{ "id": 154336, "title": "Model Performance Summaries", "url": "/competitions/prostate-cancer-grade-assessment/discussion/154336", "author_name": "Tyler Ashworth", "post_date": "2020-05-28T03:36:59.910000", "votes": 13, "comment_count": 1, "views": 0, "content": "

I trained several unfrozen models for 4 epochs to give insights into their performance over many iterations. Some preface:\n - Fixed random seed\n - Input of high resolution tiles down-sampled 2x (modified from @iafoss)\n - Only first 1000 samples used for train/val split\n - CrossEntropyLoss with optimized weighting\n - Ranger optimizer\n - Batch size of 4\n - Differential learning rate from 1e-8 to 1e-4\n - All models and data fit into 12GB of GPU memory

\n\n
\n\n

```\nresnet18\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.997914 2.129031 0.216000 0.098256 00:39\n1 2.824023 2.105198 0.244000 0.208666 00:41\n2 2.791342 1.991528 0.276000 0.241386 00:41\n3 2.582884 2.008977 0.264000 0.231312 00:42

\n\n

resnet34\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.944699 2.327567 0.200000 0.112844 01:05\n1 2.894249 2.059369 0.264000 0.255887 01:05\n2 2.706221 2.029116 0.252000 0.290333 01:05\n3 2.818668 1.975234 0.276000 0.362172 01:06

\n\n

resnet50\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 3.004521 2.056370 0.264000 0.356997 01:30\n1 2.808839 2.047106 0.248000 0.388229 01:30\n2 2.606830 1.901347 0.300000 0.382661 01:31\n3 2.582468 1.913935 0.312000 0.415221 01:31

\n\n

resnet101\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.649943 2.099514 0.256000 0.288527 02:28\n1 2.957282 1.961046 0.296000 0.298058 02:29\n2 2.591007 1.902999 0.288000 0.277532 02:30\n3 2.581020 1.781275 0.364000 0.400865 02:30

\n\n

xresnet18\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.923023 2.179759 0.200000 0.169283 00:45\n1 2.722630 1.953939 0.272000 0.333705 00:46\n2 2.774051 1.917969 0.280000 0.281580 00:47\n3 2.907430 2.070880 0.272000 0.287849 00:48

\n\n

xresnet34\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.880789 2.316653 0.256000 0.244878 01:12\n1 2.871083 2.066851 0.248000 0.299686 01:13\n2 2.672848 2.011911 0.316000 0.389572 01:13\n3 2.656465 1.961482 0.300000 0.343348 01:13

\n\n

xresnet50\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.778955 2.001766 0.300000 0.332816 01:39\n1 2.698803 1.939087 0.236000 0.389846 01:40\n2 2.525798 1.833303 0.268000 0.416991 01:40\n3 2.384005 1.783235 0.260000 0.370382 01:40

\n\n

xresnet18_deep\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.862097 2.361622 0.252000 0.253623 00:50\n1 2.796951 2.532122 0.276000 0.292569 00:51\n2 2.663010 2.295276 0.272000 0.265057 00:52\n3 2.818938 2.270162 0.248000 0.255541 00:52

\n\n

xresnet34_deep\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.859520 2.311229 0.276000 0.140029 01:17\n1 2.699953 2.170536 0.212000 0.073904 01:18\n2 2.659115 2.128215 0.272000 0.189645 01:19\n3 2.705960 2.066520 0.272000 0.273737 01:18

\n\n

xresnet50_deep\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.797954 2.104088 0.216000 0.237150 01:46\n1 2.704217 2.153629 0.260000 0.199761 01:47\n2 2.712135 1.997896 0.244000 0.155668 01:47\n3 2.530224 2.087142 0.236000 0.207157 01:47

\n\n

xresnet18_deeper\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.888201 2.363007 0.204000 0.234281 00:42\n1 2.792895 2.333706 0.244000 0.253041 00:44\n2 2.663843 2.252855 0.260000 0.258998 00:44\n3 2.692466 2.275261 0.280000 0.250343 00:44

\n\n

xresnet34_deeper\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.901855 2.251732 0.268000 0.235092 01:17\n1 2.705344 2.293610 0.252000 0.287296 01:17\n2 2.657172 2.185726 0.276000 0.279795 01:17\n3 2.568786 2.288551 0.244000 0.262008 01:16

\n\n

xresnet50_deeper\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.792145 2.137277 0.276000 0.403038 01:42\n1 2.405737 1.847772 0.328000 0.456206 01:43\n2 2.595207 1.812677 0.312000 0.498666 01:43\n3 2.739977 1.907202 0.284000 0.444319 01:43

\n\n

xresnext18\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.841253 2.205192 0.144000 0.050836 01:04\n1 2.584027 2.089776 0.216000 0.252758 01:05\n2 2.829538 1.982838 0.244000 0.299332 01:05\n3 2.604419 1.980766 0.232000 0.277531 01:06

\n\n

xresnext34\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 3.142097 2.192536 0.216000 0.153920 01:44\n1 2.877317 1.959900 0.240000 0.239767 01:45\n2 2.574792 2.030516 0.228000 0.251452 01:45\n3 2.841407 1.939851 0.240000 0.283612 01:45

\n\n

xresnext50\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.985841 1.986187 0.224000 0.222333 02:15\n1 2.860546 1.904011 0.232000 0.155318 02:17\n2 2.774284 1.917158 0.304000 0.291405 02:21\n3 2.664962 2.046517 0.248000 0.213829 02:19

\n\n

xse_resnet18\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.759439 2.240283 0.156000 0.120766 00:49\n1 2.875750 2.083892 0.212000 0.192887 00:51\n2 2.657371 1.964030 0.252000 0.269273 00:50\n3 2.616734 1.862385 0.284000 0.313874 00:51

\n\n

xse_resnet34\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 3.025935 2.161243 0.236000 0.291339 01:19\n1 2.953254 1.984489 0.264000 0.308011 01:20\n2 2.707938 2.098907 0.244000 0.294739 01:19\n3 2.714735 2.061189 0.220000 0.305739 01:19

\n\n

xse_resnext18\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 3.201864 2.006721 0.192000 0.121550 01:07\n1 2.737963 1.944592 0.256000 0.357435 01:07\n2 2.847851 1.930497 0.240000 0.249285 01:08\n3 2.579204 1.920168 0.260000 0.279745 01:09

\n\n

xse_resnext34\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.952002 2.158503 0.144000 0.168505 01:50\n1 2.690593 2.094943 0.184000 0.192831 01:51\n2 2.679636 1.976317 0.244000 0.369530 01:52\n3 2.705839 1.989389 0.232000 0.297703 01:51

\n\n

xse_resnext50\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.770859 2.048736 0.244000 0.255889 02:30\n1 2.713147 2.037284 0.224000 0.226195 02:31\n2 2.524182 1.858638 0.260000 0.310789 02:31\n3 2.587657 1.794883 0.308000 0.237041 02:30

\n\n

xse_resnext18_deep\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.912724 2.219825 0.164000 0.045606 01:10\n1 2.666023 2.134997 0.196000 0.216765 01:12\n2 2.666162 2.024561 0.236000 0.290008 01:12\n3 2.577030 2.026497 0.208000 0.236935 01:13

\n\n

xse_resnext34_deep\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.896611 2.371519 0.156000 0.036532 01:54\n1 2.788881 2.382403 0.160000 0.145151 01:56\n2 2.666806 2.265215 0.200000 0.105472 01:56\n3 2.832739 2.438379 0.248000 0.338065 01:54

\n\n

xse_resnext50_deep\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.927841 2.482784 0.132000 -0.040772 02:33\n1 2.647558 2.184246 0.136000 0.163014 02:33\n2 2.730430 2.146693 0.196000 0.168116 02:33\n3 2.557507 2.182191 0.180000 0.182766 02:33

\n\n

xse_resnext18_deeper\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.940510 2.249118 0.148000 0.015876 01:01\n1 2.746829 2.192743 0.180000 0.224508 01:02\n2 2.803038 2.213990 0.184000 0.174223 01:03\n3 2.664809 2.077724 0.228000 0.295361 01:04

\n\n

xse_resnext34_deeper\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.946931 2.115391 0.208000 0.214123 01:42\n1 2.775198 2.084097 0.220000 0.176251 01:43\n2 2.752686 2.103553 0.224000 0.251483 01:43\n3 2.584117 2.008688 0.240000 0.256399 01:43

\n\n

xse_resnext50_deeper\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.708858 2.118466 0.228000 0.233001 02:21\n1 2.615856 2.210131 0.220000 0.242481 02:22\n2 2.583560 2.081321 0.232000 0.273392 02:22\n3 2.598797 2.197201 0.224000 0.286468 02:22

\n\n

se_resnet50\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.373027 1.886143 0.244000 0.211762 01:35\n1 1.941451 1.669011 0.360000 0.501695 01:38\n2 1.802921 1.672646 0.344000 0.400246 01:39\n3 1.369761 1.426990 0.420000 0.548746 01:39

\n\n

se_resnet101\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.597258 2.157305 0.256000 0.403364 02:43\n1 2.044147 1.783164 0.332000 0.477507 02:50\n2 1.701595 1.624095 0.368000 0.495470 02:50\n3 1.531487 1.627710 0.384000 0.501729 02:50

\n\n

se_resnext50_32x4d\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.337729 1.945258 0.324000 0.322926 02:27\n1 1.798788 1.662011 0.360000 0.486547 02:26\n2 1.638118 1.603430 0.392000 0.517201 02:25\n3 1.384352 1.504506 0.400000 0.543687 02:24

\n\n

efficientnet_b0\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 1.989156 1.761777 0.288000 0.336867 01:16\n1 1.902808 1.708751 0.264000 0.371345 01:17\n2 1.831351 1.699059 0.292000 0.380623 01:16\n3 1.828497 1.710836 0.248000 0.379986 01:18

\n\n

efficientnet_b1\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 1.998489 1.785805 0.260000 0.279713 01:47\n1 1.911476 1.695981 0.300000 0.353004 01:47\n2 1.846845 1.750414 0.284000 0.337978 01:48\n3 1.821340 1.708410 0.272000 0.232455 01:48

\n\n

efficientnet_b2\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 1.937251 1.766941 0.268000 0.333090 01:50\n1 1.883277 1.838557 0.228000 0.282344 01:52\n2 1.857713 1.723782 0.276000 0.322592 01:53\n3 1.844040 1.776808 0.256000 0.319936 01:53

\n\n

efficientnet_b2a\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 1.943703 1.712293 0.296000 0.308079 01:52\n1 1.911521 1.716539 0.276000 0.293017 01:54\n2 1.875442 1.583848 0.356000 0.446829 01:54\n3 1.816793 1.599105 0.348000 0.377734 01:54

\n\n

efficientnet_b3\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 1.939427 1.818084 0.264000 0.306302 02:20\n1 1.965472 1.722801 0.288000 0.334242 02:23\n2 1.833842 1.685646 0.320000 0.403161 02:24\n3 1.869864 1.641537 0.348000 0.432743 02:24

\n\n

resnest50\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 5.060787 3.422106 0.260000 0.293238 02:05\n1 3.159662 2.845312 0.280000 0.414904 02:09\n2 2.797441 3.018130 0.236000 0.374927 02:12\n3 2.302660 2.574300 0.292000 0.400289 02:12\n```

", "messages": [ { "id": 864512, "postDate": "2020-05-28T03:36:59.910Z", "content": "

I trained several unfrozen models for 4 epochs to give insights into their performance over many iterations. Some preface:\n - Fixed random seed\n - Input of high resolution tiles down-sampled 2x (modified from @iafoss)\n - Only first 1000 samples used for train/val split\n - CrossEntropyLoss with optimized weighting\n - Ranger optimizer\n - Batch size of 4\n - Differential learning rate from 1e-8 to 1e-4\n - All models and data fit into 12GB of GPU memory

\n\n
\n\n

