{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "以下是自动学习框架autogluon的实践。" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "%matplotlib inline\n", "from autogluon.vision import ImagePredictor, ImageDataset" ] }, { "cell_type": "code", "execution_count": 7, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "train # 50000 test # 10000\n" ] }, { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
imagelabel
0/home/fangga/dachuang/AI/data/cifar/train/airp...0
1/home/fangga/dachuang/AI/data/cifar/train/airp...0
2/home/fangga/dachuang/AI/data/cifar/train/airp...0
3/home/fangga/dachuang/AI/data/cifar/train/airp...0
4/home/fangga/dachuang/AI/data/cifar/train/airp...0
\n", "
" ], "text/plain": [ " image label\n", "0 /home/fangga/dachuang/AI/data/cifar/train/airp... 0\n", "1 /home/fangga/dachuang/AI/data/cifar/train/airp... 0\n", "2 /home/fangga/dachuang/AI/data/cifar/train/airp... 0\n", "3 /home/fangga/dachuang/AI/data/cifar/train/airp... 0\n", "4 /home/fangga/dachuang/AI/data/cifar/train/airp... 0" ] }, "execution_count": 7, "metadata": {}, "output_type": "execute_result" } ], "source": [ "train_data, _, test_data = ImageDataset.from_folders('data/cifar', train='train', test='test')\n", "print('train #', len(train_data), 'test #', len(test_data))\n", "train_data.head()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "`time_limit=auto` set to `time_limit=7200`.\n", "Reset labels to [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]\n", "Randomly split train_data into train[45000]/validation[5000] splits.\n", "The number of requested GPUs is greater than the number of available GPUs.Reduce the number to 1\n", "Starting fit without HPO\n", "modified configs( != ): {\n", "root.img_cls.model resnet101 != resnet50\n", "root.train.early_stop_baseline 0.0 != -inf\n", "root.train.early_stop_max_value 1.0 != inf\n", "root.train.batch_size 32 != 16\n", "root.train.early_stop_patience -1 != 10\n", "root.train.epochs 200 != 15\n", "root.misc.seed 42 != 435\n", "root.misc.num_workers 4 != 16\n", "}\n", "Saved config to /hy-tmp/c148b04f/.trial_0/config.yaml\n", "Downloading: \"https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet50_a1_0-14fe96d1.pth\" to /root/.cache/torch/hub/checkpoints/resnet50_a1_0-14fe96d1.pth\n", "Model resnet50 created, param count: 23528522\n", "AMP not enabled. Training in float32.\n", "Disable EMA as it is not supported for now.\n", "Start training from [Epoch 0]\n", "Epoch[0] Batch [49]\tSpeed: 134.945039 samples/sec\taccuracy=0.100000\tlr=0.000100\n", "Epoch[0] Batch [99]\tSpeed: 161.723904 samples/sec\taccuracy=0.105625\tlr=0.000100\n", "Epoch[0] Batch [149]\tSpeed: 156.695006 samples/sec\taccuracy=0.109167\tlr=0.000100\n", "Epoch[0] Batch [199]\tSpeed: 199.071506 samples/sec\taccuracy=0.111875\tlr=0.000100\n", "Epoch[0] Batch [249]\tSpeed: 159.495037 samples/sec\taccuracy=0.119750\tlr=0.000100\n", "Epoch[0] Batch [299]\tSpeed: 219.798152 samples/sec\taccuracy=0.121875\tlr=0.000100\n", "Epoch[0] Batch [349]\tSpeed: 164.758666 samples/sec\taccuracy=0.125357\tlr=0.000100\n", "Epoch[0] Batch [399]\tSpeed: 186.992106 samples/sec\taccuracy=0.128125\tlr=0.000100\n", "Epoch[0] Batch [449]\tSpeed: 187.883480 samples/sec\taccuracy=0.130833\tlr=0.000100\n", "Epoch[0] Batch [499]\tSpeed: 173.124262 samples/sec\taccuracy=0.134250\tlr=0.000100\n", "Epoch[0] Batch [549]\tSpeed: 162.516105 samples/sec\taccuracy=0.140227\tlr=0.000100\n", "Epoch[0] Batch [599]\tSpeed: 219.849967 samples/sec\taccuracy=0.142813\tlr=0.000100\n", "Epoch[0] Batch [649]\tSpeed: 177.265311 samples/sec\taccuracy=0.146538\tlr=0.000100\n", "Epoch[0] Batch [699]\tSpeed: 223.355135 samples/sec\taccuracy=0.150089\tlr=0.000100\n", "Epoch[0] Batch [749]\tSpeed: 218.368730 samples/sec\taccuracy=0.154583\tlr=0.000100\n", "Epoch[0] Batch [799]\tSpeed: 165.739097 samples/sec\taccuracy=0.157031\tlr=0.000100\n", "Epoch[0] Batch [849]\tSpeed: 196.920615 samples/sec\taccuracy=0.159191\tlr=0.000100\n", "Epoch[0] Batch [899]\tSpeed: 187.972577 samples/sec\taccuracy=0.162222\tlr=0.000100\n", "Epoch[0] Batch [949]\tSpeed: 189.072213 samples/sec\taccuracy=0.164737\tlr=0.000100\n", "Epoch[0] Batch [999]\tSpeed: 188.693028 samples/sec\taccuracy=0.166437\tlr=0.000100\n", "Epoch[0] Batch [1049]\tSpeed: 187.366028 samples/sec\taccuracy=0.169286\tlr=0.000100\n", "Epoch[0] Batch [1099]\tSpeed: 189.117758 samples/sec\taccuracy=0.172045\tlr=0.000100\n", "Epoch[0] Batch [1149]\tSpeed: 187.572868 samples/sec\taccuracy=0.174783\tlr=0.000100\n", "Epoch[0] Batch [1199]\tSpeed: 188.701591 samples/sec\taccuracy=0.176875\tlr=0.000100\n", "Epoch[0] Batch [1249]\tSpeed: 190.952527 samples/sec\taccuracy=0.179100\tlr=0.000100\n", "Epoch[0] Batch [1299]\tSpeed: 188.583277 samples/sec\taccuracy=0.181442\tlr=0.000100\n", "Epoch[0] Batch [1349]\tSpeed: 157.413679 samples/sec\taccuracy=0.184907\tlr=0.000100\n", "Epoch[0] Batch [1399]\tSpeed: 225.683978 samples/sec\taccuracy=0.186429\tlr=0.000100\n", "Epoch[0] Batch [1449]\tSpeed: 219.859374 samples/sec\taccuracy=0.187802\tlr=0.000100\n", "Epoch[0] Batch [1499]\tSpeed: 193.721784 samples/sec\taccuracy=0.190167\tlr=0.000100\n", "Epoch[0] Batch [1549]\tSpeed: 172.940549 samples/sec\taccuracy=0.192218\tlr=0.000100\n", "Epoch[0] Batch [1599]\tSpeed: 206.118104 samples/sec\taccuracy=0.193438\tlr=0.000100\n", "Epoch[0] Batch [1649]\tSpeed: 190.683920 samples/sec\taccuracy=0.195795\tlr=0.000100\n", "Epoch[0] Batch [1699]\tSpeed: 189.106194 samples/sec\taccuracy=0.197721\tlr=0.000100\n", "Epoch[0] Batch [1749]\tSpeed: 189.276668 samples/sec\taccuracy=0.199107\tlr=0.000100\n", "Epoch[0] Batch [1799]\tSpeed: 191.343279 samples/sec\taccuracy=0.201042\tlr=0.000100\n", "Epoch[0] Batch [1849]\tSpeed: 190.227833 samples/sec\taccuracy=0.203041\tlr=0.000100\n", "Epoch[0] Batch [1899]\tSpeed: 189.368662 samples/sec\taccuracy=0.204342\tlr=0.000100\n", "Epoch[0] Batch [1949]\tSpeed: 