You are viewing a plain text version of this content. The canonical link for it is here.
Posted to reviews@spark.apache.org by GitBox <gi...@apache.org> on 2021/02/26 04:37:33 UTC
[GitHub] [spark] zhengruifeng commented on pull request #31657: [Spark-34448][3.0][DO NOT MERGE] Binary logistic regression incorrectly computes the intercept and coefficients when data is not centered
zhengruifeng commented on pull request #31657:
URL: https://github.com/apache/spark/pull/31657#issuecomment-786405441
logs for first commit:
```
[info] LogisticRegressionSuite:
featuresMean: [0.4999142959117828,1.4847274177074965]
featuresStd: [0.28501348037270735,0.28375633081273305]
optimizer: breeze.optimize.LBFGS@e0d13c1
initialCoefWithInterceptMatrix 0.0 0.0 -3.548107045716773
gradient: [-0.002305300414829984,-0.001540091613926333,3.625153510711243E-16]
solution: 0.0,0.0,-3.548107045716773
gradient: [1.270197141453041,3.420900875935085,0.6340310033871515]
gradient: [0.02712239947522615,0.07838873390702962,0.014843589590906155]
gradient: [5.742212568839818E-5,0.004963605142872151,0.0012130388688036885]
gradient: [-0.0011314004879724678,0.0016935454656592792,6.032615470892342E-4]
solution: 0.004180284984790963,0.002792704069058841,-3.5481070457167734
gradient: [-0.0010728634715018125,0.00160772260209591,5.917007493574007E-4]
gradient: [-0.0010434636154917003,0.0015652858941190642,5.860092709095692E-4]
gradient: [-9.99199410343589E-4,0.0015022235418333052,5.775831071855921E-4]
solution: 0.010568018954399284,7.631146700935471E-4,-3.549657937366679
gradient: [1.2390525210071498E-4,1.4717833756368236E-4,3.984793999995559E-4]
solution: 0.06279624904315896,-0.014248302675022488,-3.571101773719935
gradient: [9.7770168512076E-5,-1.373828594793572E-4,3.400255628368619E-4]
gradient: [8.481446986403304E-5,-2.790060362521768E-4,3.1093063707884655E-4]
solution: 0.06716764247861332,-0.013115419667116011,-3.588130852021863
gradient: [-9.968656875832441E-5,-0.0021390619743561936,-1.5963325443471433E-4]
solution: 0.09216063924943523,0.009066818457207741,-3.767127156279349
gradient: [-1.1037293015684862E-4,-0.001941870473764838,-2.5262563678862614E-4]
solution: 0.09825523866837622,0.03454325895969772,-3.915627606930046
gradient: [-2.0482663038072557E-5,-4.43210469688911E-4,-7.738226665332456E-5]
solution: 0.08968371353490419,0.055787393265421564,-4.005407480573733
gradient: [-7.5354046131570336E-6,-5.834078775921241E-5,-1.3492011139001316E-5]
solution: 0.08582579292847434,0.05750880834200776,-4.005067048017333
gradient: [-9.990118197823818E-7,-1.9016389487075713E-6,-8.202512294008946E-7]
solution: 0.08520584612951042,0.057117003687867866,-4.001392447236977
gradient: [-2.1269163009208114E-8,1.6634162741361534E-7,1.4415507203052513E-8]
solution: 0.08518495364769638,0.05703053481014659,-4.000866357569121
gradient: [-1.0143803239515137E-10,6.222596734389185E-9,1.0190080001226675E-9]
solution: 0.08518662793392717,0.05702725297223899,-4.000852573726639
Just the blr data
Coefficients: [0.29888631170192387,0.20097261903867272]
Intercept: -4.000852573726639
objectives: 0.12762747240529596,0.1276204770445784,0.1276095137555505,0.12756395823655467,0.12755829996951326,0.12751789363970717,0.12749589534392758,0.1274859125261507,0.12748551910072972,0.12748550724997348,0.12748550712365025,0.127485507123459
featuresMean: [0.4999142959117828,1.4847274177074965,0.9899999976158129]
featuresStd: [0.28501348037270735,0.28375633081273305,0.03000002215257344]
optimizer: breeze.optimize.LBFGS@225c477f
initialCoefWithInterceptMatrix 0.0 0.0 0.0 -3.548107045716773
gradient: [-0.002305300414829984,-0.001540091613926333,-4.4666644349528265E-5,3.625153510711243E-16]
solution: 0.0,0.0,0.0,-3.548107045716773
gradient: [1.4043881959417224,3.884356710917848,23.954743028917907,0.7257390266235928]
gradient: [0.031201599585874356,0.09016374509441172,0.5634823144261903,0.017072723505223292]
gradient: [3.232595426584435E-4,0.005753607491472995,0.044972059597789826,0.0013638840438085728]
