Skip to content

Latest commit

 

History

History
60 lines (52 loc) · 19 KB

README_detailed.md

File metadata and controls

60 lines (52 loc) · 19 KB

Deep learning for Cod-Otoliths

Species Predict validLOSS MSE MAPE ACC MCC #trained activ. f classWeights
Greenland Halibut(1) age x 2.65 0.124 0.262 x 8875 linear x
Salmon sea age -"- 0.239 0.141 0.822 x 9073 linear x
Salmon B4 river age 0.359 0.359 19.58 0.618 x 6246 linear x
Cod B4 age 0.0297 0.0297 1.588 0.984 x 6330 linear x
Cod B4 age 0.8796 0.8796 9.228 0.52597 x 1029 linear x
Cod B4 (epoch41) age 0.9695 0.9695 - 0.5805 x 1984 linear x
Cod B4 (epoch53) age 0.9785 0.9785 - 0.6174 x 1984 linear x
Cod B4 (test-metric) age 0.7814 0.7814 - 0.6409 x 1984 linear x

5-fold training after adding 3000 images - testset 15%

NN-config fold-1 (mse, acc) fold-2 fold-3 fold-4 fold-5 mean MSE mean ACC datset size
B4 [0.48629727959632874, 0.6488250494003296] [0.4687076508998871, 0.6697127819061279] [0.4820464551448822, 0.6631853580474854] [0.4878818988800049, 0.6488250494003296] [0.47346818447113037, 0.6579634547233582] 0.4216341924334377 0.6971279373368147 5150
B4,standardScalar on target, StratifiedKFold [0.4904178682121655, 0.643603133159269] [0.5347786293189214, 0.6227154046997389] [0.4968175576778982, 0.660574412532637] [0.4573790357033823, 0.6984334203655352] [0.5134489160772718, 0.6514360313315927] 0.4257378207634962 0.7010443864229765 5150
B4,standardScalar on target, StratifiedKFold, pretraining on salmon-scales 20 epochs [0.46910862052397906, 0.6631853785900783] [0.4689538583672292, 0.685378590078329] [0.5130625894583206, 0.6514360313315927] [0.4741178483435903, 0.6814621409921671] [0.47853572666031796, 0.6501305483028721] 0.43275226756928326 0.6892950391644909 5150
B5,standardScalar on target, StratifiedKFold [0.4353308031328409, 0.667098445595855 ] [0.44663970124877767, 0.6826424870466321] [0.45198405952545945, 0.677461139896373] [0.430863676385045, 0.6748704663212435] [0.44126310267337826, 0.6917098445595855] 0.40109202928591997 0.7072538860103627 5150