```\nresnet18\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.997914 2.129031 0.216000 0.098256 00:39\n1 2.824023 2.105198 0.244000 0.208666 00:41\n2 2.791342 1.991528 0.276000 0.241386 00:41\n3 2.582884 2.008977 0.264000 0.231312 00:42

\n\n

resnet34\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.944699 2.327567 0.200000 0.112844 01:05\n1 2.894249 2.059369 0.264000 0.255887 01:05\n2 2.706221 2.029116 0.252000 0.290333 01:05\n3 2.818668 1.975234 0.276000 0.362172 01:06

\n\n

resnet50\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 3.004521 2.056370 0.264000 0.356997 01:30\n1 2.808839 2.047106 0.248000 0.388229 01:30\n2 2.606830 1.901347 0.300000 0.382661 01:31\n3 2.582468 1.913935 0.312000 0.415221 01:31

\n\n

resnet101\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.649943 2.099514 0.256000 0.288527 02:28\n1 2.957282 1.961046 0.296000 0.298058 02:29\n2 2.591007 1.902999 0.288000 0.277532 02:30\n3 2.581020 1.781275 0.364000 0.400865 02:30

\n\n

xresnet18\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.923023 2.179759 0.200000 0.169283 00:45\n1 2.722630 1.953939 0.272000 0.333705 00:46\n2 2.774051 1.917969 0.280000 0.281580 00:47\n3 2.907430 2.070880 0.272000 0.287849 00:48