gradient: [-0.0012590176292179422,0.001365692563291016,0.01789481499206562,5.435186208776085E-4]
solution: 0.003361311019887606,0.0022455758391510356,6.512751349352555E-5,-3.5481070457167734
gradient: [-0.0012058526178998585,0.0013107514665489008,0.017161368441663608,5.216994435121182E-4]
gradient: [-0.0011790396852133288,0.001284089723365383,0.01679996576337007,5.109510894332127E-4]
gradient: [-0.0011385318209425725,0.0012451072263457426,0.016264501895099932,4.950297871192996E-4]
gradient: [-0.0010771199438327206,0.0011889041731884607,0.015476221524179956,4.715998596198084E-4]
solution: 0.014585851627457428,0.009492561757612174,-0.0017615363470460436,-3.5481690479854455
gradient: [1.538887596861738E-4,5.298765454822907E-4,0.003464779223103357,1.160666951868568E-4]
solution: 0.08327888456493171,0.05384078236770347,-0.012962432443477992,-3.548558824766501
gradient: [-3.2589948482520296E-6,1.0213435092852024E-6,1.9811253571497672E-5,1.1886786287249719E-5]
solution: 0.08453485443429141,0.05463738823981839,-0.013274937663181247,-3.548583974676902
gradient: [-2.385174172773077E-7,5.526130808482889E-7,-3.4939774539634526E-7,1.1294707984496084E-5]
solution: 0.0846875705174748,0.054735933110898506,-0.013299818781811643,-3.548585606117963
gradient: [-3.3810264008587864E-7,2.670868522249066E-7,-1.75510037687809E-6,1.125118225882494E-5]
gradient: [-3.878950828871268E-7,1.243242276239016E-7,-2.4579490735686702E-6,1.1229419476802874E-5]
gradient: [-4.6258353642127757E-7,-8.981909696433376E-8,-3.5122188552941224E-6,1.1196775406189872E-5]
gradient: [-5.746157420832957E-7,-4.1103270748129716E-7,-5.09361616893375E-6,1.1147809527964553E-5]
gradient: [-7.426629843554906E-7,-8.928500247250781E-7,-7.465695576017879E-6,1.1074361225055086E-5]
gradient: [-9.947314471921232E-7,-1.6155690306649716E-6,-1.1023777437736044E-5,1.0964189927105351E-5]
gradient: [-1.3728287414776048E-6,-2.699631852435118E-6,-1.636081641822193E-5,1.0798935582432377E-5]
gradient: [-1.939962530948605E-6,-4.325690796620734E-6,-2.4366186309353907E-5,1.0551059920859132E-5]
gradient: [-2.7906358765086735E-6,-6.764699811755861E-6,-3.637381684693582E-5,1.0179259601508278E-5]
gradient: [-4.0665843845300455E-6,-1.0423034688555446E-5,-5.438430802060975E-5,9.621588762150324E-6]
solution: 0.0846875705174748,0.054735933110898506,-0.013299818781811643,-3.548585606117963
gradient: [-2.385174172773077E-7,5.526130808482889E-7,-3.4939774539634526E-7,1.1294707984496084E-5]
gradient: [-3.440101174710719E-7,2.140211317141052E-7,-2.358906082786234E-6,1.123348788290265E-5]
gradient: [-3.967563806994008E-7,4.4725463667759864E-8,-3.3636584917312575E-6,1.1202877885420761E-5]
gradient: [-4.7587566701034233E-7,-2.09217654962913E-7,-4.870784943797389E-6,1.115696295650889E-5]
gradient: [-5.94554352885666E-7,-5.901314705197364E-7,-7.131469709530847E-6,1.1088090713180832E-5]
gradient: [-7.725718324804517E-7,-1.161500254371333E-6,-1.0522485813565708E-5,1.0984782686918663E-5]
gradient: [-1.0395968172452967E-6,-2.0185490637795736E-6,-1.5608985118804412E-5,1.0829821408901281E-5]
gradient: [-1.4401315155747625E-6,-3.304112455660856E-6,-2.323867816976326E-5,1.059738120594298E-5]
gradient: [-2.040927311571323E-6,-5.2324354461108556E-6,-3.468309193828281E-5,1.024872475681238E-5]
gradient: [-2.9421069396890685E-6,-8.124870208995275E-6,-5.1849429549435624E-5,9.725748758568528E-6]
gradient: [-4.293844733636149E-6,-1.2463410482831362E-5,-7.75982991109886E-5,8.94130428013014E-6]
solution: 0.0846875705174748,0.054735933110898506,-0.013299818781811643,-3.548585606117963
blr data plus one vector that is filled with 1's and .9's
Coefficients: [0.29713531586902586,0.1928976631256973,-0.44332696536594945]
Intercept: -3.548585606117963
objectives: 0.12762747240529596,0.12762184862587836,0.12758748705248976,0.12748833468678486,0.12748809482977605,0.12748809437935926,0.12748809437935926,0.12748809437935926
```
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org