1143 lines
31 KiB
JSON
1143 lines
31 KiB
JSON
{
|
|
"decision": {
|
|
"current_surface_can_show_selection_contribution": false,
|
|
"prospective_target": "Test whether internal features from a short, shared real probe reduce the number or duration of full frontier evaluations relative to an outcome-only model given the same probe.",
|
|
"reason": "The strongest frozen-calibrated SLO reading reaches zero real regret after real evaluation of its first two-cell tie bucket. A method that requires one calibration probe and one final verification cannot use this single task to demonstrate fewer real cell evaluations."
|
|
},
|
|
"next_anchor_prediction": {
|
|
"estimand": "next-anchor feasibility from the full current-anchor summary",
|
|
"external_features": [
|
|
"log_current_rate_per_gpu",
|
|
"log_next_over_current_rate",
|
|
"log2_tp",
|
|
"log2_mns",
|
|
"current_pass_rate",
|
|
"ttft_max_over_6s",
|
|
"tpot_max_over_50ms",
|
|
"exact_output_fraction",
|
|
"early_stopped"
|
|
],
|
|
"headline": {
|
|
"bootstrap": {
|
|
"accuracy_delta_instrumentation_minus_outcome": {
|
|
"ci95": [
|
|
0.0,
|
|
0.125
|
|
],
|
|
"point": 0.040000000000000036
|
|
},
|
|
"brier_delta_instrumentation_minus_outcome": {
|
|
"ci95": [
|
|
-0.019846177751851126,
|
|
0.02199527924137797
|
|
],
|
|
"point": -0.0019563088303256454
|
|
},
|
|
"replicates": 10000,
|
|
"seed": 20260714,
|
|
"semantics": "group bootstrap over cells; diagnostic confidence interval"
|
|
},
|
|
"instrumentation_aware": {
|
|
"accuracy": 0.88,
|
|
"balanced_accuracy": 0.8576388888888888,
|
|
"brier": 0.1031574576574755,
|
|
"confusion": {
|
|
"false_negative": 1,
|
|
"false_positive": 2,
|
|
"true_negative": 7,
|
|
"true_positive": 15
|
|
},
|
|
"log_loss": 0.327997342465158
|
|
},
|
|
"outcome_only": {
|
|
"accuracy": 0.84,
|
|
"balanced_accuracy": 0.8020833333333333,
|
|
"brier": 0.10511376648780114,
|
|
"confusion": {
|
|
"false_negative": 1,
|
|
"false_positive": 3,
|
|
"true_negative": 6,
|
|
"true_positive": 15
|
|
},
|
|
"log_loss": 0.3385851118121996
|
|
},
|
|
"paired_correctness": {
|
|
"both_correct": 21,
|
|
"both_wrong": 3,
|
|
"instrumentation_only_correct": 1,
|
|
"mcnemar_exact_two_sided_p": 1.0,
|
|
"outcome_only_correct": 0
|
|
}
|
|
},
|
|
"headline_regularization": 1.0,
|
|
"instrumentation_features": [
|
|
"waiting_mean",
|
|
"waiting_max",
|
|
"decode_batch_mean",
|
|
"decode_batch_cv",
|
|
"kv_usage_mean",
|
|
"kv_usage_max",
|
|
"graph_none_share",
|
|
"graph_full_share",
|
|
"padding_fraction",
|
|
"prefill_token_fraction",
|
|
"model_steps_per_second"
|
|
],
|
|
"label_sanity": {
|
|
"distinct_n": 2,
|
|
"max": 1.0,
|
|
"min": 0.0,
|
|
"n": 25,
|
|
"negative": 9,
|
|
"positive": 16
|
|
},
|
|
"model": "L2 logistic regression with train-fold standardization",
|
|
"regularization_sensitivity": {
|
|
"0.1": {
|
|
"bootstrap": {
|
|
"accuracy_delta_instrumentation_minus_outcome": {
|
|
"ci95": [
|
|
0.0,
|
|
0.13636363636363646
|
|
],
|
|
"point": 0.040000000000000036
|
|
},
|
|
"brier_delta_instrumentation_minus_outcome": {
|
|
"ci95": [
|
|
-0.029611344558941783,
|
|
0.030283692179295475
|
|
],
|
|
"point": -0.0013643768871213907
|
|
},
|
|
"replicates": 10000,
|
|
"seed": 20260714,
|
|