10-fold training - testset 10%

NN-config fold-1 (mse, acc)(max, mean, min) fold-2 fold-3 fold-4 fold-5 fold-6 fold-7 fold-8 fold-9 fold-10 mean MSE mean ACC datset size
B4 with B5 img size,standardScalar on target, StratifiedKFold [0.3197536160088654,0.6990291262135923] [0.31845087610760114,0.6893203883495146] [0.30601433207216255,0.6873786407766991] [0.31335182256242056,0.683495145631068] [0.32232335618776714,0.6893203883495146] [0.31398366708967657,0.7009708737864078] [0.3147587854708994,0.6970873786407767] [0.3164655743413814,0.6679611650485436] [0.30647813064746254,0.6893203883495146] [0.30178253861722815,0.7242718446601941] 0.27677442836796534 0.7281553398058253 5150
B4 MLP(256,32,1) middle 0.34420374858404784,
0.6854368932038835
0.3276395089426725,
0.6932038834951456
0.31646135453238877,
0.7300970873786408
0.33412137047640145,
0.6854368932038835
0.32597101450158217,
0.6776699029126214
0.320012361979369,
0.6815533980582524
0.3552662592840664,
0.6718446601941748
0.32610113857471096,
0.6718446601941748
0.3127441343499593,
0.683495145631068
0.32505598184156775,
0.6951456310679611
0.2847701138075624 0.7145631067961165 5150
B4 MLP(256,32,1) max 0.34048308447557973,
0.6407766990291263
0.31743578138623596,
0.6815533980582524
0.31839301469258846,
0.6718446601941748
0.347277618330676,
0.6621359223300971
0.3355500890052016,
0.6776699029126214
0.3355500890052016,
0.6951456310679611
0.3357910349341229,
0.6718446601941748
0.3201121420411831,
0.6932038834951456
0.3536916718387224,
0.6621359223300971
0.3357048581437831,
0.6524271844660194
0.2908639084032022 0.7087378640776699 5150
B5,standardScalar on target, StratifiedKFold, min [0.32392935016259955,0.7184466019417476] [0.3216571342235008,0.6912621359223301] [0.32478147278277225,0.6932038834951456] [0.33645739740161507,0.6679611650485436] [0.29122819849402387,0.7359223300970874] [0.3136678521029937,0.7067961165048544] [0.32013204172139337,0.6621359223300971] [0.3309105222508796,0.683495145631068] [0.32977773700250607,0.6951456310679611] [0.31682926098100384,0.6873786407766991] 0.2770159431240281 0.7436893203883496 5150
B5,standardScalar on target, StratifiedKFold, middle 0.30821068974075166,
0.7029126213592233
0.2861046477441015,
0.7203883495145631
0.3152282506349248,
0.6776699029126214
0.3490899074879489,
0.6660194174757281
0.33241087097324296,
0.6737864077669903
0.31005261853611604,
0.6990291262135923
0.27975445236855084,
0.7184466019417476
0.2746708636880597,
0.7145631067961165
0.3305822893783216,
0.6815533980582524
0.28843126783645945,
0.7223300970873786
0.2731098431413754 0.7339805825242719 5150
B6,standardScalar on target, StratifiedKFold, min [0.32510835507442315,0.683495145631068] [0.3290709168395908,0.6854368932038835] [0.33377248527623243,0.6640776699029126] [0.29291382688065226,0.7242718446601941] [0.31181813555346594, 0.7067961165048544] [0.2902428397533708,0.7087378640776699] [0.3196961505168636,0.6932038834951456] [0.3060735895344372,0.6932038834951456] [0.27629437608294805,0.7203883495145631] [0.29986697830481673,0.6893203883495146] 0.272170267061415 0.7339805825242719 5150