164.383187 samples/sec\taccuracy=0.206026\tlr=0.000100\n", "Epoch[0] Batch [1999]\tSpeed: 227.799242 samples/sec\taccuracy=0.208406\tlr=0.000100\n", "Epoch[0] Batch [2049]\tSpeed: 215.575263 samples/sec\taccuracy=0.210183\tlr=0.000100\n", "Epoch[0] Batch [2099]\tSpeed: 220.357634 samples/sec\taccuracy=0.211756\tlr=0.000100\n", "Epoch[0] Batch [2149]\tSpeed: 218.932833 samples/sec\taccuracy=0.214331\tlr=0.000100\n", "Epoch[0] Batch [2199]\tSpeed: 192.300029 samples/sec\taccuracy=0.216676\tlr=0.000100\n", "Epoch[0] Batch [2249]\tSpeed: 187.486873 samples/sec\taccuracy=0.218528\tlr=0.000100\n", "Epoch[0] Batch [2299]\tSpeed: 185.209352 samples/sec\taccuracy=0.220734\tlr=0.000100\n", "Epoch[0] Batch [2349]\tSpeed: 193.016321 samples/sec\taccuracy=0.222739\tlr=0.000100\n", "Epoch[0] Batch [2399]\tSpeed: 191.487319 samples/sec\taccuracy=0.224349\tlr=0.000100\n", "Epoch[0] Batch [2449]\tSpeed: 189.659470 samples/sec\taccuracy=0.226633\tlr=0.000100\n", "Epoch[0] Batch [2499]\tSpeed: 192.915847 samples/sec\taccuracy=0.228225\tlr=0.000100\n", "Epoch[0] Batch [2549]\tSpeed: 189.352044 samples/sec\taccuracy=0.230588\tlr=0.000100\n", "Epoch[0] Batch [2599]\tSpeed: 190.647192 samples/sec\taccuracy=0.232476\tlr=0.000100\n", "Epoch[0] Batch [2649]\tSpeed: 189.394967 samples/sec\taccuracy=0.234316\tlr=0.000100\n", "Epoch[0] Batch [2699]\tSpeed: 188.630888 samples/sec\taccuracy=0.235995\tlr=0.000100\n", "Epoch[0] Batch [2749]\tSpeed: 192.784592 samples/sec\taccuracy=0.237818\tlr=0.000100\n", "Epoch[0] Batch [2799]\tSpeed: 206.895014 samples/sec\taccuracy=0.239509\tlr=0.000100\n", "[Epoch 0] training: accuracy=0.240020\n", "[Epoch 0] speed: 188 samples/sec\ttime cost: 238.981761\n", "[Epoch 0] validation: top1=0.436200 top5=0.900800\n", "[Epoch 0] Current best top-1: 0.436200 vs previous -inf, saved to /hy-tmp/c148b04f/.trial_0/best_checkpoint.pkl\n", "Epoch[1] Batch [49]\tSpeed: 165.563224 samples/sec\taccuracy=0.255000\tlr=0.003400\n", "Epoch[1] Batch [99]\tSpeed: 187.925308 samples/sec\taccuracy=0.265000\tlr=0.003400\n", "Epoch[1] Batch [149]\tSpeed: 189.220641 samples/sec\taccuracy=0.271250\tlr=0.003400\n", "Epoch[1] Batch [199]\tSpeed: 189.928808 samples/sec\taccuracy=0.291250\tlr=0.003400\n", "Epoch[1] Batch [249]\tSpeed: 190.114739 samples/sec\taccuracy=0.297250\tlr=0.003400\n", "Epoch[1] Batch [299]\tSpeed: 189.908342 samples/sec\taccuracy=0.306667\tlr=0.003400\n", "Epoch[1] Batch [349]\tSpeed: 189.046392 samples/sec\taccuracy=0.314107\tlr=0.003400\n", "Epoch[1] Batch [399]\tSpeed: 208.726190 samples/sec\taccuracy=0.324688\tlr=0.003400\n", "Epoch[1] Batch [449]\tSpeed: 232.392286 samples/sec\taccuracy=0.331667\tlr=0.003400\n", "Epoch[1] Batch [499]\tSpeed: 200.159868 samples/sec\taccuracy=0.337625\tlr=0.003400\n", "Epoch[1] Batch [549]\tSpeed: 173.244942 samples/sec\taccuracy=0.342045\tlr=0.003400\n", "Epoch[1] Batch [599]\tSpeed: 187.723435 samples/sec\taccuracy=0.348854\tlr=0.003400\n", "Epoch[1] Batch [649]\tSpeed: 189.642458 samples/sec\taccuracy=0.354038\tlr=0.003400\n", "Epoch[1] Batch [699]\tSpeed: 191.134273 samples/sec\taccuracy=0.359286\tlr=0.003400\n", "Epoch[1] Batch [749]\tSpeed: 189.238602 samples/sec\taccuracy=0.366167\tlr=0.003400\n", "Epoch[1] Batch [799]\tSpeed: 185.074150 samples/sec\taccuracy=0.371719\tlr=0.003400\n", "Epoch[1] Batch [849]\tSpeed: 195.831585 samples/sec\taccuracy=0.377206\tlr=0.003400\n", "Epoch[1] Batch [899]\tSpeed: 187.227598 samples/sec\taccuracy=0.381319\tlr=0.003400\n", "Epoch[1] Batch [949]\tSpeed: 187.482159 samples/sec\taccuracy=0.386118\tlr=0.003400\n", "Epoch[1] Batch [999]\tSpeed: 189.886525 samples/sec\taccuracy=0.391250\tlr=0.003400\n", "Epoch[1] Batch [1049]\tSpeed: 189.753253 samples/sec\taccuracy=0.396012\tlr=0.003400\n", "Epoch[1] Batch [1099]\tSpeed: 187.584403 samples/sec\taccuracy=0.401023\tlr=0.003400\n", "Epoch[1] Batch [1149]\tSpeed: 186.309923 samples/sec\taccuracy=0.405272\tlr=0.003400\n", "Epoch[1] Batch [1199]\tSpeed: 186.529335 samples/sec\taccuracy=0.409948\tlr=0.003400\n", "Epoch[1] Batch [1249]\tSpeed: 191.928052 samples/sec\taccuracy=0.414650\tlr=0.003400\n", "Epoch[1] Batch [1299]\tSpeed: 185.336306 samples/sec\taccuracy=0.419567\tlr=0.003400\n", "Epoch[1] Batch [1349]\tSpeed: 186.830798 samples/sec\taccuracy=0.422778\tlr=0.003400\n", "Epoch[1] Batch [1399]\tSpeed: 187.798609 samples/sec\taccuracy=0.428214\tlr=0.003400\n", "Epoch[1] Batch [1449]\tSpeed: 187.567709 samples/sec\taccuracy=0.431897\tlr=0.003400\n", "Epoch[1] Batch [1499]\tSpeed: 186.440213 samples/sec\taccuracy=0.435917\tlr=0.003400\n", "Epoch[1] Batch [1549]\tSpeed: 190.899838 samples/sec\taccuracy=0.440403\tlr=0.003400\n", "Epoch[1] Batch [1599]\tSpeed: 189.622782 samples/sec\taccuracy=0.443555\tlr=0.003400\n", "Epoch[1] Batch [1649]\tSpeed: 190.321963 samples/sec\taccuracy=0.446818\tlr=0.003400\n", "Epoch[1] Batch [1699]\tSpeed: 186.500026 samples/sec\taccuracy=0.450515\tlr=0.003400\n", "Epoch[1] Batch [1749]\tSpeed: 185.102216 samples/sec\taccuracy=0.454643\tlr=0.003400\n", "Epoch[1] Batch [1799]\tSpeed: 187.592216 samples/sec\taccuracy=0.457917\tlr=0.003400\n", "Epoch[1] Batch [1849]\tSpeed: 195.538295 samples/sec\taccuracy=0.461588\tlr=0.003400\n", "Epoch[1] Batch [1899]\tSpeed: 190.991275 samples/sec\taccuracy=0.464638\tlr=0.003400\n", "Epoch[1] Batch [1949]\tSpeed: 190.035783 samples/sec\taccuracy=0.467596\tlr=0.003400\n", "Epoch[1] Batch [1999]\tSpeed: 193.008849 samples/sec\taccuracy=0.470344\tlr=0.003400\n", "Epoch[1] Batch [2049]\tSpeed: 190.844138 samples/sec\taccuracy=0.473750\tlr=0.003400\n", "Epoch[1] Batch [2099]\tSpeed: 190.005394 samples/sec\taccuracy=0.476756\tlr=0.003400\n", "Epoch[1] Batch [2149]\tSpeed: 189.376816 samples/sec\taccuracy=0.479826\tlr=0.003400\n", "Epoch[1] Batch [2199]\tSpeed: 190.164700 samples/sec\taccuracy=0.482585\tlr=0.003400\n", "Epoch[1] Batch [2249]\tSpeed: 204.398377 