"semantics": "group bootstrap over cells; diagnostic confidence interval"
|
|
},
|
|
"instrumentation_aware": {
|
|
"accuracy": 0.92,
|
|
"balanced_accuracy": 0.9131944444444444,
|
|
"brier": 0.08887235975611406,
|
|
"confusion": {
|
|
"false_negative": 1,
|
|
"false_positive": 1,
|
|
"true_negative": 8,
|
|
"true_positive": 15
|
|
},
|
|
"log_loss": 0.3015353076332512
|
|
},
|
|
"outcome_only": {
|
|
"accuracy": 0.88,
|
|
"balanced_accuracy": 0.8819444444444444,
|
|
"brier": 0.09023673664323545,
|
|
"confusion": {
|
|
"false_negative": 2,
|
|
"false_positive": 1,
|
|
"true_negative": 8,
|
|
"true_positive": 14
|
|
},
|
|
"log_loss": 0.2927458553148107
|
|
},
|
|
"paired_correctness": {
|
|
"both_correct": 22,
|
|
"both_wrong": 2,
|
|
"instrumentation_only_correct": 1,
|
|
"mcnemar_exact_two_sided_p": 1.0,
|
|
"outcome_only_correct": 0
|
|
}
|
|
},
|
|
"1.0": {
|
|
"bootstrap": {
|
|
"accuracy_delta_instrumentation_minus_outcome": {
|
|
"ci95": [
|
|
0.0,
|
|
0.125
|
|
],
|
|
"point": 0.040000000000000036
|
|
},
|
|
"brier_delta_instrumentation_minus_outcome": {
|
|
"ci95": [
|
|
-0.019846177751851126,
|
|
0.02199527924137797
|
|
],
|
|
"point": -0.0019563088303256454
|
|
},
|
|
"replicates": 10000,
|
|
"seed": 20260714,
|
|
"semantics": "group bootstrap over cells; diagnostic confidence interval"
|
|
},
|
|
"instrumentation_aware": {
|
|
"accuracy": 0.88,
|
|
"balanced_accuracy": 0.8576388888888888,
|
|
"brier": 0.1031574576574755,
|
|
"confusion": {
|
|
"false_negative": 1,
|
|
"false_positive": 2,
|
|
"true_negative": 7,
|
|
"true_positive": 15
|
|
},
|
|
"log_loss": 0.327997342465158
|
|
},
|
|
"outcome_only": {
|
|
"accuracy": 0.84,
|
|
"balanced_accuracy": 0.8020833333333333,
|
|
"brier": 0.10511376648780114,
|
|
"confusion": {
|
|
"false_negative": 1,
|
|
"false_positive": 3,
|
|
"true_negative": 6,
|
|
"true_positive": 15
|
|
},
|
|
"log_loss": 0.3385851118121996
|
|
},
|
|
"paired_correctness": {
|
|
"both_correct": 21,
|
|
"both_wrong": 3,
|
|
"instrumentation_only_correct": 1,
|
|
"mcnemar_exact_two_sided_p": 1.0,
|
|
"outcome_only_correct": 0
|
|
}
|
|
},
|
|
"10.0": {
|
|
"bootstrap": {
|
|
"accuracy_delta_instrumentation_minus_outcome": {
|
|
"ci95": [
|
|
-0.13636363636363646,
|
|
0.0
|
|
],
|
|
"point": -0.039999999999999925
|
|
},
|
|
"brier_delta_instrumentation_minus_outcome": {
|
|
"ci95": [
|
|
-0.04125619566309649,
|
|
0.012885694983237113
|
|
],
|
|
"point": -0.013003224609162978
|
|
},
|
|
"replicates": 10000,
|
|
"seed": 20260714,
|
|
"semantics": "group bootstrap over cells; diagnostic confidence interval"
|
|
},
|
|
"instrumentation_aware": {
|
|
"accuracy": 0.8,
|
|
"balanced_accuracy": 0.7465277777777778,
|
|
"brier": 0.16045670460468575,
|
|
"confusion": {
|
|
"false_negative": 1,
|
|
"false_positive": 4,
|
|
"true_negative": 5,
|
|
"true_positive": 15
|
|
},
|
|
"log_loss": 0.4963310825527336
|
|
},
|
|
"outcome_only": {
|
|
"accuracy": 0.84,
|
|
"balanced_accuracy": 0.7777777777777778,
|
|
"brier": 0.17345992921384873,
|
|
"confusion": {
|
|
"false_negative": 0,
|
|
"false_positive": 4,
|
|
"true_negative": 5,
|
|
"true_positive": 16
|
|
},
|
|
"log_loss": 0.5254987874287221
|
|
},
|
|
"paired_correctness": {
|
|
"both_correct": 20,
|
|
"both_wrong": 4,
|
|
"instrumentation_only_correct": 0,