B6,standardScalar on target, StratifiedKFold, middle [0.32259526220834905,0.6854368932038835] [0.30107642192557743,0.6990291262135923] [0.31192925963007273,0.6757281553398058] [0.2681776623502772,0.7359223300970874] [0.2941738409037171,0.7281553398058253] [0.26564604307906664,0.7203883495145631] [0.3091506689318128,0.6796116504854369] [0.3109753967546189,0.6932038834951456] [0.2784018680409038,0.7203883495145631] [0.2892850218536424,0.7106796116504854] 0.2622701387186297 0.7436893203883496 5150
B6,standardScalar on target, StratifiedKFold, max 0.43451196999200425,
0.7048543689320388
0.3060707878247112,
0.6815533980582524
0.3060707878247112,
0.6524271844660194
0.2702834854954916,
0.7320388349514563
0.38975394919619233,
0.6912621359223301
0.3209367247336584,
0.6776699029126214
0.41142236959285977,
0.6796116504854369
0.3205649040133468,
0.6796116504854369
0.2941406328857888,
0.7281553398058253
0.4479741380215303,
0.6854368932038835
0.30471818752107327 0.7145631067961165 5150
EfficientNetV2-m baseline [0.43593598646472737, 0.5864077669902913], [11.656976699829102, 5.509386144332515, 1.0240179300308228] [0.329003091574722, 0.6757281553398058],[11.689010620117188, 5.1254937412669355, 0.9654581546783447] [0.33623309593114964, 0.6776699029126214], [11.889851570129395, 5.285319744498985, 0.9146838188171387] [0.3737735574605572, 0.6368932038834951], [12.078588485717773, 5.409902389535626, 0.9477762579917908] [0.39238114248926304,0.625242718446602],[11.572626113891602,5.190874819500932,0.9491563439369202] [0.36080863358942594,0.654368932038835],[11.696592330932617,5.305954855854072,0.8887564539909363] [0.34378273482522753, 0.6601941747572816],[11.557056427001953, 5.2136024237836445, 0.920559823513031] [0.37456772209432077, 0.6388349514563106],[11.856122016906738,5.3887680463420535,0.9428082704544067] [0.32232482847267624,0.658252427184466],[11.902791023254395,5.312223237000623,0.913942813873291] [0.32846365935074273,0.6660194174757281],[12.20690631866455, 5.246973541292172, 0.9430772662162781] 0.33051156280698374 0.6699029126213593 5150
EfficientNetV2-l baseline [0.3631332187061773, 0.654368932038835], [12.01773738861084, 5.310474924439365,0.8895091414451599] [0.35952411192975386, 0.6524271844660194], [11.634198188781738, 5.277551628663702, 0.8856195211410522] [0.434872708049315,0.6407766990291263],[11.776787757873535,5.303626792176256,0.9191138744354248] [0.343894515699138,0.6699029126213593],[11.503342628479004,5.10492112439813,0.8544027209281921] [0.38062042388052225,0.6310679611650486],[11.66337776184082, 5.360615915928072, 0.8762474060058594] [0.35183026052933125,0.6640776699029126],[11.736310958862305,5.2492036546318275,0.8503969311714172] [0.3770195560160116,0.6485436893203883],[11.890049934387207,5.2943437687401635,0.9406237006187439] [0.3549831283558617,0.658252427184466],[11.888720512390137,5.279939203586393,0.9145044088363647] [0.33879768520217496,0.6563106796116505],[11.994569778442383,5.31596553163621,0.9251725673675537] [0.3497640113286335,0.658252427184466],[11.943987846374512,5.207454828845645,0.8693790435791016] 0.3482863627148823 0.6757281553398058 5150
