diff options
| author | haoyuren <13851610112@163.com> | 2026-07-28 20:54:56 +0800 |
|---|---|---|
| committer | haoyuren <13851610112@163.com> | 2026-07-28 20:54:56 +0800 |
| commit | 18bf25f34f6d352605086aa8fe63977baacc315b (patch) | |
| tree | fb140a2bf0079ca3afc41d937b34a09d6dd89d53 /outputs/learning_rate_compensation/regime_summary.csv | |
| parent | 99b95217a30f75e9e35b5044850ced4ab802bc87 (diff) | |
Track learning-rate and parameter-matched row-level outputs
Pulled from the original experiment machine. These are the authoritative
source rows behind the Appendix H tables (regime_summary matches the
paper values; the depth-3 compensated gap rounds to 0.16262).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Diffstat (limited to 'outputs/learning_rate_compensation/regime_summary.csv')
| -rw-r--r-- | outputs/learning_rate_compensation/regime_summary.csv | 16 |
1 files changed, 16 insertions, 0 deletions
diff --git a/outputs/learning_rate_compensation/regime_summary.csv b/outputs/learning_rate_compensation/regime_summary.csv new file mode 100644 index 0000000..12d8446 --- /dev/null +++ b/outputs/learning_rate_compensation/regime_summary.csv @@ -0,0 +1,16 @@ +depth,regime,rows,stable_rows,bp_lr,fa_lr_mean,rho_mean,bp_loss_mean,fa_loss_mean,gap_mean,gap_std,gap_sem
+1,same_lr,48,48,0.001,0.001,0.9610069350407984,2.1666989888152424,2.2257618018338,0.059062813018558,0.009435428579739853,0.00385199758747267
+1,initial_compensation,48,48,0.001,0.0010407171219455268,0.9610069350407984,2.1666989888152424,2.207225477295465,0.04052648848022297,0.007618800905898918,0.0031103624452177643
+1,independently_tuned,48,48,0.128,0.064,0.9610069350407984,0.294261348016096,0.8636599107003535,0.5693985626842576,0.025470956849676082,0.010398474590359084
+2,same_lr,48,48,0.001,0.001,0.7612398883286953,2.0219677440519996,2.2211329917145335,0.199165247662534,0.03444824281209884,0.014063436237523408
+2,initial_compensation,48,48,0.001,0.0013168083270719375,0.7612398883286953,2.0219677440519996,2.119259691680273,0.09729194762827353,0.02543136611942341,0.010382311742415214
+2,independently_tuned,48,48,0.128,0.032,0.7612398883286953,0.15235488385676207,0.7944569102207383,0.6421020263639763,0.12610115797280932,0.051480582167912944
+3,same_lr,48,48,0.001,0.001,0.6295068544264688,1.847377871664352,2.1208343429545216,0.2734564712901692,0.06428969293109411,0.026246157233565874
+3,initial_compensation,48,48,0.001,0.0016066791854717785,0.6295068544264688,1.847377871664352,2.0100027235277964,0.1626248518634443,0.05875562787394225,0.023986884634667825
+3,independently_tuned,48,48,0.064,0.016,0.6295068544264688,0.0855589967594977,1.215578000751055,1.1300190039915574,0.057596675687040334,0.023513744385635763
+4,same_lr,48,48,0.001,0.001,0.5314180794284439,1.7554493999414322,2.048847847137616,0.2933984471961841,0.05201060249040933,0.021233239552705147
+4,initial_compensation,48,48,0.001,0.0019168307305947902,0.5314180794284439,1.7554493999414322,1.9398655534801694,0.18441615353873775,0.04066007177452844,0.016599404792089205
+4,independently_tuned,48,48,0.064,0.016,0.5314180794284439,0.07042385696633152,1.3785391365213364,1.3081152795550048,0.14724117672303647,0.060110958683067185
+6,same_lr,48,48,0.001,0.001,0.39886513049466427,1.6381389737072565,2.044780538842934,0.40664156513567756,0.09378365477021289,0.03828701673339253
+6,initial_compensation,48,48,0.001,0.002590508159488245,0.39886513049466427,1.6381389737072565,1.9266386829320636,0.28849970922480717,0.06681916348673785,0.027278809263686105
+6,independently_tuned,48,48,0.064,0.004,0.39886513049466427,0.08353873014787917,1.840925633337892,1.757386903190013,0.06337128253452119,0.02587121775922074
|