|
|
"mcnemar_exact_two_sided_p": 1.0,
|
|
"outcome_only_correct": 1
|
|
}
|
|
}
|
|
},
|
|
"sensitivity_summary": {
|
|
"accuracy_delta_min_max": [
|
|
-0.039999999999999925,
|
|
0.040000000000000036
|
|
],
|
|
"brier_delta_min_max": [
|
|
-0.013003224609162978,
|
|
-0.0013643768871213907
|
|
],
|
|
"incremental_signal_verdict": "NEEDS_PROSPECTIVE_EVIDENCE"
|
|
},
|
|
"split": "leave-one-cell-out",
|
|
"status": "RETROSPECTIVE_DIAGNOSTIC_ONLY"
|
|
},
|
|
"provenance": {
|
|
"phase6_metrics": "/home/gahow/phd/aituner/runs/opprof-phase6/phase6/metrics.json",
|
|
"phase6_sha256": "290ba7fcb8727291166de7e4d47afdc84e230052495c81dd087db0ace9f93a16",
|
|
"simfid_metrics": "/home/gahow/phd/replayserve/runs/simfid_s2rb/results/metrics.json",
|
|
"simfid_sha256": "55edb37d5692e979ab6f6dc6c65913a9db0aa0a836c350e4c05d9c38eee78206"
|
|
},
|
|
"sanity": {
|
|
"invariants": {
|
|
"probabilities_bounded": true,
|
|
"same_cells_all_readings": true,
|
|
"scores_nonnegative": true,
|
|
"transition_features_finite": true
|
|
},
|
|
"real_scores": {
|
|
"distinct_n": 8,
|
|
"max": 3.283333333333333,
|
|
"min": 1.2833333333333334,
|
|
"n": 12
|
|
},
|
|
"red_flags": [],
|
|
"simulator_readings": 4,
|
|
"transition_cells": 12,
|
|
"transitions": 25
|
|
},
|
|
"schema": "fidelity-headroom-v1",
|
|
"scope": "retrospective single-workload premise audit; not prospective contribution evidence",
|
|
"status": "PASS",
|
|
"topk_headroom": {
|
|
"frozen-calibrated/SLO-gated": {
|
|
"minimum_k": {
|
|
"five_percent": {
|
|
"expanded_k": 2,
|
|
"nominal_k": 1
|
|
},
|
|
"one_percent": {
|
|
"expanded_k": 2,
|
|
"nominal_k": 1
|
|
},
|
|
"zero": {
|
|
"expanded_k": 2,
|
|
"nominal_k": 1
|
|
}
|
|
},
|
|
"points": [
|
|
{
|
|
"candidates": [
|
|
"tp2_mns32",
|
|
"tp2_mns64"
|
|
],
|
|
"expanded_k": 2,
|
|
"nominal_k": 1,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp2_mns32",
|
|
"tp2_mns64"
|
|
],
|
|
"expanded_k": 2,
|
|
"nominal_k": 2,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp4_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64"
|
|
],
|
|
"expanded_k": 5,
|
|
"nominal_k": 3,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp4_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64"
|
|
],
|
|
"expanded_k": 5,
|
|
"nominal_k": 4,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp4_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64"
|
|
],
|
|
"expanded_k": 5,
|
|
"nominal_k": 5,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp4_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp1_mns16",
|
|
"tp1_mns32",
|
|
"tp1_mns64"
|
|
],
|
|
"expanded_k": 8,
|
|
"nominal_k": 6,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp4_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp1_mns16",
|
|
"tp1_mns32",
|
|
"tp1_mns64"
|
|
],
|
|
"expanded_k": 8,
|
|
"nominal_k": 7,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp4_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp1_mns16",
|
|
"tp1_mns32",
|
|
"tp1_mns64"
|
|
],
|
|
"expanded_k": 8,
|
|
"nominal_k": 8,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp4_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp1_mns16",
|
|
"tp1_mns32",
|
|
"tp1_mns64",
|
|
"tp2_mns16"
|
|
],
|
|
"expanded_k": 9,
|
|