\n\n

xresnet34\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.880789 2.316653 0.256000 0.244878 01:12\n1 2.871083 2.066851 0.248000 0.299686 01:13\n2 2.672848 2.011911 0.316000 0.389572 01:13\n3 2.656465 1.961482 0.300000 0.343348 01:13

\n\n

xresnet50\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.778955 2.001766 0.300000 0.332816 01:39\n1 2.698803 1.939087 0.236000 0.389846 01:40\n2 2.525798 1.833303 0.268000 0.416991 01:40\n3 2.384005 1.783235 0.260000 0.370382 01:40

\n\n

xresnet18_deep\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.862097 2.361622 0.252000 0.253623 00:50\n1 2.796951 2.532122 0.276000 0.292569 00:51\n2 2.663010 2.295276 0.272000 0.265057 00:52\n3 2.818938 2.270162 0.248000 0.255541 00:52

\n\n

xresnet34_deep\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.859520 2.311229 0.276000 0.140029 01:17\n1 2.699953 2.170536 0.212000 0.073904 01:18\n2 2.659115 2.128215 0.272000 0.189645 01:19\n3 2.705960 2.066520 0.272000 0.273737 01:18

\n\n

xresnet50_deep\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.797954 2.104088 0.216000 0.237150 01:46\n1 2.704217 2.153629 0.260000 0.199761 01:47\n2 2.712135 1.997896 0.244000 0.155668 01:47\n3 2.530224 2.087142 0.236000 0.207157 01:47

\n\n

xresnet18_deeper\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.888201 2.363007 0.204000 0.234281 00:42\n1 2.792895 2.333706 0.244000 0.253041 00:44\n2 2.663843 2.252855 0.260000 0.258998 00:44\n3 2.692466 2.275261 0.280000 0.250343 00:44

\n\n

xresnet34_deeper\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.901855 2.251732 0.268000 0.235092 01:17\n1 2.705344 2.293610 0.252000 0.287296 01:17\n2 2.657172 2.185726 0.276000 0.279795 01:17\n3 2.568786 2.288551 0.244000 0.262008 01:16

\n\n

xresnet50_deeper\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.792145 2.137277 0.276000 0.403038 01:42\n1 2.405737 1.847772 0.328000 0.456206 01:43\n2 2.595207 1.812677 0.312000 0.498666 01:43\n3 2.739977 1.907202 0.284000 0.444319 01:43

\n\n

xresnext18\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.841253 2.205192 0.144000 0.050836 01:04\n1 2.584027 2.089776 0.216000 0.252758 01:05\n2 2.829538 1.982838 0.244000 0.299332 01:05\n3 2.604419 1.980766 0.232000 0.277531 01:06

\n\n

xresnext34\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 3.142097 2.192536 0.216000 0.153920 01:44\n1 2.877317 1.959900 0.240000 0.239767 01:45\n2 2.574792 2.030516 0.228000 0.251452 01:45\n3 2.841407 1.939851 0.240000 0.283612 01:45

\n\n

xresnext50\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.985841 1.986187 0.224000 0.222333 02:15\n1 2.860546 1.904011 0.232000 0.155318 02:17\n2 2.774284 1.917158 0.304000 0.291405 02:21\n3 2.664962 2.046517 0.248000 0.213829 02:19

\n\n

xse_resnet18\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.759439 2.240283 0.156000 0.120766 00:49\n1 2.875750 2.083892 0.212000 0.192887 00:51\n2 2.657371 1.964030 0.252000 0.269273 00:50\n3 2.616734 1.862385 0.284000 0.313874 00:51

\n\n

xse_resnet34\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 3.025935 2.161243 0.236000 0.291339 01:19\n1 2.953254 1.984489 0.264000 0.308011 01:20\n2 2.707938 2.098907 0.244000 0.294739 01:19\n3 2.714735 2.061189 0.220000 0.305739 01:19

\n\n

xse_resnext18\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 3.201864 2.006721 0.192000 0.121550 01:07\n1 2.737963 1.944592 0.256000 0.357435 01:07\n2 2.847851 1.930497 0.240000 0.249285 01:08\n3 2.579204 1.920168 0.260000 0.279745 01:09

\n\n

xse_resnext34\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.952002 2.158503 0.144000 0.168505 01:50\n1 2.690593 2.094943 0.184000 0.192831 01:51\n2 2.679636 1.976317 0.244000 0.369530 01:52\n3 2.705839 1.989389 0.232000 0.297703 01:51

\n\n

xse_resnext50\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.770859 2.048736 0.244000 0.255889 02:30\n1 2.713147 2.037284 0.224000 0.226195 02:31\n2 2.524182 1.858638 0.260000 0.310789 02:31\n3 2.587657 1.794883 0.308000 0.237041 02:30

\n\n

xse_resnext18_deep\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.912724 2.219825 0.164000 0.045606 01:10\n1 2.666023 2.134997 0.196000 0.216765 01:12\n2 2.666162 2.024561 0.236000 0.290008 01:12\n3 2.577030 2.026497 0.208000 0.236935 01:13

\n\n

xse_resnext34_deep\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.896611 2.371519 0.156000 0.036532 01:54\n1 2.788881 2.382403 0.160000 0.145151 01:56\n2 2.666806 2.265215 0.200000 0.105472 01:56\n3 2.832739 2.438379 0.248000 0.338065 01:54

\n\n

xse_resnext50_deep\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.927841 2.482784 0.132000 -0.040772 02:33\n1 2.647558 2.184246 0.136000 0.163014 02:33\n2 2.730430 2.146693 0.196000 0.168116 02:33\n3 2.557507 2.182191 0.180000 0.182766 02:33

\n\n

xse_resnext18_deeper\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.940510 2.249118 0.148000 0.015876 01:01\n1 2.746829 2.192743 0.180000 0.224508 01:02\n2 2.803038 2.213990 0.184000 0.174223 01:03\n3 2.664809 2.077724 0.228000 0.295361 01:04

\n\n

xse_resnext34_deeper\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.946931 2.115391 0.208000 0.214123 01:42\n1 2.775198 2.084097 0.220000 0.176251 01:43\n2 2.752686 2.103553 0.224000 0.251483 01:43\n3 2.584117 2.008688 0.240000 0.256399 01:43

\n\n

xse_resnext50_deeper\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.708858 2.118466 0.228000 0.233001 02:21\n1 2.615856 2.210131 0.220000 0.242481 02:22\n2 2.583560 2.081321 0.232000 0.273392 02:22\n3 2.598797 2.197201 0.224000 0.286468 02:22