EfficientNetV2-m exposure="middle" [0.39718897058209524, 0.6077669902912621],[11.72133731842041,5.447287622701775,0.9450588226318359] [0.37383119606970494,0.6524271844660194],[11.627776145935059,5.1725509268566245,1.0156219005584717] [0.35610648247128274,0.6601941747572816],[11.998383522033691,5.361631473985691,0.8636019229888916] [0.3840084548166239,0.6271844660194175],[11.898012161254883, 5.398851113064775, 0.9133579134941101] [0.349523375805845,0.654368932038835],[11.257091522216797,5.246484787255815,0.8743095993995667] [0.33748664972898557,0.6679611650485436],[11.714927673339844,5.247008727707909,0.9867938160896301] [0.3256831318805641,0.658252427184466],[11.796988487243652,5.233834440384096,0.9097902178764343] [0.36527177271722916,0.6213592233009708],[11.842670440673828,5.338287988796975,0.9638106226921082] [0.35276076129821055,0.6640776699029126],[11.92320728302002,5.212955970555833,0.9720532298088074] [0.3345734053043579,0.6640776699029126],[12.203539848327637,5.287658938620854,0.8788002729415894] 0.3357187723002123 0.6427184466019418 5150
EfficientNetV2-m exposure="max" [0.4552132929754933,0.5883495145631068],[11.542473793029785,5.508298445905297,1.0317736864089966] [0.36893291988509574,0.6524271844660194],[11.887615203857422,5.27592860066775,0.8879614472389221] [0.41200427720744315,0.6097087378640776],[12.223085403442383,5.459770287124856,0.8846766352653503] [0.3505811724148065,0.6446601941747573],[11.851435661315918,5.365713037101968,0.9331003427505493] [0.34315351951040934,0.6796116504854369],[11.596426963806152,5.156253770369928,0.919688880443573] [0.4127972184036007,0.6038834951456311],[11.885137557983398,5.45255562101753,0.9943726062774658] [0.3575387501974067,0.658252427184466],[11.667533874511719,5.304152626319996,0.9186959862709045] [0.36465211985087903,0.6485436893203883],[11.978194236755371,5.313120795337899,0.9499825239181519] [0.44050087972013446,0.5805825242718446],[11.716320037841797,5.491412600151543,0.9512141346931458] [0.3538976686818368,0.654368932038835],[11.76646900177002,5.304721099427603,0.9214164614677429] 0.3601233019856572 0.6524271844660194 5150
EfficientNetV2-m exposure="max" without mixed precision (amp.GradScaler()) [0.455899366829671, 0.5786407766990291][11.624638557434082,5.510402609075157,1.029883861541748] [0.3955207692059546,0.6388349514563106] [0.3870887313689006,0.6310679611650486] [0.37190759645808213,0.6427184466019418] [0.3954467958043927,0.6349514563106796] [0.3806486774090985,0.6310679611650486] [0.3694079017430422,0.6349514563106796] [0.4474771832160993,0.5786407766990291] [0.4329707149162454,0.6097087378640776] [0.3630835861605383,0.6330097087378641] 0.38274013996009026 0.6271844660194175 5150
EfficientNetV2-l MLP(256,32,1) [0.3629718165260899,0.6640776699029126],[11.326478958129883,5.176436455967357,0.7270090579986572] [0.3778153902815894,0.654368932038835],[11.756083488464355,5.078827020496998,0.7515597939491272] [0.40490400097933393,0.6621359223300971],[11.678875923156738,5.0282682075083835,0.6951654553413391] [0.3422837114291397,0.6601941747572816],[11.539137840270996,5.027335057906734,0.7135916948318481] [0.3930839488501394,0.654368932038835],[11.62857723236084,5.065871095078663,0.8129523992538452] [0.36973989926423256, 0.6679611650485436],[11.52272891998291, 5.0117800284357905,0.7823890447616577] [0.44649657125737163,0.6388349514563106],[11.628984451293945,5.012184920820218,0.45771318674087524] [0.3442822287250434,0.6679611650485436],[11.645936965942383,5.125562575951363,0.709739625453949] [0.3326264110015347,0.6660194174757281],[11.753767967224121,5.1002887210799654,0.7644101977348328] [0.3626785866047162,0.6563106796116505],[11.611091613769531,5.029804865249153,0.7668406367301941] 0.3578598940821187 0.6621359223300971 5150