"nominal_k": 9,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp4_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp1_mns16",
|
|
"tp1_mns32",
|
|
"tp1_mns64",
|
|
"tp2_mns16",
|
|
"tp2_mns8"
|
|
],
|
|
"expanded_k": 10,
|
|
"nominal_k": 10,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp4_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp1_mns16",
|
|
"tp1_mns32",
|
|
"tp1_mns64",
|
|
"tp2_mns16",
|
|
"tp2_mns8",
|
|
"tp1_mns8"
|
|
],
|
|
"expanded_k": 11,
|
|
"nominal_k": 11,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp4_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp1_mns16",
|
|
"tp1_mns32",
|
|
"tp1_mns64",
|
|
"tp2_mns16",
|
|
"tp2_mns8",
|
|
"tp1_mns8",
|
|
"tp4_mns8"
|
|
],
|
|
"expanded_k": 12,
|
|
"nominal_k": 12,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
}
|
|
],
|
|
"real_best": 3.283333333333333
|
|
},
|
|
"frozen-calibrated/throughput-proxy": {
|
|
"minimum_k": {
|
|
"five_percent": {
|
|
"expanded_k": 4,
|
|
"nominal_k": 3
|
|
},
|
|
"one_percent": {
|
|
"expanded_k": 4,
|
|
"nominal_k": 3
|
|
},
|
|
"zero": {
|
|
"expanded_k": 4,
|
|
"nominal_k": 3
|
|
}
|
|
},
|
|
"points": [
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64"
|
|
],
|
|
"expanded_k": 1,
|
|
"nominal_k": 1,
|
|
"real_regret": 0.30456852791878175,
|
|
"selected_cell_after_real_final": "tp1_mns64",
|
|
"selected_real_score": 2.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32"
|
|
],
|
|
"expanded_k": 2,
|
|
"nominal_k": 2,
|
|
"real_regret": 0.30456852791878175,
|
|
"selected_cell_after_real_final": "tp1_mns64",
|
|
"selected_real_score": 2.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns32",
|
|
"tp2_mns64"
|
|
],
|
|
"expanded_k": 4,
|
|
"nominal_k": 3,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns32",
|
|
"tp2_mns64"
|
|
],
|
|
"expanded_k": 4,
|
|
"nominal_k": 4,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16"
|
|
],
|
|
"expanded_k": 5,
|
|
"nominal_k": 5,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64"
|
|
],
|
|
"expanded_k": 7,
|
|
"nominal_k": 6,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64"
|
|
],
|
|
"expanded_k": 7,
|
|
"nominal_k": 7,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp4_mns16"
|
|
],
|
|
"expanded_k": 8,
|
|
"nominal_k": 8,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp4_mns16",
|
|
"tp2_mns16"
|
|
],
|
|
"expanded_k": 9,
|
|
"nominal_k": 9,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp4_mns16",
|
|
"tp2_mns16",
|
|
"tp1_mns8"
|
|
],
|
|
"expanded_k": 10,
|
|
"nominal_k": 10,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp4_mns16",
|
|
"tp2_mns16",
|
|
"tp1_mns8",
|
|
"tp2_mns8"
|
|
],
|
|
"expanded_k": 11,
|
|
"nominal_k": 11,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16",
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp4_mns16",
|
|
"tp2_mns16",
|
|
"tp1_mns8",
|
|
"tp2_mns8",
|
|
"tp4_mns8"
|
|
],
|
|
"expanded_k": 12,
|
|
"nominal_k": 12,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
}
|
|
],
|
|
"real_best": 3.283333333333333
|
|
},
|
|
"uncalibrated/SLO-gated": {
|
|
"minimum_k": {
|
|
"five_percent": {
|
|
"expanded_k": 12,
|
|
"nominal_k": 8
|
|
},
|
|
"one_percent": {
|
|
"expanded_k": 12,
|
|
"nominal_k": 8
|
|
},
|
|
"zero": {
|
|