\n\n

se_resnet50\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.373027 1.886143 0.244000 0.211762 01:35\n1 1.941451 1.669011 0.360000 0.501695 01:38\n2 1.802921 1.672646 0.344000 0.400246 01:39\n3 1.369761 1.426990 0.420000 0.548746 01:39

\n\n

se_resnet101\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.597258 2.157305 0.256000 0.403364 02:43\n1 2.044147 1.783164 0.332000 0.477507 02:50\n2 1.701595 1.624095 0.368000 0.495470 02:50\n3 1.531487 1.627710 0.384000 0.501729 02:50

\n\n

se_resnext50_32x4d\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 2.337729 1.945258 0.324000 0.322926 02:27\n1 1.798788 1.662011 0.360000 0.486547 02:26\n2 1.638118 1.603430 0.392000 0.517201 02:25\n3 1.384352 1.504506 0.400000 0.543687 02:24

\n\n

efficientnet_b0\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 1.989156 1.761777 0.288000 0.336867 01:16\n1 1.902808 1.708751 0.264000 0.371345 01:17\n2 1.831351 1.699059 0.292000 0.380623 01:16\n3 1.828497 1.710836 0.248000 0.379986 01:18

\n\n

efficientnet_b1\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 1.998489 1.785805 0.260000 0.279713 01:47\n1 1.911476 1.695981 0.300000 0.353004 01:47\n2 1.846845 1.750414 0.284000 0.337978 01:48\n3 1.821340 1.708410 0.272000 0.232455 01:48

\n\n

efficientnet_b2\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 1.937251 1.766941 0.268000 0.333090 01:50\n1 1.883277 1.838557 0.228000 0.282344 01:52\n2 1.857713 1.723782 0.276000 0.322592 01:53\n3 1.844040 1.776808 0.256000 0.319936 01:53

\n\n

efficientnet_b2a\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 1.943703 1.712293 0.296000 0.308079 01:52\n1 1.911521 1.716539 0.276000 0.293017 01:54\n2 1.875442 1.583848 0.356000 0.446829 01:54\n3 1.816793 1.599105 0.348000 0.377734 01:54

\n\n

efficientnet_b3\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 1.939427 1.818084 0.264000 0.306302 02:20\n1 1.965472 1.722801 0.288000 0.334242 02:23\n2 1.833842 1.685646 0.320000 0.403161 02:24\n3 1.869864 1.641537 0.348000 0.432743 02:24

\n\n

resnest50\nepoch train_loss valid_loss accuracy cohen_kappa_score time\n0 5.060787 3.422106 0.260000 0.293238 02:05\n1 3.159662 2.845312 0.280000 0.414904 02:09\n2 2.797441 3.018130 0.236000 0.374927 02:12\n3 2.302660 2.574300 0.292000 0.400289 02:12\n```