10-fold training - testset 10% on EffNetV2 with albumenation (-90,90) rotation

NN-config fold-1 (val_mse,val_acc),(mse, acc) fold-2 fold-3 fold-4 fold-5 fold-6 fold-7 fold-8 fold-9 fold-10 mean MSE mean ACC datset size
EfficientNetV2-m exposure="max" [0.37055935391494754,0.6621359223300971] [0.45588673297899246,0.6233009708737864] [0.3547869452993763,0.6446601941747573] [0.4045900893110558,0.6135922330097088] [0.88552916,0.4406047516198704] [0.48096464346891804,0.6233009708737864] [0.37039235639865814,0.654368932038835] [0.45940533538042655,0.6330097087378641] [0.8034557,0.5205183585313174] [0.5949553335231461,0.6] 0.3811391933261682 0.658252427184466 5150
EfficientNetV2-m exposure="max" MLE savepoints 0.39006542761913027,0.6349514563106796 0.3975690467732811,0.6194174757281553 0.3456209812726769,0.6504854368932039 0.392271179482109,0.6466019417475728 0.3944861143057912,0.6194174757281553 0.36419188526757085,0.6621359223300971 0.32908238390696287,0.6718446601941748 0.4586693277976992,0.5805825242718446 0.4482269191065332,0.6135922330097088 0.38145080447413454,0.6446601941747573 0.40194174757281553 0.6504854368932039 5150
EfficientNetV2-m MLP(256,32,1) exposure="middle" MLE savepoints 0.32134962560309055, 0.6873786407766991 0.37654773499747174,0.6757281553398058 0.3317787046957071,0.683495145631068 0.2848554733474878,0.7106796116504854 0.28457994675059795,0.7009708737864078 0.3254396787970704,0.7048543689320388 0.3107698348896557,0.6990291262135923 0.3481675475700843,0.683495145631068 0.29518738573095393,0.6990291262135923 0.37277993194640846, 0.6601941747572816 0.29181617978158675 0.7242718446601941 5150
EfficientNetV2-l MLP(256,32,1) 9 channels, mse savepoints,test_img=384 0.2919331574757366,
0.7087378640776699
0.28862907527393483,
0.7067961165048544
0.2886753393807174,
0.7067961165048544
0.3264675816582527,
0.7067961165048544
0.30704787144179796,
0.7067961165048544
0.32707414210911223,
0.7067961165048544
0.2831883007978934,
0.7067961165048544
0.2999658127942135,
0.7067961165048544
0.3347793650017895,
0.7067961165048544
0.2951520435189703,
0.7067961165048544
0.2812634011698365 0.7165048543689321 5150
EfficientNetV2-m MLP(256,32,1) 9 channels, mse savepoints,test_img=384 0.28924956303381794,
0.7165048543689321
0.29942306690937437,
0.7067961165048544
0.3025252460057941,
0.6932038834951456
0.28377943850540877,
0.7126213592233009
0.2915233471041366,
0.7184466019417476
0.2871269724253753,
0.7184466019417476
0.3030619876887209,
0.7126213592233009
0.2879829908232323,
0.7165048543689321
0.28862248577395033,
0.7106796116504854
0.29375312456653485,
0.7067961165048544
0.2728000533332227 0.7398058252427184 5150
EfficientNetV2-m MLP(256,32,1) max, mse savepoints,test_img=384,150 epochs 0.30484069945668224,
0.6893203883495146
0.4133824266231807,
0.625242718446602
0.31930824312116274,
0.6679611650485436
0.32717935896508477,
0.7048543689320388
0.30969602468386354,
0.6893203883495146
0.28448606067270993,
0.7087378640776699
0.30904694431735247,
0.6932038834951456
0.3150160931097954,
0.7067961165048544
0.30198777594516435,
0.6970873786407767
0.2869042119366531,
0.7262135922330097
0.28988731306984655 0.7106796116504854 5150
EfficientNetV2-m MLP(256,32,1) max, mse savepoints,test_img=384,450 epochs 0.33681427373177025,
0.6893203883495146
, 0.29654104862479846,
, 0.7009708737864078
, 0.3021051977439794,
, 0.7029126213592233
, 0.29052846705153995,
, 0.7126213592233009
0.3153618507145613,
0.7067961165048544
0.34705555720142651,
0.6854368932038835
0.3384720333945969,
0.6970873786407767
0.3211546217945373,
0.6796116504854369
0.3125481445122048,
0.6912621359223301
0.283231936652271,
0.7184466019417476
0.2891016447191845 0.7126213592233009 5150
EfficientNetV2-m MLP(256,32,1) min, mse savepoints,test_img=384 0.2921728504806332,
0.7106796116504854
0.2915168880493239,
0.7106796116504854
0.29415177969506007,
0.6951456310679611
0.27463061557075813,
0.7339805825242719
0.2978112364938812,
0.7184466019417476
0.3038186086169834,
0.7087378640776699
0.3043179297837437,
0.7087378640776699
0.33064210104040115,
0.6970873786407767
0.30677957823914664,
0.7009708737864078
0.29461632788989833,
0.7145631067961165
0.2733061445705727 0.7398058252427184 5150
EfficientNetV2-l MLP(256,32,1) MLE savepoints, same test image size [0.31076611375098306,0.6893203883495146] [0.4524055561189933,0.5941747572815534]
EfficientNetV2-l old MLP(256,32,1) MLE savepoints, same test image size middle [0.30097150575771525,0.6970873786407767] [0.28132935946744175,0.7339805825242719] [0.29904481883102846,0.6912621359223301 [0.3176863822144107,0.6699029126213593] [0.28227414303887455,0.7184466019417476] [0.3049059022718125,0.6990291262135923] [0.2796412552907852, 0.7262135922330097] 0.33390750726134205,0.6815533980582524] [0.2998588231157967,0.7048543689320388] [0.31008770317437595,0.7029126213592233] 0.27972858674427004 0.7184466019417476 5150
EfficientNetV2-l MLP(256,32,1), mse savepoints, max 0.32196506924850776,
0.7106796116504854
0.2946665601392249,
0.7009708737864078
0.32414991299737056,
0.6990291262135923
0.3531381614789506,
0.7417475728155339
0.29510834368141653,
0.7281553398058253
0.3059995273245565,
0.7106796116504854
0.2705381382875259,
0.7223300970873786
0.2921056222257979,
0.7106796116504854
0.38004589090507696,
0.7106796116504854
0.29924078041383567,
0.7009708737864078
0.28599091634838775 0.7242718446601941 5150
EfficientNetV2-l MLP(256,32,1), mse savepoints, middle 0.29953838222096635,
0.6873786407766991
0.33232208673970726,
0.6796116504854369
0.32039824947911555,
0.6970873786407767
0.30037839009590983,
0.7184466019417476
0.27221360980857046,
0.7106796116504854
0.30174664335615214,
0.7106796116504854
0.29355417419590435,
0.6970873786407767
0.28472557798644504,
0.7048543689320388
0.30734402115092485,
0.7106796116504854
0.2849121101485401,
0.7203883495145631
0.2749396896282028 0.7281553398058253 5150