"expanded_k": 12,
|
|
"nominal_k": 8
|
|
}
|
|
},
|
|
"points": [
|
|
{
|
|
"candidates": [
|
|
"tp4_mns32",
|
|
"tp4_mns64"
|
|
],
|
|
"expanded_k": 2,
|
|
"nominal_k": 1,
|
|
"real_regret": 0.25634517766497456,
|
|
"selected_cell_after_real_final": "tp4_mns64",
|
|
"selected_real_score": 2.441666666666667
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp4_mns32",
|
|
"tp4_mns64"
|
|
],
|
|
"expanded_k": 2,
|
|
"nominal_k": 2,
|
|
"real_regret": 0.25634517766497456,
|
|
"selected_cell_after_real_final": "tp4_mns64",
|
|
"selected_real_score": 2.441666666666667
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp2_mns16"
|
|
],
|
|
"expanded_k": 3,
|
|
"nominal_k": 3,
|
|
"real_regret": 0.25634517766497456,
|
|
"selected_cell_after_real_final": "tp4_mns64",
|
|
"selected_real_score": 2.441666666666667
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp2_mns16",
|
|
"tp1_mns16",
|
|
"tp1_mns32",
|
|
"tp1_mns64",
|
|
"tp1_mns8"
|
|
],
|
|
"expanded_k": 7,
|
|
"nominal_k": 4,
|
|
"real_regret": 0.25634517766497456,
|
|
"selected_cell_after_real_final": "tp4_mns64",
|
|
"selected_real_score": 2.441666666666667
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp2_mns16",
|
|
"tp1_mns16",
|
|
"tp1_mns32",
|
|
"tp1_mns64",
|
|
"tp1_mns8"
|
|
],
|
|
"expanded_k": 7,
|
|
"nominal_k": 5,
|
|
"real_regret": 0.25634517766497456,
|
|
"selected_cell_after_real_final": "tp4_mns64",
|
|
"selected_real_score": 2.441666666666667
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp2_mns16",
|
|
"tp1_mns16",
|
|
"tp1_mns32",
|
|
"tp1_mns64",
|
|
"tp1_mns8"
|
|
],
|
|
"expanded_k": 7,
|
|
"nominal_k": 6,
|
|
"real_regret": 0.25634517766497456,
|
|
"selected_cell_after_real_final": "tp4_mns64",
|
|
"selected_real_score": 2.441666666666667
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp2_mns16",
|
|
"tp1_mns16",
|
|
"tp1_mns32",
|
|
"tp1_mns64",
|
|
"tp1_mns8"
|
|
],
|
|
"expanded_k": 7,
|
|
"nominal_k": 7,
|
|
"real_regret": 0.25634517766497456,
|
|
"selected_cell_after_real_final": "tp4_mns64",
|
|
"selected_real_score": 2.441666666666667
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp2_mns16",
|
|
"tp1_mns16",
|
|
"tp1_mns32",
|
|
"tp1_mns64",
|
|
"tp1_mns8",
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp2_mns8",
|
|
"tp4_mns16",
|
|
"tp4_mns8"
|
|
],
|
|
"expanded_k": 12,
|
|
"nominal_k": 8,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp2_mns16",
|
|
"tp1_mns16",
|
|
"tp1_mns32",
|
|
"tp1_mns64",
|
|
"tp1_mns8",
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp2_mns8",
|
|
"tp4_mns16",
|
|
"tp4_mns8"
|
|
],
|
|
"expanded_k": 12,
|
|
"nominal_k": 9,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp2_mns16",
|
|
"tp1_mns16",
|
|
"tp1_mns32",
|
|
"tp1_mns64",
|
|
"tp1_mns8",
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp2_mns8",
|
|
"tp4_mns16",
|
|
"tp4_mns8"
|
|
],
|
|
"expanded_k": 12,
|
|
"nominal_k": 10,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp2_mns16",
|
|
"tp1_mns16",
|
|
"tp1_mns32",
|
|
"tp1_mns64",
|
|
"tp1_mns8",
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp2_mns8",
|
|
"tp4_mns16",
|
|
"tp4_mns8"
|
|
],
|
|
"expanded_k": 12,
|
|
"nominal_k": 11,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp4_mns32",
|
|