", "rawMarkdown": "I trained several unfrozen models for 4 epochs to give insights into their performance over many iterations. Some preface:\n - Fixed random seed\n - Input of high resolution tiles down-sampled 2x (modified from @iafoss)\n - Only first 1000 samples used for train/val split\n - CrossEntropyLoss with optimized weighting\n - Ranger optimizer\n - Batch size of 4\n - Differential learning rate from 1e-8 to 1e-4\n - All models and data fit into 12GB of GPU memory\n\n---\n\n```\nresnet18\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.997914 \t2.129031 \t0.216000 \t0.098256 \t 00:39\n1 \t 2.824023 \t2.105198 \t0.244000 \t0.208666 \t 00:41\n2 \t 2.791342 \t1.991528 \t0.276000 \t0.241386 \t 00:41\n3 \t 2.582884 \t2.008977 \t0.264000 \t0.231312 \t 00:42\n\nresnet34\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.944699 \t2.327567 \t0.200000 \t0.112844 \t 01:05\n1 \t 2.894249 \t2.059369 \t0.264000 \t0.255887 \t 01:05\n2 \t 2.706221 \t2.029116 \t0.252000 \t0.290333 \t 01:05\n3 \t 2.818668 \t1.975234 \t0.276000 \t0.362172 \t 01:06\n\nresnet50\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 3.004521 \t2.056370 \t0.264000 \t0.356997 \t 01:30\n1 \t 2.808839 \t2.047106 \t0.248000 \t0.388229 \t 01:30\n2 \t 2.606830 \t1.901347 \t0.300000 \t0.382661 \t 01:31\n3 \t 2.582468 \t1.913935 \t0.312000 \t0.415221 \t 01:31\n\nresnet101\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.649943 \t2.099514 \t0.256000 \t0.288527 \t 02:28\n1 \t 2.957282 \t1.961046 \t0.296000 \t0.298058 \t 02:29\n2 \t 2.591007 \t1.902999 \t0.288000 \t0.277532 \t 02:30\n3 \t 2.581020 \t1.781275 \t0.364000 \t0.400865 \t 02:30\n\nxresnet18\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.923023 \t2.179759 \t0.200000 \t0.169283 \t 00:45\n1 \t 2.722630 \t1.953939 \t0.272000 \t0.333705 \t 00:46\n2 \t 2.774051 \t1.917969 \t0.280000 \t0.281580 \t 00:47\n3 \t 2.907430 \t2.070880 \t0.272000 \t0.287849 \t 00:48\n\nxresnet34\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.880789 \t2.316653 \t0.256000 \t0.244878 \t 01:12\n1 \t 2.871083 \t2.066851 \t0.248000 \t0.299686 \t 01:13\n2 \t 2.672848 \t2.011911 \t0.316000 \t0.389572 \t 01:13\n3 \t 2.656465 \t1.961482 \t0.300000 \t0.343348 \t 01:13\n\nxresnet50\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.778955 \t2.001766 \t0.300000 \t0.332816 \t 01:39\n1 \t 2.698803 \t1.939087 \t0.236000 \t0.389846 \t 01:40\n2 \t 2.525798 \t1.833303 \t0.268000 \t0.416991 \t 01:40\n3 \t 2.384005 \t1.783235 \t0.260000 \t0.370382 \t 01:40\n\nxresnet18_deep\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.862097 \t2.361622 \t0.252000 \t0.253623 \t 00:50\n1 \t 2.796951 \t2.532122 \t0.276000 \t0.292569 \t 00:51\n2 \t 2.663010 \t2.295276 \t0.272000 \t0.265057 \t 00:52\n3 \t 2.818938 \t2.270162 \t0.248000 \t0.255541 \t 00:52\n\nxresnet34_deep\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.859520 \t2.311229 \t0.276000 \t0.140029 \t 01:17\n1 \t 2.699953 \t2.170536 \t0.212000 \t0.073904 \t 01:18\n2 \t 2.659115 \t2.128215 \t0.272000 \t0.189645 \t 01:19\n3 \t 2.705960 \t2.066520 \t0.272000 \t0.273737 \t 01:18\n\nxresnet50_deep\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.797954 \t2.104088 \t0.216000 \t0.237150 \t 01:46\n1 \t 2.704217 \t2.153629 \t0.260000 \t0.199761 \t 01:47\n2 \t 2.712135 \t1.997896 \t0.244000 \t0.155668 \t 01:47\n3 \t 2.530224 \t2.087142 \t0.236000 \t0.207157 \t 01:47\n\nxresnet18_deeper\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.888201 \t2.363007 \t0.204000 \t0.234281 \t 00:42\n1 \t 2.792895 \t2.333706 \t0.244000 \t0.253041 \t 00:44\n2 \t 2.663843 \t2.252855 \t0.260000 \t0.258998 \t 00:44\n3 \t 2.692466 \t2.275261 \t0.280000 \t0.250343 \t 00:44\n\nxresnet34_deeper\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.901855 \t2.251732 \t0.268000 \t0.235092 \t 01:17\n1 \t 2.705344 \t2.293610 \t0.252000 \t0.287296 \t 01:17\n2 \t 2.657172 \t2.185726 \t0.276000 \t0.279795 \t 01:17\n3 \t 2.568786 \t2.288551 \t0.244000 \t0.262008 \t 01:16\n\nxresnet50_deeper\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.792145 \t2.137277 \t0.276000 \t0.403038 \t 01:42\n1 \t 2.405737 \t1.847772 \t0.328000 \t0.456206 \t 01:43\n2 \t 2.595207 \t1.812677 \t0.312000 \t0.498666 \t 01:43\n3 \t 2.739977 \t1.907202 \t0.284000 \t0.444319 \t 01:43\n\nxresnext18\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.841253 \t2.205192 \t0.144000 \t0.050836 \t 01:04\n1 \t 2.584027 \t2.089776 \t0.216000 \t0.252758 \t 01:05\n2 \t 2.829538 \t1.982838 \t0.244000 \t0.299332 \t 01:05\n3 \t 2.604419 \t1.980766 \t0.232000 \t0.277531 \t 01:06\n\nxresnext34\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 3.142097 \t2.192536 \t0.216000 \t0.153920 \t 01:44\n1 \t 2.877317 \t1.959900 \t0.240000 \t0.239767 \t 01:45\n2 \t 2.574792 \t2.030516 \t0.228000 \t0.251452 \t 01:45\n3 \t 2.841407 \t1.939851 \t0.240000 \t0.283612 \t 01:45\n\nxresnext50\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.985841 \t1.986187 \t0.224000 \t0.222333 \t 02:15\n1 \t 2.860546 \t1.904011 \t0.232000 \t0.155318 \t 02:17\n2 \t 2.774284 \t1.917158 \t0.304000 \t0.291405 \t 02:21\n3 \t 2.664962 \t2.046517 \t0.248000 \t0.213829 \t 02:19\n\nxse_resnet18\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.759439 \t2.240283 \t0.156000 \t0.120766 \t 00:49\n1 \t 2.875750 \t2.083892 \t0.212000 \t0.192887 \t 00:51\n2 \t 2.657371 \t1.964030 \t0.252000 \t0.269273 \t 00:50\n3 \t 2.616734 \t1.862385 \t0.284000 \t0.313874 \t 00:51\n\nxse_resnet34\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 3.025935 \t2.161243 \t0.236000 \t0.291339 \t 01:19\n1 \t 2.953254 \t1.984489 \t0.264000 \t0.308011 \t 01:20\n2 \t 2.707938 \t2.098907 \t0.244000 \t0.294739 \t 01:19\n3 \t 2.714735 \t2.061189 \t0.220000 \t0.305739 \t 01:19\n\nxse_resnext18\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 3.201864 \t2.006721 \t0.192000 \t0.121550 \t 01:07\n1 \t 2.737963 \t1.944592 \t0.256000 \t0.357435 \t 01:07\n2 \t 2.847851 \t1.930497 \t0.240000 \t0.249285 \t 01:08\n3 \t 2.579204 \t1.920168 \t0.260000 \t0.279745 \t 01:09\n\nxse_resnext34\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.952002 \t2.158503 \t0.144000 \t0.168505 \t 01:50\n1 \t 2.690593 \t2.094943 \t0.184000 \t0.192831 \t 01:51\n2 \t 2.679636 \t1.976317 \t0.244000 \t0.369530 \t 01:52\n3 \t 2.705839 \t1.989389 \t0.232000 \t0.297703 \t 01:51\n\nxse_resnext50\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.770859 \t2.048736 \t0.244000 \t0.255889 \t 02:30\n1 \t 2.713147 \t2.037284 \t0.224000 \t0.226195 \t 02:31\n2 \t 2.524182 \t1.858638 \t0.260000 \t0.310789 \t 02:31\n3 \t 2.587657 \t1.794883 \t0.308000 \t0.237041 \t 02:30\n\nxse_resnext18_deep\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.912724 \t2.219825 \t0.164000 \t0.045606 \t 01:10\n1 \t 2.666023 \t2.134997 \t0.196000 \t0.216765 \t 01:12\n2 \t 2.666162 \t2.024561 \t0.236000 \t0.290008 \t 01:12\n3 \t 2.577030 \t2.026497 \t0.208000 \t0.236935 \t 01:13\n\nxse_resnext34_deep\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.896611 \t2.371519 \t0.156000 \t0.036532 \t 01:54\n1 \t 2.788881 \t2.382403 \t0.160000 \t0.145151 \t 01:56\n2 \t 2.666806 \t2.265215 \t0.200000 \t0.105472 \t 01:56\n3 \t 2.832739 \t2.438379 \t0.248000 \t0.338065 \t 01:54\n\nxse_resnext50_deep\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.927841 \t2.482784 \t0.132000 \t-0.040772 \t 02:33\n1 \t 2.647558 \t2.184246 \t0.136000 \t0.163014 \t 02:33\n2 \t 2.730430 \t2.146693 \t0.196000 \t0.168116 \t 02:33\n3 \t 2.557507 \t2.182191 \t0.180000 \t0.182766 \t 02:33\n\nxse_resnext18_deeper\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.940510 \t2.249118 \t0.148000 \t0.015876 \t 01:01\n1 \t 2.746829 \t2.192743 \t0.180000 \t0.224508 \t 01:02\n2 \t 2.803038 \t2.213990 \t0.184000 \t0.174223 \t 01:03\n3 \t 2.664809 \t2.077724 \t0.228000 \t0.295361 \t 01:04\n\nxse_resnext34_deeper\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.946931 \t2.115391 \t0.208000 \t0.214123 \t 01:42\n1 \t 2.775198 \t2.084097 \t0.220000 \t0.176251 \t 01:43\n2 \t 2.752686 \t2.103553 \t0.224000 \t0.251483 \t 01:43\n3 \t 2.584117 \t2.008688 \t0.240000 \t0.256399 \t 01:43\n\nxse_resnext50_deeper\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.708858 \t2.118466 \t0.228000 \t0.233001 \t 02:21\n1 \t 2.615856 \t2.210131 \t0.220000 \t0.242481 \t 02:22\n2 \t 2.583560 \t2.081321 \t0.232000 \t0.273392 \t 02:22\n3 \t 2.598797 \t2.197201 \t0.224000 \t0.286468 \t 02:22\n\nse_resnet50\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.373027 \t1.886143 \t0.244000 \t0.211762 \t 01:35\n1 \t 1.941451 \t1.669011 \t0.360000 \t0.501695 \t 01:38\n2 \t 1.802921 \t1.672646 \t0.344000 \t0.400246 \t 01:39\n3 \t 1.369761 \t1.426990 \t0.420000 \t0.548746 \t 01:39\n\nse_resnet101\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.597258 \t2.157305 \t0.256000 \t0.403364 \t 02:43\n1 \t 2.044147 \t1.783164 \t0.332000 \t0.477507 \t 02:50\n2 \t 1.701595 \t1.624095 \t0.368000 \t0.495470 \t 02:50\n3 \t 1.531487 \t1.627710 \t0.384000 \t0.501729 \t 02:50\n\nse_resnext50_32x4d\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.337729 \t1.945258 \t0.324000 \t0.322926 \t 02:27\n1 \t 1.798788 \t1.662011 \t0.360000 \t0.486547 \t 02:26\n2 \t 1.638118 \t1.603430 \t0.392000 \t0.517201 \t 02:25\n3 \t 1.384352 \t1.504506 \t0.400000 \t0.543687 \t 02:24\n\nefficientnet_b0\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 1.989156 \t1.761777 \t0.288000 \t0.336867 \t 01:16\n1 \t 1.902808 \t1.708751 \t0.264000 \t0.371345 \t 01:17\n2 \t 1.831351 \t1.699059 \t0.292000 \t0.380623 \t 01:16\n3 \t 1.828497 \t1.710836 \t0.248000 \t0.379986 \t 01:18\n\nefficientnet_b1\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 1.998489 \t1.785805 \t0.260000 \t0.279713 \t 01:47\n1 \t 1.911476 \t1.695981 \t0.300000 \t0.353004 \t 01:47\n2 \t 1.846845 \t1.750414 \t0.284000 \t0.337978 \t 01:48\n3 \t 1.821340 \t1.708410 \t0.272000 \t0.232455 \t 01:48\n\nefficientnet_b2\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 1.937251 \t1.766941 \t0.268000 \t0.333090 \t 01:50\n1 \t 1.883277 \t1.838557 \t0.228000 \t0.282344 \t 01:52\n2 \t 1.857713 \t1.723782 \t0.276000 \t0.322592 \t 01:53\n3 \t 1.844040 \t1.776808 \t0.256000 \t0.319936 \t 01:53\n\nefficientnet_b2a\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 1.943703 \t1.712293 \t0.296000 \t0.308079 \t 01:52\n1 \t 1.911521 \t1.716539 \t0.276000 \t0.293017 \t 01:54\n2 \t 1.875442 \t1.583848 \t0.356000 \t0.446829 \t 01:54\n3 \t 1.816793 \t1.599105 \t0.348000 \t0.377734 \t 01:54\n\nefficientnet_b3\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 1.939427 \t1.818084 \t0.264000 \t0.306302 \t 02:20\n1 \t 1.965472 \t1.722801 \t0.288000 \t0.334242 \t 02:23\n2 \t 1.833842 \t1.685646 \t0.320000 \t0.403161 \t 02:24\n3 \t 1.869864 \t1.641537 \t0.348000 \t0.432743 \t 02:24\n\nresnest50\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 5.060787 \t3.422106 \t0.260000 \t0.293238 \t 02:05\n1 \t 3.159662 \t2.845312 \t0.280000 \t0.414904 \t 02:09\n2 \t 2.797441 \t3.018130 \t0.236000 \t0.374927 \t 02:12\n3 \t 2.302660 \t2.574300 \t0.292000 \t0.400289 \t 02:12\n```", "votes": 13 }, { "id": 865986, "postDate": "2020-05-29T02:50:38.040Z", "content": "