samples/sec\taccuracy=0.485583\tlr=0.003400\n", "Epoch[1] Batch [2299]\tSpeed: 195.676808 samples/sec\taccuracy=0.488804\tlr=0.003400\n", "Epoch[1] Batch [2349]\tSpeed: 192.037830 samples/sec\taccuracy=0.491809\tlr=0.003400\n", "Epoch[1] Batch [2399]\tSpeed: 191.625906 samples/sec\taccuracy=0.494948\tlr=0.003400\n", "Epoch[1] Batch [2449]\tSpeed: 204.301230 samples/sec\taccuracy=0.497577\tlr=0.003400\n", "Epoch[1] Batch [2499]\tSpeed: 204.661094 samples/sec\taccuracy=0.500150\tlr=0.003400\n", "Epoch[1] Batch [2549]\tSpeed: 206.414960 samples/sec\taccuracy=0.502549\tlr=0.003400\n", "Epoch[1] Batch [2599]\tSpeed: 206.072255 samples/sec\taccuracy=0.505625\tlr=0.003400\n", "Epoch[1] Batch [2649]\tSpeed: 206.889337 samples/sec\taccuracy=0.507901\tlr=0.003400\n", "Epoch[1] Batch [2699]\tSpeed: 194.202937 samples/sec\taccuracy=0.509606\tlr=0.003400\n", "Epoch[1] Batch [2749]\tSpeed: 207.551373 samples/sec\taccuracy=0.511932\tlr=0.003400\n", "Epoch[1] Batch [2799]\tSpeed: 275.113608 samples/sec\taccuracy=0.514442\tlr=0.003400\n", "[Epoch 1] training: accuracy=0.515003\n", "[Epoch 1] speed: 192 samples/sec\ttime cost: 233.134034\n", "[Epoch 1] validation: top1=0.876000 top5=0.996000\n", "[Epoch 1] Current best top-1: 0.876000 vs previous 0.436200, saved to /hy-tmp/c148b04f/.trial_0/best_checkpoint.pkl\n", "Epoch[2] Batch [49]\tSpeed: 169.880814 samples/sec\taccuracy=0.626250\tlr=0.006700\n", "Epoch[2] Batch [99]\tSpeed: 190.112337 samples/sec\taccuracy=0.637500\tlr=0.006700\n", "Epoch[2] Batch [149]\tSpeed: 188.506732 samples/sec\taccuracy=0.630417\tlr=0.006700\n", "Epoch[2] Batch [199]\tSpeed: 189.962388 samples/sec\taccuracy=0.630625\tlr=0.006700\n", "Epoch[2] Batch [249]\tSpeed: 191.852629 samples/sec\taccuracy=0.637500\tlr=0.006700\n", "Epoch[2] Batch [299]\tSpeed: 191.373465 samples/sec\taccuracy=0.643542\tlr=0.006700\n", "Epoch[2] Batch [349]\tSpeed: 188.455415 samples/sec\taccuracy=0.643036\tlr=0.006700\n", "Epoch[2] Batch [399]\tSpeed: 188.676359 samples/sec\taccuracy=0.646250\tlr=0.006700\n", "Epoch[2] Batch [449]\tSpeed: 191.834564 samples/sec\taccuracy=0.646944\tlr=0.006700\n", "Epoch[2] Batch [499]\tSpeed: 190.232459 samples/sec\taccuracy=0.650125\tlr=0.006700\n", "Epoch[2] Batch [549]\tSpeed: 189.380515 samples/sec\taccuracy=0.651364\tlr=0.006700\n", "Epoch[2] Batch [599]\tSpeed: 189.160596 samples/sec\taccuracy=0.652813\tlr=0.006700\n", "Epoch[2] Batch [649]\tSpeed: 188.919454 samples/sec\taccuracy=0.653269\tlr=0.006700\n", "Epoch[2] Batch [699]\tSpeed: 187.861222 samples/sec\taccuracy=0.654464\tlr=0.006700\n", "Epoch[2] Batch [749]\tSpeed: 194.734178 samples/sec\taccuracy=0.655833\tlr=0.006700\n", "Epoch[2] Batch [799]\tSpeed: 188.881787 samples/sec\taccuracy=0.656641\tlr=0.006700\n", "Epoch[2] Batch [849]\tSpeed: 193.530420 samples/sec\taccuracy=0.656250\tlr=0.006700\n", "Epoch[2] Batch [899]\tSpeed: 194.345181 samples/sec\taccuracy=0.655694\tlr=0.006700\n", "Epoch[2] Batch [949]\tSpeed: 193.999439 samples/sec\taccuracy=0.658158\tlr=0.006700\n", "Epoch[2] Batch [999]\tSpeed: 183.166158 samples/sec\taccuracy=0.659625\tlr=0.006700\n", "Epoch[2] Batch [1049]\tSpeed: 198.096400 samples/sec\taccuracy=0.659643\tlr=0.006700\n", "Epoch[2] Batch [1099]\tSpeed: 187.453210 samples/sec\taccuracy=0.661705\tlr=0.006700\n", "Epoch[2] Batch [1149]\tSpeed: 191.404872 samples/sec\taccuracy=0.663478\tlr=0.006700\n", "Epoch[2] Batch [1199]\tSpeed: 194.395172 samples/sec\taccuracy=0.664531\tlr=0.006700\n", "Epoch[2] Batch [1249]\tSpeed: 191.883316 samples/sec\taccuracy=0.664800\tlr=0.006700\n", "Epoch[2] Batch [1299]\tSpeed: 192.366275 samples/sec\taccuracy=0.665481\tlr=0.006700\n", "Epoch[2] Batch [1349]\tSpeed: 191.402797 samples/sec\taccuracy=0.667037\tlr=0.006700\n", "Epoch[2] Batch [1399]\tSpeed: 191.662782 samples/sec\taccuracy=0.668393\tlr=0.006700\n", "Epoch[2] Batch [1449]\tSpeed: 191.722477 samples/sec\taccuracy=0.668836\tlr=0.006700\n", "Epoch[2] Batch [1499]\tSpeed: 190.396803 samples/sec\taccuracy=0.668792\tlr=0.006700\n", "Epoch[2] Batch [1549]\tSpeed: 192.400402 samples/sec\taccuracy=0.670242\tlr=0.006700\n", "Epoch[2] Batch [1599]\tSpeed: 189.492469 samples/sec\taccuracy=0.671250\tlr=0.006700\n", "Epoch[2] Batch [1649]\tSpeed: 190.082785 samples/sec\taccuracy=0.672424\tlr=0.006700\n", "Epoch[2] Batch [1699]\tSpeed: 191.251766 samples/sec\taccuracy=0.672500\tlr=0.006700\n", "Epoch[2] Batch [1749]\tSpeed: 191.812369 samples/sec\taccuracy=0.673250\tlr=0.006700\n", "Epoch[2] Batch [1799]\tSpeed: 188.804627 samples/sec\taccuracy=0.673646\tlr=0.006700\n", "Epoch[2] Batch [1849]\tSpeed: 187.653840 samples/sec\taccuracy=0.674392\tlr=0.006700\n", "Epoch[2] Batch [1899]\tSpeed: 187.164091 samples/sec\taccuracy=0.675329\tlr=0.006700\n", "Epoch[2] Batch [1949]\tSpeed: 189.955796 samples/sec\taccuracy=0.675769\tlr=0.006700\n", "Epoch[2] Batch [1999]\tSpeed: 189.464842 samples/sec\taccuracy=0.676875\tlr=0.006700\n", "Epoch[2] Batch [2049]\tSpeed: 189.855905 samples/sec\taccuracy=0.677805\tlr=0.006700\n", "Epoch[2] Batch [2099]\tSpeed: 190.015541 samples/sec\taccuracy=0.678095\tlr=0.006700\n", "Epoch[2] Batch [2149]\tSpeed: 187.923076 samples/sec\taccuracy=0.678895\tlr=0.006700\n", "Epoch[2] Batch [2199]\tSpeed: 186.848474 samples/sec\taccuracy=0.679375\tlr=0.006700\n", "Epoch[2] Batch [2249]\tSpeed: 188.653817 samples/sec\taccuracy=0.680250\tlr=0.006700\n", "Epoch[2] Batch [2299]\tSpeed: 191.376194 samples/sec\taccuracy=0.681766\tlr=0.006700\n", "Epoch[2] Batch [2349]\tSpeed: 192.655319 samples/sec\taccuracy=0.682314\tlr=0.006700\n", "Epoch[2] Batch [2399]\tSpeed: 198.361918 samples/sec\taccuracy=0.683255\tlr=0.006700\n", "Epoch[2] Batch [2449]\tSpeed: 190.241530 samples/sec\taccuracy=0.684388\tlr=0.006700\n", "Epoch[2] Batch [2499]\tSpeed: 187.874949 samples/sec\taccuracy=0.685225\tlr=0.006700\n", "Epoch[2] Batch [2549]\tSpeed: 191.809551 samples/sec\taccuracy=0.685613\tlr=0.006700\n", "Epoch[2] Batch [2599]\tSpeed: 202.764288 samples/sec\taccuracy=0.685937\tlr=0.006700\n", "Epoch[2] Batch [2649]\tSpeed: 203.165155 samples/sec\taccuracy=0.686981\tlr=0.006700\n", "Epoch[2] Batch [2699]\tSpeed: 202.297702 samples/sec\taccuracy=0.687269\tlr=0.006700\n", "Epoch[2] Batch [2749]\tSpeed: 203.503868 samples/sec\taccuracy=0.688068\tlr=0.006700\n", "Epoch[2] Batch [2799]\tSpeed: 205.877489 samples/sec\taccuracy=0.688750\tlr=0.006700\n", "[Epoch 2] training: accuracy=0.688989\n", "[Epoch 2] speed: 191 samples/sec\ttime cost: 235.289050\n", "[Epoch 2] validation: top1=0.920200 top5=0.998400\n", "[Epoch 2] Current best top-1: 0.920200 vs previous 0.876000, saved to /hy-tmp/c148b04f/.trial_0/best_checkpoint.pkl\n", "Epoch[3] Batch [49]\tSpeed: 166.967726 samples/sec\taccuracy=0.746250\tlr=0.010000\n", "Epoch[3] Batch [99]\tSpeed: 265.275746 samples/sec\taccuracy=0.727500\tlr=0.010000\n", "Epoch[3] Batch [149]\tSpeed: 183.765048 samples/sec\taccuracy=0.728333\tlr=0.010000\n", "Epoch[3] Batch [199]\tSpeed: 210.901680 samples/sec\taccuracy=0.729688\tlr=0.010000\n", "Epoch[3] Batch [249]\tSpeed: 190.549797 samples/sec\taccuracy=0.729000\tlr=0.010000\n", "Epoch[3] Batch [299]\tSpeed: 184.439790 samples/sec\taccuracy=0.722083\tlr=0.010000\n", "Epoch[3] Batch [349]\tSpeed: 188.357412 samples/sec\taccuracy=0.723036\tlr=0.010000\n", "Epoch[3] Batch [399]\tSpeed: 188.831297 samples/sec\taccuracy=0.721562\tlr=0.010000\n", "Epoch[3] Batch [449]\tSpeed: 189.627711 samples/sec\taccuracy=0.720694\tlr=0.010000\n", "Epoch[3] Batch [499]\tSpeed: 188.741172 samples/sec\taccuracy=0.721625\tlr=0.010000\n", "Epoch[3] Batch [549]\tSpeed: 188.642055 samples/sec\taccuracy=0.723977\tlr=0.010000\n", "Epoch[3] Batch [599]\tSpeed: 190.032479 samples/sec\taccuracy=0.724063\tlr=0.010000\n", "Epoch[3] Batch [649]\tSpeed: 192.128909 samples/sec\taccuracy=0.724712\tlr=0.010000\n", "Epoch[3] Batch [699]\tSpeed: 190.442880 samples/sec\taccuracy=0.724554\tlr=0.010000\n", "Epoch[3] Batch [749]\tSpeed: 191.439281 samples/sec\taccuracy=0.725583\tlr=0.010000\n", "Epoch[3] Batch [799]\tSpeed: 166.776877 samples/sec\taccuracy=0.725391\tlr=0.010000\n", "Epoch[3] Batch [849]\tSpeed: 192.810071 samples/sec\taccuracy=0.726176\tlr=0.010000\n", "Epoch[3] Batch [899]\tSpeed: 240.275856 samples/sec\taccuracy=0.727083\tlr=0.010000\n", "Epoch[3] Batch [949]\tSpeed: 238.989689 samples/sec\taccuracy=0.728026\tlr=0.010000\n", "Epoch[3] Batch [999]\tSpeed: 241.769412 samples/sec\taccuracy=0.727812\tlr=0.010000\n", "Epoch[3] Batch [1049]\tSpeed: 199.274535 samples/sec\taccuracy=0.727857\tlr=0.010000\n", "Epoch[3] Batch [1099]\tSpeed: 200.123243 samples/sec\taccuracy=0.728352\tlr=0.010000\n", "Epoch[3] Batch [1149]\tSpeed: 203.702228 samples/sec\taccuracy=0.728533\tlr=0.010000\n", "Epoch[3] Batch [1199]\tSpeed: 190.264173 samples/sec\taccuracy=0.728281\tlr=0.010000\n", "Epoch[3] Batch [1249]\tSpeed: 190.388193 samples/sec\taccuracy=0.729550\tlr=0.010000\n", "Epoch[3] Batch [1299]\tSpeed: 191.027657 samples/sec\taccuracy=0.730721\tlr=0.010000\n", "Epoch[3] Batch [1349]\tSpeed: 189.755989 samples/sec\taccuracy=0.730602\tlr=0.010000\n", "Epoch[3] Batch [1399]\tSpeed: 199.677143 samples/sec\taccuracy=0.731161\tlr=0.010000\n", "Epoch[3] Batch [1449]\tSpeed: 193.199271 samples/sec\taccuracy=0.732069\tlr=0.010000\n", "Epoch[3] Batch [1499]\tSpeed: 188.614103 samples/sec\taccuracy=0.732667\tlr=0.010000\n", "Epoch[3] Batch [1549]\tSpeed: 183.159359 samples/sec\taccuracy=0.732742\tlr=0.010000\n", "Epoch[3] Batch [1599]\tSpeed: 157.483348 samples/sec\taccuracy=0.733125\tlr=0.010000\n", "Epoch[3] Batch [1649]\tSpeed: 189.593082 samples/sec\taccuracy=0.733333\tlr=0.010000\n", "Epoch[3] Batch [1699]\tSpeed: 189.831533 samples/sec\taccuracy=0.733272\tlr=0.010000\n", "Epoch[3] Batch [1749]\tSpeed: 191.365497 samples/sec\taccuracy=0.733964\tlr=0.010000\n", "Epoch[3] Batch [1799]\tSpeed: 189.916586 samples/sec\taccuracy=0.734618\tlr=0.010000\n", "Epoch[3] Batch [1849]\tSpeed: 192.270343 samples/sec\taccuracy=0.735338\tlr=0.010000\n", "Epoch[3] Batch [1899]\tSpeed: 191.585161 samples/sec\taccuracy=0.735691\tlr=0.010000\n", "Epoch[3] Batch [1949]\tSpeed: 190.332813 samples/sec\taccuracy=0.735673\tlr=0.010000\n", "Epoch[3] Batch [1999]\tSpeed: 190.622173 samples/sec\taccuracy=0.736094\tlr=0.010000\n", "Epoch[3] Batch [2049]\tSpeed: 189.112333 samples/sec\taccuracy=0.735762\tlr=0.010000\n", "Epoch[3] Batch [2099]\tSpeed: 196.736759 samples/sec\taccuracy=0.736458\tlr=0.010000\n", "Epoch[3] Batch [2149]\tSpeed: 188.769756 samples/sec\taccuracy=0.736570\tlr=0.010000\n", "Epoch[3] Batch [2199]\tSpeed: 187.500462 samples/sec\taccuracy=0.736648\tlr=0.010000\n", "Epoch[3] Batch [2249]\tSpeed: 188.085961 samples/sec\taccuracy=0.736611\tlr=0.010000\n", "Epoch[3] Batch [2299]\tSpeed: 188.472743 samples/sec\taccuracy=0.737446\tlr=0.010000\n", "Epoch[3] Batch [2349]\tSpeed: 187.924739 samples/sec\taccuracy=0.737261\tlr=0.010000\n", "Epoch[3] Batch [2399]\tSpeed: 183.355428 samples/sec\taccuracy=0.737917\tlr=0.010000\n", "Epoch[3] Batch [2449]\tSpeed: 183.187868 samples/sec\taccuracy=0.738393\tlr=0.010000\n", "Epoch[3] Batch [2499]\tSpeed: 185.084644 samples/sec\taccuracy=0.738500\tlr=0.010000\n", "Epoch[3] Batch [2549]\tSpeed: 187.033319 samples/sec\taccuracy=0.738725\tlr=0.010000\n", "Epoch[3] Batch [2599]\tSpeed: 189.504145 samples/sec\taccuracy=0.739087\tlr=0.010000\n", "Epoch[3] Batch [2649]\tSpeed: 186.224070 samples/sec\taccuracy=0.739458\tlr=0.010000\n", "Epoch[3] Batch [2699]\tSpeed: 188.133786 samples/sec\taccuracy=0.739630\tlr=0.010000\n", "Epoch[3] Batch [2749]\tSpeed: 191.439915 samples/sec\taccuracy=0.739636\tlr=0.010000\n", "Epoch[3] Batch [2799]\tSpeed: 194.092343 samples/sec\taccuracy=0.740246\tlr=0.010000\n", "[Epoch 3] training: accuracy=0.740287\n", "[Epoch 3] speed: 191 samples/sec\ttime cost: 234.526291\n", "[Epoch 3] validation: top1=0.930400 top5=0.998200\n", "[Epoch 3] Current best top-1: 0.930400 vs previous 0.920200, saved to /hy-tmp/c148b04f/.trial_0/best_checkpoint.pkl\n", "Epoch[4] Batch [49]\tSpeed: 166.350594 samples/sec\taccuracy=0.761250\tlr=0.010000\n", "Epoch[4] Batch [99]\tSpeed: 188.201404 samples/sec\taccuracy=0.760000\tlr=0.010000\n", "Epoch[4] Batch [149]\tSpeed: 188.357655 samples/sec\taccuracy=0.765000\tlr=0.010000\n", "Epoch[4] Batch [199]\tSpeed: 182.544312 samples/sec\taccuracy=0.769062\tlr=0.010000\n", "Epoch[4] Batch [249]\tSpeed: 193.545189 samples/sec\taccuracy=0.762250\tlr=0.010000\n", "Epoch[4] Batch [299]\tSpeed: 193.312958 samples/sec\taccuracy=0.761458\tlr=0.010000\n", "Epoch[4] Batch [349]\tSpeed: 167.500349 samples/sec\taccuracy=0.762857\tlr=0.010000\n", "Epoch[4] Batch [399]\tSpeed: 226.087267 samples/sec\taccuracy=0.763750\tlr=0.010000\n", "Epoch[4] Batch [449]\tSpeed: 219.916335 samples/sec\taccuracy=0.763472\tlr=0.010000\n", "Epoch[4] Batch [499]\tSpeed: 215.597342 samples/sec\taccuracy=0.765125\tlr=0.010000\n", "Epoch[4] Batch [549]\tSpeed: 203.264339 samples/sec\taccuracy=0.762159\tlr=0.010000\n", "Epoch[4] Batch [599]\tSpeed: 206.603773 samples/sec\taccuracy=0.763021\tlr=0.010000\n", "Epoch[4] Batch [649]\tSpeed: 204.492724 samples/sec\taccuracy=0.761346\tlr=0.010000\n", "Epoch[4] Batch [699]\tSpeed: 201.202317 samples/sec\taccuracy=0.761607\tlr=0.010000\n", "Epoch[4] Batch [749]\tSpeed: 204.774878 samples/sec\taccuracy=0.762000\tlr=0.010000\n", "Epoch[4] Batch [799]\tSpeed: 201.119973 samples/sec\taccuracy=0.760078\tlr=0.010000\n", "Epoch[4] Batch [849]\tSpeed: 190.312237 samples/sec\taccuracy=0.759779\tlr=0.010000\n", "Epoch[4] Batch [899]\tSpeed: 204.424353 samples/sec\taccuracy=0.759722\tlr=0.010000\n", "Epoch[4] Batch [949]\tSpeed: 219.947674 samples/sec\taccuracy=0.759934\tlr=0.010000\n", "Epoch[4] Batch [999]\tSpeed: 195.833814 samples/sec\taccuracy=0.759188\tlr=0.010000\n", "Epoch[4] Batch [1049]\tSpeed: 189.678607 samples/sec\taccuracy=0.760417\tlr=0.010000\n", "Epoch[4] Batch [1099]\tSpeed: 187.663328 samples/sec\taccuracy=0.760341\tlr=0.010000\n", "Epoch[4] Batch [1149]\tSpeed: 191.511100 samples/sec\taccuracy=0.760163\tlr=0.010000\n", "Epoch[4] Batch [1199]\tSpeed: 187.694946 samples/sec\taccuracy=0.760677\tlr=0.010000\n", "Epoch[4] Batch [1249]\tSpeed: 189.702778 samples/sec\taccuracy=0.760950\tlr=0.010000\n", "Epoch[4] Batch [1299]\tSpeed: 190.255877 samples/sec\taccuracy=0.761827\tlr=0.010000\n", "Epoch[4] Batch [1349]\tSpeed: 192.064717 samples/sec\taccuracy=0.762130\tlr=0.010000\n", "Epoch[4] Batch [1399]\tSpeed: 191.463281 samples/sec\taccuracy=0.762768\tlr=0.010000\n", "Epoch[4] Batch [1449]\tSpeed: 188.943102 samples/sec\taccuracy=0.762974\tlr=0.010000\n", "Epoch[4] Batch [1499]\tSpeed: 185.002028 samples/sec\taccuracy=0.764208\tlr=0.010000\n", "Epoch[4] Batch [1549]\tSpeed: 189.909599 samples/sec\taccuracy=0.764556\tlr=0.010000\n", "Epoch[4] Batch [1599]\tSpeed: 197.214720 samples/sec\taccuracy=0.764570\tlr=0.010000\n", "Epoch[4] Batch [1649]\tSpeed: 192.922668 samples/sec\taccuracy=0.764205\tlr=0.010000\n", "Epoch[4] Batch [1699]\tSpeed: 187.978684 samples/sec\taccuracy=0.764412\tlr=0.010000\n", "Epoch[4] Batch [1749]\tSpeed: 188.860843 samples/sec\taccuracy=0.765143\tlr=0.010000\n", "Epoch[4] Batch [1799]\tSpeed: 180.906750 samples/sec\taccuracy=0.765799\tlr=0.010000\n", "Epoch[4] Batch [1849]\tSpeed: 187.243750 samples/sec\taccuracy=0.765980\tlr=0.010000\n", "Epoch[4] Batch [1899]\tSpeed: 187.015097 samples/sec\taccuracy=0.766414\tlr=0.010000\n", "Epoch[4] Batch [1949]\tSpeed: 182.048000 samples/sec\taccuracy=0.767276\tlr=0.010000\n", "Epoch[4] Batch [1999]\tSpeed: 187.361812 samples/sec\taccuracy=0.767469\tlr=0.010000\n", "Epoch[4] Batch [2049]\tSpeed: 186.958318 samples/sec\taccuracy=0.767713\tlr=0.010000\n", "Epoch[4] Batch [2099]\tSpeed: 190.515522 samples/sec\taccuracy=0.768304\tlr=0.010000\n", "Epoch[4] Batch [2149]\tSpeed: 191.107330 samples/sec\taccuracy=0.768343\tlr=0.010000\n", "Epoch[4] Batch [2199]\tSpeed: 189.260248 samples/sec\taccuracy=0.768920\tlr=0.010000\n", "Epoch[4] Batch [2249]\tSpeed: 188.018626 samples/sec\taccuracy=0.768917\tlr=0.010000\n", "Epoch[4] Batch [2299]\tSpeed: 203.424217 samples/sec\taccuracy=0.769158\tlr=0.010000\n", "Epoch[4] Batch [2349]\tSpeed: 203.679947 samples/sec\taccuracy=0.769734\tlr=0.010000\n", "Epoch[4] Batch [2399]\tSpeed: 203.892663 samples/sec\taccuracy=0.769557\tlr=0.010000\n", "Epoch[4] Batch [2449]\tSpeed: 203.680429 samples/sec\taccuracy=0.769439\tlr=0.010000\n", "Epoch[4] Batch [2499]\tSpeed: 204.333838 samples/sec\taccuracy=0.769275\tlr=0.010000\n", "Epoch[4] Batch [2549]\tSpeed: 202.989891 samples/sec\taccuracy=0.769289\tlr=0.010000\n", "Epoch[4] Batch [2599]\tSpeed: 165.550955 samples/sec\taccuracy=0.769183\tlr=0.010000\n", "Epoch[4] Batch [2649]\tSpeed: 221.085518 samples/sec\taccuracy=0.769599\tlr=0.010000\n", "Epoch[4] Batch [2699]\tSpeed: 212.336857 samples/sec\taccuracy=0.769213\tlr=0.010000\n", "Epoch[4] Batch [2749]\tSpeed: 213.847809 samples/sec\taccuracy=0.769386\tlr=0.010000\n", "Epoch[4] Batch [2799]\tSpeed: 225.311199 samples/sec\taccuracy=0.769375\tlr=0.010000\n", "[Epoch 4] training: accuracy=0.769403\n", "[Epoch 4] speed: 194 samples/sec\ttime cost: 230.936548\n", "[Epoch 4] validation: top1=0.943400 top5=0.999400\n", "[Epoch 4] Current best top-1: 0.943400 vs previous 0.930400, saved to /hy-tmp/c148b04f/.trial_0/best_checkpoint.pkl\n", "Epoch[5] Batch [49]\tSpeed: 167.101781 samples/sec\taccuracy=0.782500\tlr=0.010000\n", "Epoch[5] Batch [99]\tSpeed: 192.832985 samples/sec\taccuracy=0.780000\tlr=0.010000\n", "Epoch[5] Batch [149]\tSpeed: 195.030750 samples/sec\taccuracy=0.787083\tlr=0.010000\n", "Epoch[5] Batch [199]\tSpeed: 190.191970 samples/sec\taccuracy=0.783125\tlr=0.010000\n", "Epoch[5] Batch [249]\tSpeed: 187.959952 