Skip to content

Run B is decided against: the power analysis, the 8-epoch cosine rung, and the gate applied (#135) #138

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
jonfroehlich merged 13 commits into from
Sep 4, 2026
Merged

Run B is decided against: the power analysis, the 8-epoch cosine rung, and the gate applied (#135) #138

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
13 commits
Select commit Hold shift + click to select a range
a6fb658
The benchmark can resolve Run B, but only read paired -- and pooling …
jonfroehlich Aug 18, 2026
bedbeb9
Recover per-pano detections from the Run A heatmap cache, with no GPU…
jonfroehlich Aug 18, 2026
bd9486a
The bracket is closed: paired MDE on manual_gold is 0.0063, measured …
jonfroehlich Aug 18, 2026
2cdb01c
Pre-register the 8-epoch cosine rung, and give train.py a preemption-…
jonfroehlich Aug 18, 2026
cb2fb59
Cosine rung is submitted: job 38640313, with the pre-flight verified …
jonfroehlich Aug 18, 2026
2617d0d
Check the applied LR across requeue boundaries, don't assume it (#135)
jonfroehlich Aug 19, 2026
d9c4fcd
Run A was requeued five times, not twice -- plain sacct hides incarna…
jonfroehlich Aug 19, 2026
7bbdace
Merge remote-tracking branch 'origin/main' into analysis/run-b-power-135
jonfroehlich Aug 20, 2026
fa02b37
Make resume cheap and checkpoints frequent, so preemption stops costi…
jonfroehlich Aug 20, 2026
af2da14
The cosine rung completed, and the LR schedule verifies over 100% of …
jonfroehlich Aug 29, 2026
82b9056
manual_gold: the cosine rung is TIED with Run A at every epoch (#135)
jonfroehlich Aug 29, 2026
4171a6e
Apply the Run B gate that was pre-registered and never evaluated: Run…
jonfroehlich Sep 3, 2026
cc94b34
Review fixes: the rung is tied on the primary, not at every epoch; th…
jonfroehlich Sep 4, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5,309 changes: 5,309 additions & 0 deletions docs/data/benchmark_power_135.json
View file
Open in desktop

Large diffs are not rendered by default.

