_a": 0.9953532162760084, "raw": { "nll": 0.7873009438915944, "brier": 0.4666413141016571, "tv_to_exact_q": 2.034366598790971e-17, "ece": 0.012771470251124017, "set90_coverage": 0.9734, "mean_set_size": 2.3876 }, "calibrated": { "nll": 0.7879509007202342, "brier": 0.46677964644116265, "tv_to_exact_q": 0.006602013699090465, "ece": 0.012340774685794783, "set90_coverage": 0.9734, "mean_set_size": 2.392 }, "optimizer_success": true }, { "gamma": 1.5, "predicted_a": 0.6666666666666666, "fitted_a_mean": 0.6530733297958197, "gamma_times_fitted_a": 0.9796099946937296, "raw": { "nll": 0.8414786648391069, "brier": 0.48738748659195336, "tv_to_exact_q": 0.08651805518125863, "ece": 0.08809918856795493, "set90_coverage": 0.9234, "mean_set_size": 2.0218 }, "calibrated": { "nll": 0.8020212579724905, "brier": 0.47353656511301123, "tv_to_exact_q": 0.008298574265717416, "ece": 0.010963217301649434, "set90_coverage": 0.9718, "mean_set_size": 2.4056 }, "optimizer_success": true }, { "gamma": 2.0, "predicted_a": 0.5, "fitted_a_mean": 0.5130615254222709, "gamma_times_fitted_a": 1.0261230508445418, "raw": { "nll": 0.889404327234238, "brier": 0.4903964225682859, "tv_to_exact_q": 0.14200148153616524, "ece": 0.1324196675129269, "set90_coverage": 0.8884, "mean_set_size": 1.778 }, "calibrated": { "nll": 0.7790706624494645, "brier": 0.4598805465689491, "tv_to_exact_q": 0.00697078504150675, "ece": 0.01752786632853536, "set90_coverage": 0.968, "mean_set_size": 2.357 }, "optimizer_success": false } ], "comparison_gamma_2": { "gamma": 2.0, "predicted_a": 0.5, "fitted_a_mean": 0.5130615254222709, "gamma_times_fitted_a": 1.0261230508445418, "raw": { "nll": 0.889404327234238, "brier": 0.4903964225682859, "tv_to_exact_q": 0.14200148153616524, "ece": 0.1324196675129269, "set90_coverage": 0.8884, "mean_set_size": 1.778 }, "calibrated": { "nll": 0.7790706624494645, "brier": 0.4598805465689491, "tv_to_exact_q": 0.00697078504150675, "ece": 0.01752786632853536, "set90_coverage": 0.968, "mean_set_size": 2.357 }, "optimizer_success": false } }