samples/sec\taccuracy=0.782750\tlr=0.010000\n", "Epoch[5] Batch [299]\tSpeed: 191.209578 samples/sec\taccuracy=0.785417\tlr=0.010000\n", "Epoch[5] Batch [349]\tSpeed: 198.091746 samples/sec\taccuracy=0.785893\tlr=0.010000\n", "Epoch[5] Batch [399]\tSpeed: 189.087524 samples/sec\taccuracy=0.786563\tlr=0.010000\n", "Epoch[5] Batch [449]\tSpeed: 193.353851 samples/sec\taccuracy=0.785417\tlr=0.010000\n", "Epoch[5] Batch [499]\tSpeed: 190.990394 samples/sec\taccuracy=0.787375\tlr=0.010000\n", "Epoch[5] Batch [549]\tSpeed: 188.380697 samples/sec\taccuracy=0.786023\tlr=0.010000\n", "Epoch[5] Batch [599]\tSpeed: 188.925241 samples/sec\taccuracy=0.787396\tlr=0.010000\n", "Epoch[5] Batch [649]\tSpeed: 189.083262 samples/sec\taccuracy=0.786827\tlr=0.010000\n", "Epoch[5] Batch [699]\tSpeed: 192.584153 samples/sec\taccuracy=0.787857\tlr=0.010000\n", "Epoch[5] Batch [749]\tSpeed: 194.823218 samples/sec\taccuracy=0.786417\tlr=0.010000\n", "Epoch[5] Batch [799]\tSpeed: 191.438856 samples/sec\taccuracy=0.787500\tlr=0.010000\n", "Epoch[5] Batch [849]\tSpeed: 186.999985 samples/sec\taccuracy=0.786471\tlr=0.010000\n", "Epoch[5] Batch [899]\tSpeed: 182.037748 samples/sec\taccuracy=0.786111\tlr=0.010000\n", "Epoch[5] Batch [949]\tSpeed: 187.155541 samples/sec\taccuracy=0.787303\tlr=0.010000\n", "Epoch[5] Batch [999]\tSpeed: 190.177450 samples/sec\taccuracy=0.786500\tlr=0.010000\n", "Epoch[5] Batch [1049]\tSpeed: 187.189338 samples/sec\taccuracy=0.786250\tlr=0.010000\n", "Epoch[5] Batch [1099]\tSpeed: 187.960752 samples/sec\taccuracy=0.785852\tlr=0.010000\n", "Epoch[5] Batch [1149]\tSpeed: 190.878368 samples/sec\taccuracy=0.785163\tlr=0.010000\n", "Epoch[5] Batch [1199]\tSpeed: 186.104763 samples/sec\taccuracy=0.785573\tlr=0.010000\n", "Epoch[5] Batch [1249]\tSpeed: 191.702509 samples/sec\taccuracy=0.785050\tlr=0.010000\n", "Epoch[5] Batch [1299]\tSpeed: 189.737748 samples/sec\taccuracy=0.785721\tlr=0.010000\n", "Epoch[5] Batch [1349]\tSpeed: 189.163646 samples/sec\taccuracy=0.785556\tlr=0.010000\n", "Epoch[5] Batch [1399]\tSpeed: 203.149657 samples/sec\taccuracy=0.785179\tlr=0.010000\n", "Epoch[5] Batch [1449]\tSpeed: 202.480410 samples/sec\taccuracy=0.785129\tlr=0.010000\n", "Epoch[5] Batch [1499]\tSpeed: 188.572446 samples/sec\taccuracy=0.785917\tlr=0.010000\n", "Epoch[5] Batch [1549]\tSpeed: 191.171385 samples/sec\taccuracy=0.785887\tlr=0.010000\n", "Epoch[5] Batch [1599]\tSpeed: 191.107319 samples/sec\taccuracy=0.785859\tlr=0.010000\n", "Epoch[5] Batch [1649]\tSpeed: 202.388740 samples/sec\taccuracy=0.785720\tlr=0.010000\n", "Epoch[5] Batch [1699]\tSpeed: 193.723719 samples/sec\taccuracy=0.785699\tlr=0.010000\n", "Epoch[5] Batch [1749]\tSpeed: 192.504216 samples/sec\taccuracy=0.785643\tlr=0.010000\n", "Epoch[5] Batch [1799]\tSpeed: 189.614831 samples/sec\taccuracy=0.785521\tlr=0.010000\n", "Epoch[5] Batch [1849]\tSpeed: 189.687829 samples/sec\taccuracy=0.784831\tlr=0.010000\n", "Epoch[5] Batch [1899]\tSpeed: 190.839210 samples/sec\taccuracy=0.785000\tlr=0.010000\n", "Epoch[5] Batch [1949]\tSpeed: 190.537288 samples/sec\taccuracy=0.785064\tlr=0.010000\n", "Epoch[5] Batch [1999]\tSpeed: 190.559309 samples/sec\taccuracy=0.785281\tlr=0.010000\n", "Epoch[5] Batch [2049]\tSpeed: 190.793646 samples/sec\taccuracy=0.785366\tlr=0.010000\n", "Epoch[5] Batch [2099]\tSpeed: 189.302659 samples/sec\taccuracy=0.784940\tlr=0.010000\n", "Epoch[5] Batch [2149]\tSpeed: 189.398088 samples/sec\taccuracy=0.784942\tlr=0.010000\n", "Epoch[5] Batch [2199]\tSpeed: 189.994356 samples/sec\taccuracy=0.784375\tlr=0.010000\n", "Epoch[5] Batch [2249]\tSpeed: 189.561603 samples/sec\taccuracy=0.784861\tlr=0.010000\n", "Epoch[5] Batch [2299]\tSpeed: 187.811759 samples/sec\taccuracy=0.784891\tlr=0.010000\n", "Epoch[5] Batch [2349]\tSpeed: 196.868913 samples/sec\taccuracy=0.784601\tlr=0.010000\n", "Epoch[5] Batch [2399]\tSpeed: 184.917386 samples/sec\taccuracy=0.784453\tlr=0.010000\n", "Epoch[5] Batch [2449]\tSpeed: 188.754592 samples/sec\taccuracy=0.784821\tlr=0.010000\n", "Epoch[5] Batch [2499]\tSpeed: 187.715548 samples/sec\taccuracy=0.785225\tlr=0.010000\n", "Epoch[5] Batch [2549]\tSpeed: 183.595906 samples/sec\taccuracy=0.784828\tlr=0.010000\n", "Epoch[5] Batch [2599]\tSpeed: 186.794926 samples/sec\taccuracy=0.784880\tlr=0.010000\n", "Epoch[5] Batch [2649]\tSpeed: 191.660658 samples/sec\taccuracy=0.784953\tlr=0.010000\n", "Epoch[5] Batch [2699]\tSpeed: 190.291526 samples/sec\taccuracy=0.785069\tlr=0.010000\n", "Epoch[5] Batch [2749]\tSpeed: 195.245314 samples/sec\taccuracy=0.785341\tlr=0.010000\n", "Epoch[5] Batch [2799]\tSpeed: 193.504204 samples/sec\taccuracy=0.785469\tlr=0.010000\n", "[Epoch 5] training: accuracy=0.785384\n", "[Epoch 5] speed: 190 samples/sec\ttime cost: 236.528347\n", "[Epoch 5] validation: top1=0.946200 top5=0.998400\n", "[Epoch 5] Current best top-1: 0.946200 vs previous 0.943400, saved to /hy-tmp/c148b04f/.trial_0/best_checkpoint.pkl\n", "Epoch[6] Batch [49]\tSpeed: 170.107075 samples/sec\taccuracy=0.788750\tlr=0.010000\n", "Epoch[6] Batch [99]\tSpeed: 188.285360 samples/sec\taccuracy=0.791250\tlr=0.010000\n", "Epoch[6] Batch [149]\tSpeed: 179.372263 samples/sec\taccuracy=0.790833\tlr=0.010000\n", "Epoch[6] Batch [199]\tSpeed: 199.625016 samples/sec\taccuracy=0.789687\tlr=0.010000\n", "Epoch[6] Batch [249]\tSpeed: 203.220083 samples/sec\taccuracy=0.789250\tlr=0.010000\n", "Epoch[6] Batch [299]\tSpeed: 204.365947 samples/sec\taccuracy=0.787500\tlr=0.010000\n", "Epoch[6] Batch [349]\tSpeed: 185.470240 samples/sec\taccuracy=0.792679\tlr=0.010000\n", "Epoch[6] Batch [399]\tSpeed: 191.657856 samples/sec\taccuracy=0.794063\tlr=0.010000\n", "Epoch[6] Batch [449]\tSpeed: 198.002615 samples/sec\taccuracy=0.796111\tlr=0.010000\n", "Epoch[6] Batch [499]\tSpeed: 189.700751 samples/sec\taccuracy=0.796750\tlr=0.010000\n", "Epoch[6] Batch [549]\tSpeed: 190.702863 samples/sec\taccuracy=0.798750\tlr=0.010000\n", "Epoch[6] Batch [599]\tSpeed: 201.282447 samples/sec\taccuracy=0.797604\tlr=0.010000\n", "Epoch[6] Batch [649]\tSpeed: 207.416583 samples/sec\taccuracy=0.797788\tlr=0.010000\n", "Epoch[6] Batch [699]\tSpeed: 212.875159 samples/sec\taccuracy=0.797411\tlr=0.010000\n", "Epoch[6] Batch [749]\tSpeed: 235.479729 samples/sec\taccuracy=0.799000\tlr=0.010000\n", "Epoch[6] Batch [799]\tSpeed: 209.175305 samples/sec\taccuracy=0.797813\tlr=0.010000\n", "Epoch[6] Batch [849]\tSpeed: 189.489280 samples/sec\taccuracy=0.798603\tlr=0.010000\n", "Epoch[6] Batch [899]\tSpeed: 189.878155 samples/sec\taccuracy=0.797292\tlr=0.010000\n", "Epoch[6] Batch [949]\tSpeed: 189.514270 samples/sec\taccuracy=0.796645\tlr=0.010000\n", "Epoch[6] Batch [999]\tSpeed: 187.033517 samples/sec\taccuracy=0.796813\tlr=0.010000\n", "Epoch[6] Batch [1049]\tSpeed: 191.507340 samples/sec\taccuracy=0.796548\tlr=0.010000\n", "Epoch[6] Batch [1099]\tSpeed: 187.767702 samples/sec\taccuracy=0.795170\tlr=0.010000\n", "Epoch[6] Batch [1149]\tSpeed: 201.352248 samples/sec\taccuracy=0.795163\tlr=0.010000\n", "Epoch[6] Batch [1199]\tSpeed: 198.538018 samples/sec\taccuracy=0.795937\tlr=0.010000\n", "Epoch[6] Batch [1249]\tSpeed: 191.058395 samples/sec\taccuracy=0.795250\tlr=0.010000\n", "Epoch[6] Batch [1299]\tSpeed: 193.909503 samples/sec\taccuracy=0.794904\tlr=0.010000\n", "Epoch[6] Batch [1349]\tSpeed: 200.059527 samples/sec\taccuracy=0.794120\tlr=0.010000\n", "Epoch[6] Batch [1399]\tSpeed: 190.510211 samples/sec\taccuracy=0.794420\tlr=0.010000\n", "Epoch[6] Batch [1449]\tSpeed: 189.579756 samples/sec\taccuracy=0.794957\tlr=0.010000\n", "Epoch[6] Batch [1499]\tSpeed: 191.098481 samples/sec\taccuracy=0.795625\tlr=0.010000\n", "Epoch[6] Batch [1549]\tSpeed: 188.210239 samples/sec\taccuracy=0.796089\tlr=0.010000\n", "Epoch[6] Batch [1599]\tSpeed: 182.441705 samples/sec\taccuracy=0.796875\tlr=0.010000\n", "Epoch[6] Batch [1649]\tSpeed: 188.610774 samples/sec\taccuracy=0.797045\tlr=0.010000\n", "Epoch[6] Batch [1699]\tSpeed: 183.134448 samples/sec\taccuracy=0.796875\tlr=0.010000\n", "Epoch[6] Batch [1749]\tSpeed: 188.652544 samples/sec\taccuracy=0.797357\tlr=0.010000\n", "Epoch[6] Batch [1799]\tSpeed: 187.064308 samples/sec\taccuracy=0.797847\tlr=0.010000\n", "Epoch[6] Batch [1849]\tSpeed: 188.470510 samples/sec\taccuracy=0.797669\tlr=0.010000\n", "Epoch[6] Batch [1899]\tSpeed: 188.387953 samples/sec\taccuracy=0.797763\tlr=0.010000\n", "Epoch[6] Batch [1949]\tSpeed: 189.551869 samples/sec\taccuracy=0.798782\tlr=0.010000\n", "Epoch[6] Batch [1999]\tSpeed: 186.815361 samples/sec\taccuracy=0.799219\tlr=0.010000\n", "Epoch[6] Batch [2049]\tSpeed: 187.632476 samples/sec\taccuracy=0.799268\tlr=0.010000\n", "Epoch[6] Batch [2099]\tSpeed: 189.282412 samples/sec\taccuracy=0.798988\tlr=0.010000\n", "Epoch[6] Batch [2149]\tSpeed: 188.842168 samples/sec\taccuracy=0.799041\tlr=0.010000\n", "Epoch[6] Batch [2199]\tSpeed: 187.658983 samples/sec\taccuracy=0.799432\tlr=0.010000\n", "Epoch[6] Batch [2249]\tSpeed: 188.034989 samples/sec\taccuracy=0.799278\tlr=0.010000\n", "Epoch[6] Batch [2299]\tSpeed: 182.269760 samples/sec\taccuracy=0.799375\tlr=0.010000\n", "Epoch[6] Batch [2349]\tSpeed: 188.221166 samples/sec\taccuracy=0.798617\tlr=0.010000\n", "Epoch[6] Batch [2399]\tSpeed: 213.170006 samples/sec\taccuracy=0.799193\tlr=0.010000\n", "Epoch[6] Batch [2449]\tSpeed: 216.877554 samples/sec\taccuracy=0.799617\tlr=0.010000\n", "Epoch[6] Batch [2499]\tSpeed: 223.609512 samples/sec\taccuracy=0.799650\tlr=0.010000\n", "Epoch[6] Batch [2549]\tSpeed: 223.323665 samples/sec\taccuracy=0.799510\tlr=0.010000\n", "Epoch[6] Batch [2599]\tSpeed: 225.317417 samples/sec\taccuracy=0.799736\tlr=0.010000\n", "Epoch[6] Batch [2649]\tSpeed: 225.017848 samples/sec\taccuracy=0.799575\tlr=0.010000\n", "Epoch[6] Batch [2699]\tSpeed: 221.601766 samples/sec\taccuracy=0.799630\tlr=0.010000\n", "Epoch[6] Batch [2749]\tSpeed: 222.408144 samples/sec\taccuracy=0.799682\tlr=0.010000\n", "Epoch[6] Batch [2799]\tSpeed: 226.685576 samples/sec\taccuracy=0.799665\tlr=0.010000\n", "[Epoch 6] training: accuracy=0.799809\n", "[Epoch 6] speed: 196 samples/sec\ttime cost: 229.265203\n", "[Epoch 6] validation: top1=0.954200 top5=0.999200\n", "[Epoch 6] Current best top-1: 0.954200 vs previous 0.946200, saved to /hy-tmp/c148b04f/.trial_0/best_checkpoint.pkl\n", "Epoch[7] Batch [49]\tSpeed: 180.973244 samples/sec\taccuracy=0.830000\tlr=0.010000\n", "Epoch[7] Batch [99]\tSpeed: 187.415246 samples/sec\taccuracy=0.805000\tlr=0.010000\n", "Epoch[7] Batch [149]\tSpeed: 188.827641 samples/sec\taccuracy=0.797917\tlr=0.010000\n", "Epoch[7] Batch [199]\tSpeed: 193.603281 samples/sec\taccuracy=0.799687\tlr=0.010000\n", "Epoch[7] Batch [249]\tSpeed: 186.924938 samples/sec\taccuracy=0.801000\tlr=0.010000\n", "Epoch[7] Batch [299]\tSpeed: 188.943826 samples/sec\taccuracy=0.805208\tlr=0.010000\n", "Epoch[7] Batch [349]\tSpeed: 191.746470 samples/sec\taccuracy=0.802679\tlr=0.010000\n", "Epoch[7] Batch [399]\tSpeed: 196.025227 samples/sec\taccuracy=0.803906\tlr=0.010000\n", "Epoch[7] Batch [449]\tSpeed: 196.110053 samples/sec\taccuracy=0.802778\tlr=0.010000\n", "Epoch[7] Batch [499]\tSpeed: 195.578698 samples/sec\taccuracy=0.803375\tlr=0.010000\n", "Epoch[7] Batch [549]\tSpeed: 234.114937 samples/sec\taccuracy=0.801705\tlr=0.010000\n", "Epoch[7] Batch [599]\tSpeed: 266.281579 samples/sec\taccuracy=0.802188\tlr=0.010000\n", "Epoch[7] Batch [649]\tSpeed: 262.022733 samples/sec\taccuracy=0.801538\tlr=0.010000\n", "Epoch[7] Batch [699]\tSpeed: 222.158256 samples/sec\taccuracy=0.801607\tlr=0.010000\n", "Epoch[7] Batch [749]\tSpeed: 232.463497 samples/sec\taccuracy=0.802833\tlr=0.010000\n", "Epoch[7] Batch [799]\tSpeed: 267.142848 samples/sec\taccuracy=0.803516\tlr=0.010000\n", "Epoch[7] Batch [849]\tSpeed: 269.133934 samples/sec\taccuracy=0.803529\tlr=0.010000\n", "Epoch[7] Batch [899]\tSpeed: 262.323343 