Congratulations!!!

", "rawMarkdown": "Congratulations!!!" } ], "comments": [ { "id": 865986, "author_name": "Alexandre Vilela Calmon", "author_url": "", "post_date": "2020-05-29T02:50:38.040000", "content": "

Congratulations!!!

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "864512": "I trained several unfrozen models for 4 epochs to give insights into their performance over many iterations. Some preface:\n - Fixed random seed\n - Input of high resolution tiles down-sampled 2x (modified from @iafoss)\n - Only first 1000 samples used for train/val split\n - CrossEntropyLoss with optimized weighting\n - Ranger optimizer\n - Batch size of 4\n - Differential learning rate from 1e-8 to 1e-4\n - All models and data fit into 12GB of GPU memory\n\n---\n\n```\nresnet18\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.997914 \t2.129031 \t0.216000 \t0.098256 \t 00:39\n1 \t 2.824023 \t2.105198 \t0.244000 \t0.208666 \t 00:41\n2 \t 2.791342 \t1.991528 \t0.276000 \t0.241386 \t 00:41\n3 \t 2.582884 \t2.008977 \t0.264000 \t0.231312 \t 00:42\n\nresnet34\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.944699 \t2.327567 \t0.200000 \t0.112844 \t 01:05\n1 \t 2.894249 \t2.059369 \t0.264000 \t0.255887 \t 01:05\n2 \t 2.706221 \t2.029116 \t0.252000 \t0.290333 \t 01:05\n3 \t 2.818668 \t1.975234 \t0.276000 \t0.362172 \t 01:06\n\nresnet50\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 3.004521 \t2.056370 \t0.264000 \t0.356997 \t 01:30\n1 \t 2.808839 \t2.047106 \t0.248000 \t0.388229 \t 01:30\n2 \t 2.606830 \t1.901347 \t0.300000 \t0.382661 \t 01:31\n3 \t 2.582468 \t1.913935 \t0.312000 \t0.415221 \t 01:31\n\nresnet101\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.649943 \t2.099514 \t0.256000 \t0.288527 \t 02:28\n1 \t 2.957282 \t1.961046 \t0.296000 \t0.298058 \t 02:29\n2 \t 2.591007 \t1.902999 \t0.288000 \t0.277532 \t 02:30\n3 \t 2.581020 \t1.781275 \t0.364000 \t0.400865 \t 02:30\n\nxresnet18\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.923023 \t2.179759 \t0.200000 \t0.169283 \t 00:45\n1 \t 2.722630 \t1.953939 \t0.272000 \t0.333705 \t 00:46\n2 \t 2.774051 \t1.917969 \t0.280000 \t0.281580 \t 00:47\n3 \t 2.907430 \t2.070880 \t0.272000 \t0.287849 \t 00:48\n\nxresnet34\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.880789 \t2.316653 \t0.256000 \t0.244878 \t 01:12\n1 \t 2.871083 \t2.066851 \t0.248000 \t0.299686 \t 01:13\n2 \t 2.672848 \t2.011911 \t0.316000 \t0.389572 \t 01:13\n3 \t 2.656465 \t1.961482 \t0.300000 \t0.343348 \t 01:13\n\nxresnet50\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.778955 \t2.001766 \t0.300000 \t0.332816 \t 01:39\n1 \t 2.698803 \t1.939087 \t0.236000 \t0.389846 \t 01:40\n2 \t 2.525798 \t1.833303 \t0.268000 \t0.416991 \t 01:40\n3 \t 2.384005 \t1.783235 \t0.260000 \t0.370382 \t 01:40\n\nxresnet18_deep\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.862097 \t2.361622 \t0.252000 \t0.253623 \t 00:50\n1 \t 2.796951 \t2.532122 \t0.276000 \t0.292569 \t 00:51\n2 \t 2.663010 \t2.295276 \t0.272000 \t0.265057 \t 00:52\n3 \t 2.818938 \t2.270162 \t0.248000 \t0.255541 \t 00:52\n\nxresnet34_deep\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.859520 \t2.311229 \t0.276000 \t0.140029 \t 01:17\n1 \t 2.699953 \t2.170536 \t0.212000 \t0.073904 \t 01:18\n2 \t 2.659115 \t2.128215 \t0.272000 \t0.189645 \t 01:19\n3 \t 2.705960 \t2.066520 \t0.272000 \t0.273737 \t 01:18\n\nxresnet50_deep\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.797954 \t2.104088 \t0.216000 \t0.237150 \t 01:46\n1 \t 2.704217 \t2.153629 \t0.260000 \t0.199761 \t 01:47\n2 \t 2.712135 \t1.997896 \t0.244000 \t0.155668 \t 01:47\n3 \t 2.530224 \t2.087142 \t0.236000 \t0.207157 \t 01:47\n\nxresnet18_deeper\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.888201 \t2.363007 \t0.204000 \t0.234281 \t 00:42\n1 \t 2.792895 \t2.333706 \t0.244000 \t0.253041 \t 00:44\n2 \t 2.663843 \t2.252855 \t0.260000 \t0.258998 \t 00:44\n3 \t 2.692466 \t2.275261 \t0.280000 \t0.250343 \t 00:44\n\nxresnet34_deeper\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.901855 \t2.251732 \t0.268000 \t0.235092 \t 01:17\n1 \t 2.705344 \t2.293610 \t0.252000 \t0.287296 \t 01:17\n2 \t 2.657172 \t2.185726 \t0.276000 \t0.279795 \t 01:17\n3 \t 2.568786 \t2.288551 \t0.244000 \t0.262008 \t 01:16\n\nxresnet50_deeper\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.792145 \t2.137277 \t0.276000 \t0.403038 \t 01:42\n1 \t 2.405737 \t1.847772 \t0.328000 \t0.456206 \t 01:43\n2 \t 2.595207 \t1.812677 \t0.312000 \t0.498666 \t 01:43\n3 \t 2.739977 \t1.907202 \t0.284000 \t0.444319 \t 01:43\n\nxresnext18\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.841253 \t2.205192 \t0.144000 \t0.050836 \t 01:04\n1 \t 2.584027 \t2.089776 \t0.216000 \t0.252758 \t 01:05\n2 \t 2.829538 \t1.982838 \t0.244000 \t0.299332 \t 01:05\n3 \t 2.604419 \t1.980766 \t0.232000 \t0.277531 \t 01:06\n\nxresnext34\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 3.142097 \t2.192536 \t0.216000 \t0.153920 \t 01:44\n1 \t 2.877317 \t1.959900 \t0.240000 \t0.239767 \t 01:45\n2 \t 2.574792 \t2.030516 \t0.228000 \t0.251452 \t 01:45\n3 \t 2.841407 \t1.939851 \t0.240000 \t0.283612 \t 01:45\n\nxresnext50\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.985841 \t1.986187 \t0.224000 \t0.222333 \t 02:15\n1 \t 2.860546 \t1.904011 \t0.232000 \t0.155318 \t 02:17\n2 \t 2.774284 \t1.917158 \t0.304000 \t0.291405 \t 02:21\n3 \t 2.664962 \t2.046517 \t0.248000 \t0.213829 \t 02:19\n\nxse_resnet18\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.759439 \t2.240283 \t0.156000 \t0.120766 \t 00:49\n1 \t 2.875750 \t2.083892 \t0.212000 \t0.192887 \t 00:51\n2 \t 2.657371 \t1.964030 \t0.252000 \t0.269273 \t 00:50\n3 \t 2.616734 \t1.862385 \t0.284000 \t0.313874 \t 00:51\n\nxse_resnet34\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 3.025935 \t2.161243 \t0.236000 \t0.291339 \t 01:19\n1 \t 2.953254 \t1.984489 \t0.264000 \t0.308011 \t 01:20\n2 \t 2.707938 \t2.098907 \t0.244000 \t0.294739 \t 01:19\n3 \t 2.714735 \t2.061189 \t0.220000 \t0.305739 \t 01:19\n\nxse_resnext18\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 3.201864 \t2.006721 \t0.192000 \t0.121550 \t 01:07\n1 \t 2.737963 \t1.944592 \t0.256000 \t0.357435 \t 01:07\n2 \t 2.847851 \t1.930497 \t0.240000 \t0.249285 \t 01:08\n3 \t 2.579204 \t1.920168 \t0.260000 \t0.279745 \t 01:09\n\nxse_resnext34\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.952002 \t2.158503 \t0.144000 \t0.168505 \t 01:50\n1 \t 2.690593 \t2.094943 \t0.184000 \t0.192831 \t 01:51\n2 \t 2.679636 \t1.976317 \t0.244000 \t0.369530 \t 01:52\n3 \t 2.705839 \t1.989389 \t0.232000 \t0.297703 \t 01:51\n\nxse_resnext50\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.770859 \t2.048736 \t0.244000 \t0.255889 \t 02:30\n1 \t 2.713147 \t2.037284 \t0.224000 \t0.226195 \t 02:31\n2 \t 2.524182 \t1.858638 \t0.260000 \t0.310789 \t 02:31\n3 \t 2.587657 \t1.794883 \t0.308000 \t0.237041 \t 02:30\n\nxse_resnext18_deep\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.912724 \t2.219825 \t0.164000 \t0.045606 \t 01:10\n1 \t 2.666023 \t2.134997 \t0.196000 \t0.216765 \t 01:12\n2 \t 2.666162 \t2.024561 \t0.236000 \t0.290008 \t 01:12\n3 \t 2.577030 \t2.026497 \t0.208000 \t0.236935 \t 01:13\n\nxse_resnext34_deep\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.896611 \t2.371519 \t0.156000 \t0.036532 \t 01:54\n1 \t 2.788881 \t2.382403 \t0.160000 \t0.145151 \t 01:56\n2 \t 2.666806 \t2.265215 \t0.200000 \t0.105472 \t 01:56\n3 \t 2.832739 \t2.438379 \t0.248000 \t0.338065 \t 01:54\n\nxse_resnext50_deep\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.927841 \t2.482784 \t0.132000 \t-0.040772 \t 02:33\n1 \t 2.647558 \t2.184246 \t0.136000 \t0.163014 \t 02:33\n2 \t 2.730430 \t2.146693 \t0.196000 \t0.168116 \t 02:33\n3 \t 2.557507 \t2.182191 \t0.180000 \t0.182766 \t 02:33\n\nxse_resnext18_deeper\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.940510 \t2.249118 \t0.148000 \t0.015876 \t 01:01\n1 \t 2.746829 \t2.192743 \t0.180000 \t0.224508 \t 01:02\n2 \t 2.803038 \t2.213990 \t0.184000 \t0.174223 \t 01:03\n3 \t 2.664809 \t2.077724 \t0.228000 \t0.295361 \t 01:04\n\nxse_resnext34_deeper\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.946931 \t2.115391 \t0.208000 \t0.214123 \t 01:42\n1 \t 2.775198 \t2.084097 \t0.220000 \t0.176251 \t 01:43\n2 \t 2.752686 \t2.103553 \t0.224000 \t0.251483 \t 01:43\n3 \t 2.584117 \t2.008688 \t0.240000 \t0.256399 \t 01:43\n\nxse_resnext50_deeper\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.708858 \t2.118466 \t0.228000 \t0.233001 \t 02:21\n1 \t 2.615856 \t2.210131 \t0.220000 \t0.242481 \t 02:22\n2 \t 2.583560 \t2.081321 \t0.232000 \t0.273392 \t 02:22\n3 \t 2.598797 \t2.197201 \t0.224000 \t0.286468 \t 02:22\n\nse_resnet50\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.373027 \t1.886143 \t0.244000 \t0.211762 \t 01:35\n1 \t 1.941451 \t1.669011 \t0.360000 \t0.501695 \t 01:38\n2 \t 1.802921 \t1.672646 \t0.344000 \t0.400246 \t 01:39\n3 \t 1.369761 \t1.426990 \t0.420000 \t0.548746 \t 01:39\n\nse_resnet101\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.597258 \t2.157305 \t0.256000 \t0.403364 \t 02:43\n1 \t 2.044147 \t1.783164 \t0.332000 \t0.477507 \t 02:50\n2 \t 1.701595 \t1.624095 \t0.368000 \t0.495470 \t 02:50\n3 \t 1.531487 \t1.627710 \t0.384000 \t0.501729 \t 02:50\n\nse_resnext50_32x4d\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 2.337729 \t1.945258 \t0.324000 \t0.322926 \t 02:27\n1 \t 1.798788 \t1.662011 \t0.360000 \t0.486547 \t 02:26\n2 \t 1.638118 \t1.603430 \t0.392000 \t0.517201 \t 02:25\n3 \t 1.384352 \t1.504506 \t0.400000 \t0.543687 \t 02:24\n\nefficientnet_b0\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 1.989156 \t1.761777 \t0.288000 \t0.336867 \t 01:16\n1 \t 1.902808 \t1.708751 \t0.264000 \t0.371345 \t 01:17\n2 \t 1.831351 \t1.699059 \t0.292000 \t0.380623 \t 01:16\n3 \t 1.828497 \t1.710836 \t0.248000 \t0.379986 \t 01:18\n\nefficientnet_b1\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 1.998489 \t1.785805 \t0.260000 \t0.279713 \t 01:47\n1 \t 1.911476 \t1.695981 \t0.300000 \t0.353004 \t 01:47\n2 \t 1.846845 \t1.750414 \t0.284000 \t0.337978 \t 01:48\n3 \t 1.821340 \t1.708410 \t0.272000 \t0.232455 \t 01:48\n\nefficientnet_b2\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 1.937251 \t1.766941 \t0.268000 \t0.333090 \t 01:50\n1 \t 1.883277 \t1.838557 \t0.228000 \t0.282344 \t 01:52\n2 \t 1.857713 \t1.723782 \t0.276000 \t0.322592 \t 01:53\n3 \t 1.844040 \t1.776808 \t0.256000 \t0.319936 \t 01:53\n\nefficientnet_b2a\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 1.943703 \t1.712293 \t0.296000 \t0.308079 \t 01:52\n1 \t 1.911521 \t1.716539 \t0.276000 \t0.293017 \t 01:54\n2 \t 1.875442 \t1.583848 \t0.356000 \t0.446829 \t 01:54\n3 \t 1.816793 \t1.599105 \t0.348000 \t0.377734 \t 01:54\n\nefficientnet_b3\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 1.939427 \t1.818084 \t0.264000 \t0.306302 \t 02:20\n1 \t 1.965472 \t1.722801 \t0.288000 \t0.334242 \t 02:23\n2 \t 1.833842 \t1.685646 \t0.320000 \t0.403161 \t 02:24\n3 \t 1.869864 \t1.641537 \t0.348000 \t0.432743 \t 02:24\n\nresnest50\nepoch \ttrain_loss \tvalid_loss \taccuracy \tcohen_kappa_score \ttime\n0 \t 5.060787 \t3.422106 \t0.260000 \t0.293238 \t 02:05\n1 \t 3.159662 \t2.845312 \t0.280000 \t0.414904 \t 02:09\n2 \t 2.797441 \t3.018130 \t0.236000 \t0.374927 \t 02:12\n3 \t 2.302660 \t2.574300 \t0.292000 \t0.400289 \t 02:12\n```", "865986": "Congratulations!!!" } }