202 changes: 202 additions & 0 deletions docs/data/cosine_rung_135_manual_gold/epoch_1.csv
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,202 @@
confidence_threshold,precision,recall
0.000,0.027689,0.933657
0.005,0.327572,0.933401
0.010,0.496604,0.932891
0.015,0.576911,0.932126
0.020,0.624615,0.931105
0.025,0.657344,0.929574
0.030,0.685381,0.928298
0.035,0.707782,0.928298
0.040,0.723273,0.927022
0.045,0.738308,0.926512
0.050,0.750465,0.926257
0.055,0.761545,0.925746
0.060,0.770391,0.925491
0.065,0.778064,0.924981
0.070,0.783304,0.924215
0.075,0.788935,0.924215
0.080,0.796174,0.923960
0.085,0.802795,0.923450
0.090,0.808906,0.922429
0.095,0.812416,0.921664
0.100,0.818141,0.920643
0.105,0.821307,0.920643
0.110,0.826346,0.920388
0.115,0.829078,0.919622
0.120,0.833064,0.919367
0.125,0.836782,0.918347
0.130,0.841920,0.917326
0.135,0.845176,0.916560
0.140,0.848771,0.916560
0.145,0.852697,0.915795
0.150,0.855712,0.915540
0.155,0.859128,0.915029
0.160,0.861679,0.914009
0.165,0.865087,0.912988
0.170,0.868383,0.912478
0.175,0.869618,0.912223
0.180,0.872164,0.912223
0.185,0.874847,0.911457
0.190,0.876598,0.909926
0.195,0.879162,0.909671
0.200,0.880317,0.908395
0.205,0.881974,0.907629
0.210,0.883917,0.907374
0.215,0.885842,0.906864
0.220,0.887722,0.905843
0.225,0.889920,0.905588
0.230,0.891206,0.905078
0.235,0.893569,0.904057
0.240,0.895070,0.903292
0.245,0.897768,0.903036
0.250,0.899822,0.903036
0.255,0.900892,0.902271
0.260,0.904018,0.901250
0.265,0.905006,0.899464
0.270,0.905680,0.899209
0.275,0.906516,0.898188
0.280,0.908105,0.897678
0.285,0.909232,0.897168
0.290,0.911307,0.896657
0.295,0.912919,0.896147
0.300,0.914822,0.896147
0.305,0.915927,0.895126
0.310,0.916819,0.894361
0.315,0.916972,0.893340
0.320,0.918550,0.892064
0.325,0.918990,0.891554
0.330,0.919895,0.890788
0.335,0.921046,0.890023
0.340,0.922466,0.889513
0.345,0.923668,0.889257
0.350,0.924118,0.888747
0.355,0.924323,0.888237
0.360,0.924980,0.887216
0.365,0.925926,0.886706
0.370,0.926836,0.885685
0.375,0.928495,0.884664
0.380,0.928935,0.883899
0.385,0.931090,0.882623
0.390,0.932775,0.881602
0.395,0.933225,0.880837
0.400,0.933694,0.880327
0.405,0.934093,0.878796
0.410,0.935458,0.876499
0.415,0.935897,0.875478
0.420,0.936868,0.874713
0.425,0.938065,0.873437
0.430,0.938529,0.872672
0.435,0.939511,0.871906
0.440,0.940512,0.871396
0.445,0.942557,0.870885
0.450,0.942818,0.870885
0.455,0.943047,0.870375
0.460,0.943721,0.868589
0.465,0.944954,0.867313
0.470,0.945434,0.866548
0.475,0.945667,0.866037
0.480,0.946399,0.865017
0.485,0.946588,0.863741
0.490,0.947383,0.863741
0.495,0.948092,0.862210
0.500,0.949129,0.861699
0.505,0.949592,0.860424
0.510,0.949817,0.859658
0.515,0.951385,0.858893
0.520,0.952435,0.858382
0.525,0.953746,0.857617
0.530,0.954494,0.856341
0.535,0.954986,0.855320
0.540,0.955518,0.855065
0.545,0.956323,0.854810
0.550,0.957343,0.853279
0.555,0.957233,0.850982
0.560,0.957435,0.849451
0.565,0.958225,0.848686
0.570,0.958983,0.847155
0.575,0.959757,0.845879
0.580,0.959977,0.844603
0.585,0.960744,0.843072
0.590,0.960722,0.842562
0.595,0.962077,0.841541
0.600,0.962573,0.840010
0.605,0.963093,0.838990
0.610,0.963615,0.837969
0.615,0.964097,0.835928
0.620,0.965192,0.834907
0.625,0.965976,0.833121
0.630,0.966479,0.831335
0.635,0.967579,0.830059
0.640,0.967819,0.828783
0.645,0.968638,0.827507
0.650,0.969488,0.826997
0.655,0.969733,0.825721
0.660,0.970288,0.824955
0.665,0.970801,0.822914
0.670,0.971282,0.819852
0.675,0.972702,0.818321
0.680,0.973508,0.815769
0.685,0.973732,0.813473
0.690,0.974241,0.810666
0.695,0.974154,0.807859
0.700,0.974637,0.804032
0.705,0.975784,0.801990
0.710,0.976302,0.798928
0.715,0.977145,0.796377
0.720,0.978336,0.795101
0.725,0.979186,0.792294
0.730,0.979121,0.789742
0.735,0.978974,0.784129
0.740,0.979207,0.781067
0.745,0.979743,0.777494
0.750,0.979651,0.773922
0.755,0.980832,0.770350
0.760,0.981077,0.767288
0.765,0.982301,0.764736
0.770,0.982225,0.761419
0.775,0.982741,0.755550
0.780,0.982960,0.750702
0.785,0.982897,0.747895
0.790,0.983707,0.739474
0.795,0.984594,0.733861
0.800,0.984472,0.727992
0.805,0.984337,0.721613
0.810,0.984561,0.715999
0.815,0.985138,0.710385
0.820,0.985310,0.701710
0.825,0.986212,0.693544
0.830,0.987198,0.688696
0.835,0.987449,0.682572
0.840,0.987256,0.672110
0.845,0.987429,0.661393
0.850,0.987611,0.650931
0.855,0.987824,0.641745
0.860,0.987966,0.628477
0.865,0.988163,0.617760
0.870,0.989189,0.607043
0.875,0.989398,0.595305
0.880,0.990048,0.583822
0.885,0.990218,0.568257
0.890,0.990340,0.549375
0.895,0.990047,0.533044
0.900,0.990594,0.510589
0.905,0.990688,0.488645
0.910,0.990695,0.461853
0.915,0.990207,0.438632
0.920,0.990338,0.418474
0.925,0.990385,0.394233
0.930,0.991096,0.369227
0.935,0.990262,0.337331
0.940,0.989414,0.310028
0.945,0.988351,0.281449
0.950,0.989141,0.255677
0.955,0.988901,0.227354
0.960,0.990062,0.203368
0.965,0.989914,0.175300
0.970,0.989983,0.151314
0.975,0.990079,0.127328
0.980,0.988038,0.105384
0.985,0.985876,0.089053
0.990,0.986301,0.073488
0.995,0.983402,0.060475
1.000,0.979798,0.049502
Loading
Loading