"tp4_mns64",
|
|
"tp2_mns16",
|
|
"tp1_mns16",
|
|
"tp1_mns32",
|
|
"tp1_mns64",
|
|
"tp1_mns8",
|
|
"tp2_mns32",
|
|
"tp2_mns64",
|
|
"tp2_mns8",
|
|
"tp4_mns16",
|
|
"tp4_mns8"
|
|
],
|
|
"expanded_k": 12,
|
|
"nominal_k": 12,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
}
|
|
],
|
|
"real_best": 3.283333333333333
|
|
},
|
|
"uncalibrated/throughput-proxy": {
|
|
"minimum_k": {
|
|
"five_percent": {
|
|
"expanded_k": 3,
|
|
"nominal_k": 3
|
|
},
|
|
"one_percent": {
|
|
"expanded_k": 3,
|
|
"nominal_k": 3
|
|
},
|
|
"zero": {
|
|
"expanded_k": 5,
|
|
"nominal_k": 5
|
|
}
|
|
},
|
|
"points": [
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64"
|
|
],
|
|
"expanded_k": 1,
|
|
"nominal_k": 1,
|
|
"real_regret": 0.30456852791878175,
|
|
"selected_cell_after_real_final": "tp1_mns64",
|
|
"selected_real_score": 2.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32"
|
|
],
|
|
"expanded_k": 2,
|
|
"nominal_k": 2,
|
|
"real_regret": 0.30456852791878175,
|
|
"selected_cell_after_real_final": "tp1_mns64",
|
|
"selected_real_score": 2.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns64"
|
|
],
|
|
"expanded_k": 3,
|
|
"nominal_k": 3,
|
|
"real_regret": 0.0076142131979695105,
|
|
"selected_cell_after_real_final": "tp2_mns64",
|
|
"selected_real_score": 3.2583333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16"
|
|
],
|
|
"expanded_k": 4,
|
|
"nominal_k": 4,
|
|
"real_regret": 0.0076142131979695105,
|
|
"selected_cell_after_real_final": "tp2_mns64",
|
|
"selected_real_score": 3.2583333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16",
|
|
"tp2_mns32"
|
|
],
|
|
"expanded_k": 5,
|
|
"nominal_k": 5,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16",
|
|
"tp2_mns32",
|
|
"tp4_mns64"
|
|
],
|
|
"expanded_k": 6,
|
|
"nominal_k": 6,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16",
|
|
"tp2_mns32",
|
|
"tp4_mns64",
|
|
"tp1_mns8"
|
|
],
|
|
"expanded_k": 7,
|
|
"nominal_k": 7,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16",
|
|
"tp2_mns32",
|
|
"tp4_mns64",
|
|
"tp1_mns8",
|
|
"tp4_mns32"
|
|
],
|
|
"expanded_k": 8,
|
|
"nominal_k": 8,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16",
|
|
"tp2_mns32",
|
|
"tp4_mns64",
|
|
"tp1_mns8",
|
|
"tp4_mns32",
|
|
"tp2_mns16"
|
|
],
|
|
"expanded_k": 9,
|
|
"nominal_k": 9,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16",
|
|
"tp2_mns32",
|
|
"tp4_mns64",
|
|
"tp1_mns8",
|
|
"tp4_mns32",
|
|
"tp2_mns16",
|
|
"tp2_mns8"
|
|
],
|
|
"expanded_k": 10,
|
|
"nominal_k": 10,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16",
|
|
"tp2_mns32",
|
|
"tp4_mns64",
|
|
"tp1_mns8",
|
|
"tp4_mns32",
|
|
"tp2_mns16",
|
|
"tp2_mns8",
|
|
"tp4_mns16"
|
|
],
|
|
"expanded_k": 11,
|
|
"nominal_k": 11,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
},
|
|
{
|
|
"candidates": [
|
|
"tp1_mns64",
|
|
"tp1_mns32",
|
|
"tp2_mns64",
|
|
"tp1_mns16",
|
|
"tp2_mns32",
|
|
"tp4_mns64",
|
|
"tp1_mns8",
|
|
"tp4_mns32",
|
|
"tp2_mns16",
|
|
"tp2_mns8",
|
|
"tp4_mns16",
|
|
"tp4_mns8"
|
|
],
|
|
"expanded_k": 12,
|
|
"nominal_k": 12,
|
|
"real_regret": 0.0,
|
|
"selected_cell_after_real_final": "tp2_mns32",
|
|
"selected_real_score": 3.283333333333333
|
|
}
|
|
],
|
|
"real_best": 3.283333333333333
|
|
}
|
|
}
|
|
}
|