ar-regularisation-i6-86094df-3levels-9b9b1699
completed —
Measured behaviour
modelposition-marginal floor
Table view
| optimizer step | model | position-marginal floor |
|---|---|---|
| 300 | 3.6612 | 3.9290 |
| 600 | 3.6480 | 3.9290 |
| 900 | 3.8384 | 3.9290 |
| 1200 | 3.9849 | 3.9290 |
| 1500 | 4.2191 | 3.9290 |
| 1800 | 4.6009 | 3.9290 |
| 2100 | 4.7426 | 3.9290 |
| 2400 | 5.0071 | 3.9290 |
| 2700 | 5.2550 | 3.9290 |
| 3000 | 5.4942 | 3.9290 |
| 300 | 3.6732 | 3.9290 |
| 600 | 3.5977 | 3.9290 |
| 900 | 3.6403 | 3.9290 |
| 1200 | 3.7151 | 3.9290 |
| 1500 | 3.8083 | 3.9290 |
| 1800 | 3.8673 | 3.9290 |
| 2100 | 3.8980 | 3.9290 |
| 2400 | 3.9446 | 3.9290 |
| 2700 | 4.0504 | 3.9290 |
| 3000 | 4.0480 | 3.9290 |
| 300 | 3.7090 | 3.9290 |
| 600 | 3.6164 | 3.9290 |
| 900 | 3.5921 | 3.9290 |
| 1200 | 3.5994 | 3.9290 |
| 1500 | 3.6253 | 3.9290 |
| 1800 | 3.6801 | 3.9290 |
| 2100 | 3.6735 | 3.9290 |
| 2400 | 3.6924 | 3.9290 |
| 2700 | 3.7389 | 3.9290 |
| 3000 | 3.7491 | 3.9290 |
| 3300 | 3.7736 | 3.9290 |
Measured result
Verbatim from the run's summary.json.
arms
dropout
0.0
epochs_to_best
3.5807534502051475
fraction
1.0
held_out_nll
3.64796488139269
icons
2681
label
none
marginal_nll
3.9289882210903344
parameters
524674
ratio_to_own_floor
0.9284743746012917
selected_step
600
steps_run
3000
trace
fraction
1.0
held_out_nll
3.6611640545050923
label
none
marginal_nll
3.9289882210903344
step
300
train_loss
3.2960057258605957
fraction
1.0
held_out_nll
3.64796488139269
label
none
marginal_nll
3.9289882210903344
step
600
train_loss
3.3200597763061523
fraction
1.0
held_out_nll
3.8384189372997466
label
none
marginal_nll
3.9289882210903344
step
900
train_loss
2.903935194015503
fraction
1.0
held_out_nll
3.984894900990683
label
none
marginal_nll
3.9289882210903344
step
1200
train_loss
1.603338360786438
fraction
1.0
held_out_nll
4.219101442814329
label
none
marginal_nll
3.9289882210903344
step
1500
train_loss
2.4615190029144287
fraction
1.0
held_out_nll
4.600927306373807
label
none
marginal_nll
3.9289882210903344
step
1800
train_loss
1.678338646888733
… and 4 more
train_seconds
305.60176272000535
weight_decay
0.01
dropout
0.1
epochs_to_best
3.5807534502051475
fraction
1.0
held_out_nll
3.5977242662007285
icons
2681
label
dropout-0.1
marginal_nll
3.9289882210903344
parameters
524674
ratio_to_own_floor
0.9156872109945708
selected_step
600
steps_run
3000
trace
fraction
1.0
held_out_nll
3.6731803649732435
label
dropout-0.1
marginal_nll
3.9289882210903344
step
300
train_loss
3.421252727508545
fraction
1.0
held_out_nll
3.5977242662007285
label
dropout-0.1
marginal_nll
3.9289882210903344
step
600
train_loss
3.4300286769866943
fraction
1.0
held_out_nll
3.640337257682916
label
dropout-0.1
marginal_nll
3.9289882210903344
step
900
train_loss
3.2496330738067627
fraction
1.0
held_out_nll
3.7151111867983517
label
dropout-0.1
marginal_nll
3.9289882210903344
step
1200
train_loss
2.272975444793701
fraction
1.0
held_out_nll
3.808261396217343
label
dropout-0.1
marginal_nll
3.9289882210903344
step
1500
train_loss
3.0613815784454346
fraction
1.0
held_out_nll
3.867266619247664
label
dropout-0.1
marginal_nll
3.9289882210903344
step
1800
train_loss
2.3356008529663086
… and 4 more
train_seconds
306.15317193399824
weight_decay
0.01
dropout
0.3
epochs_to_best
5.371130175307721
fraction
1.0
held_out_nll
3.592118464805465
icons
2681
label
dropout-0.3
marginal_nll
3.9289882210903344
parameters
524674
ratio_to_own_floor
0.9142604311011692
selected_step
900
steps_run
3300
trace
fraction
1.0
held_out_nll
3.709010581521979
label
dropout-0.3
marginal_nll
3.9289882210903344
step
300
train_loss
3.567981481552124
fraction
1.0
held_out_nll
3.616420899002616
label
dropout-0.3
marginal_nll
3.9289882210903344
step
600
train_loss
3.518282413482666
fraction
1.0
held_out_nll
3.592118464805465
label
dropout-0.3
marginal_nll
3.9289882210903344
step
900
train_loss
3.475635290145874
fraction
1.0
held_out_nll
3.599358769742663
label
dropout-0.3
marginal_nll
3.9289882210903344
step
1200
train_loss
2.826960802078247
fraction
1.0
held_out_nll
3.6253005883014704
label
dropout-0.3
marginal_nll
3.9289882210903344
step
1500
train_loss
3.457867383956909
fraction
1.0
held_out_nll
3.680139994520633
label
dropout-0.3
marginal_nll
3.9289882210903344
step
1800
train_loss
2.757734775543213
… and 5 more
train_seconds
336.6650172630034
weight_decay
0.01
axis
regularisation
best_ratio_to_own_floor
0.9142604311011692
checks
best_arm_clears_magnitude_bar
false
config_sha256
2680463269a80492…criteria
expect_monotone
false
max_best_ratio
0.85
min_second_doubling_share
0.25
deterministic_algorithms
true
device
cuda
gpu
NVIDIA GeForce RTX 4080
improvement_first_doubling
0.050240615191961435
improvement_second_doubling
0.0056058013952635655
predeclared_outcome
not_limited_by_regularisation
prepare_seconds
102.37748816399835
schema_version
1
second_doubling_share
0.11157907549190403
study_version
ar-regularisation-i6
torch_version
2.14.0a0+4fdf77b940.nv26.08
validation_icons
339
State transitions
- planned2026-09-21T13:55:00Z
- completed2026-09-21T14:30:00Zdropout is the largest single effect measured in this gate and is still an order of magnitude too small; 0.9143 against a standing bar of 0.5
Run record
This run has no run.yaml. What follows is the identity and configuration carried by its rows in state/runs.jsonl, the append-only registry.
run_id
ar-regularisation-i6-86094df-3levels-9b9b1699
config
configs/learning/ar-regularisation-i6.yaml