samples/sec\taccuracy=0.803194\tlr=0.010000\n", "Epoch[7] Batch [949]\tSpeed: 192.934083 samples/sec\taccuracy=0.805066\tlr=0.010000\n", "Epoch[7] Batch [999]\tSpeed: 167.953071 samples/sec\taccuracy=0.805000\tlr=0.010000\n", "Epoch[7] Batch [1049]\tSpeed: 216.516400 samples/sec\taccuracy=0.804762\tlr=0.010000\n", "Epoch[7] Batch [1099]\tSpeed: 222.431704 samples/sec\taccuracy=0.805739\tlr=0.010000\n", "Epoch[7] Batch [1149]\tSpeed: 224.709263 samples/sec\taccuracy=0.804620\tlr=0.010000\n", "Epoch[7] Batch [1199]\tSpeed: 199.316379 samples/sec\taccuracy=0.805052\tlr=0.010000\n", "Epoch[7] Batch [1249]\tSpeed: 187.178572 samples/sec\taccuracy=0.805450\tlr=0.010000\n", "Epoch[7] Batch [1299]\tSpeed: 166.676992 samples/sec\taccuracy=0.805721\tlr=0.010000\n", "Epoch[7] Batch [1349]\tSpeed: 173.847207 samples/sec\taccuracy=0.805046\tlr=0.010000\n", "Epoch[7] Batch [1399]\tSpeed: 192.131626 samples/sec\taccuracy=0.804509\tlr=0.010000\n", "Epoch[7] Batch [1449]\tSpeed: 194.351822 samples/sec\taccuracy=0.804526\tlr=0.010000\n", "Epoch[7] Batch [1499]\tSpeed: 200.367648 samples/sec\taccuracy=0.804125\tlr=0.010000\n", "Epoch[7] Batch [1549]\tSpeed: 199.553630 samples/sec\taccuracy=0.803790\tlr=0.010000\n", "Epoch[7] Batch [1599]\tSpeed: 196.020990 samples/sec\taccuracy=0.803633\tlr=0.010000\n", "Epoch[7] Batch [1649]\tSpeed: 186.614163 samples/sec\taccuracy=0.803144\tlr=0.010000\n", "Epoch[7] Batch [1699]\tSpeed: 155.695361 samples/sec\taccuracy=0.804265\tlr=0.010000\n", "Epoch[7] Batch [1749]\tSpeed: 186.900959 samples/sec\taccuracy=0.804071\tlr=0.010000\n", "Epoch[7] Batch [1799]\tSpeed: 200.001383 samples/sec\taccuracy=0.804583\tlr=0.010000\n", "Epoch[7] Batch [1849]\tSpeed: 191.455404 samples/sec\taccuracy=0.805000\tlr=0.010000\n", "Epoch[7] Batch [1899]\tSpeed: 193.141143 samples/sec\taccuracy=0.805033\tlr=0.010000\n", "Epoch[7] Batch [1949]\tSpeed: 190.895070 samples/sec\taccuracy=0.805353\tlr=0.010000\n", "Epoch[7] Batch [1999]\tSpeed: 189.011474 samples/sec\taccuracy=0.805031\tlr=0.010000\n", "Epoch[7] Batch [2049]\tSpeed: 188.460940 samples/sec\taccuracy=0.805518\tlr=0.010000\n", "Epoch[7] Batch [2099]\tSpeed: 191.269602 samples/sec\taccuracy=0.805327\tlr=0.010000\n", "Epoch[7] Batch [2149]\tSpeed: 186.411461 samples/sec\taccuracy=0.805087\tlr=0.010000\n", "Epoch[7] Batch [2199]\tSpeed: 191.930906 samples/sec\taccuracy=0.805000\tlr=0.010000\n", "Epoch[7] Batch [2249]\tSpeed: 191.596418 samples/sec\taccuracy=0.805694\tlr=0.010000\n", "Epoch[7] Batch [2299]\tSpeed: 186.940132 samples/sec\taccuracy=0.805516\tlr=0.010000\n", "Epoch[7] Batch [2349]\tSpeed: 181.821134 samples/sec\taccuracy=0.805319\tlr=0.010000\n", "Epoch[7] Batch [2399]\tSpeed: 187.323278 samples/sec\taccuracy=0.805755\tlr=0.010000\n", "Epoch[7] Batch [2449]\tSpeed: 191.761363 samples/sec\taccuracy=0.805893\tlr=0.010000\n", "Epoch[7] Batch [2499]\tSpeed: 187.485354 samples/sec\taccuracy=0.805925\tlr=0.010000\n", "Epoch[7] Batch [2549]\tSpeed: 189.871611 samples/sec\taccuracy=0.806544\tlr=0.010000\n", "Epoch[7] Batch [2599]\tSpeed: 190.453528 samples/sec\taccuracy=0.806635\tlr=0.010000\n", "Epoch[7] Batch [2649]\tSpeed: 190.087652 samples/sec\taccuracy=0.806698\tlr=0.010000\n", "Epoch[7] Batch [2699]\tSpeed: 193.047703 samples/sec\taccuracy=0.806644\tlr=0.010000\n", "Epoch[7] Batch [2749]\tSpeed: 191.639619 samples/sec\taccuracy=0.806545\tlr=0.010000\n", "Epoch[7] Batch [2799]\tSpeed: 190.514137 samples/sec\taccuracy=0.807076\tlr=0.010000\n", "[Epoch 7] training: accuracy=0.807099\n", "[Epoch 7] speed: 196 samples/sec\ttime cost: 228.446615\n", "[Epoch 7] validation: top1=0.955600 top5=0.999600\n", "[Epoch 7] Current best top-1: 0.955600 vs previous 0.954200, saved to /hy-tmp/c148b04f/.trial_0/best_checkpoint.pkl\n", "Epoch[8] Batch [49]\tSpeed: 186.266092 samples/sec\taccuracy=0.812500\tlr=0.010000\n", "Epoch[8] Batch [99]\tSpeed: 204.641672 samples/sec\taccuracy=0.811250\tlr=0.010000\n", "Epoch[8] Batch [149]\tSpeed: 187.825153 samples/sec\taccuracy=0.815417\tlr=0.010000\n", "Epoch[8] Batch [199]\tSpeed: 186.296734 samples/sec\taccuracy=0.817813\tlr=0.010000\n", "Epoch[8] Batch [249]\tSpeed: 199.789603 samples/sec\taccuracy=0.818750\tlr=0.010000\n", "Epoch[8] Batch [299]\tSpeed: 201.166033 samples/sec\taccuracy=0.820833\tlr=0.010000\n", "Epoch[8] Batch [349]\tSpeed: 201.232966 samples/sec\taccuracy=0.819464\tlr=0.010000\n", "Epoch[8] Batch [399]\tSpeed: 201.976897 samples/sec\taccuracy=0.818750\tlr=0.010000\n", "Epoch[8] Batch [449]\tSpeed: 197.683829 samples/sec\taccuracy=0.818889\tlr=0.010000\n", "Epoch[8] Batch [499]\tSpeed: 188.052240 samples/sec\taccuracy=0.818750\tlr=0.010000\n", "Epoch[8] Batch [549]\tSpeed: 188.891293 samples/sec\taccuracy=0.817955\tlr=0.010000\n", "Epoch[8] Batch [599]\tSpeed: 190.114319 samples/sec\taccuracy=0.818542\tlr=0.010000\n", "Epoch[8] Batch [649]\tSpeed: 188.937496 samples/sec\taccuracy=0.818654\tlr=0.010000\n", "Epoch[8] Batch [699]\tSpeed: 183.878340 samples/sec\taccuracy=0.818214\tlr=0.010000\n", "Epoch[8] Batch [749]\tSpeed: 189.270625 samples/sec\taccuracy=0.817417\tlr=0.010000\n", "Epoch[8] Batch [799]\tSpeed: 190.430213 samples/sec\taccuracy=0.818281\tlr=0.010000\n", "Epoch[8] Batch [849]\tSpeed: 194.897021 samples/sec\taccuracy=0.819118\tlr=0.010000\n", "Epoch[8] Batch [899]\tSpeed: 189.658505 samples/sec\taccuracy=0.818125\tlr=0.010000\n", "Epoch[8] Batch [949]\tSpeed: 186.283516 samples/sec\taccuracy=0.817895\tlr=0.010000\n", "Epoch[8] Batch [999]\tSpeed: 186.147744 samples/sec\taccuracy=0.818312\tlr=0.010000\n", "Epoch[8] Batch [1049]\tSpeed: 188.476248 samples/sec\taccuracy=0.817857\tlr=0.010000\n", "Epoch[8] Batch [1099]\tSpeed: 190.127924 samples/sec\taccuracy=0.818011\tlr=0.010000\n", "Epoch[8] Batch [1149]\tSpeed: 192.562547 samples/sec\taccuracy=0.818424\tlr=0.010000\n", "Epoch[8] Batch [1199]\tSpeed: 188.755463 samples/sec\taccuracy=0.818646\tlr=0.010000\n", "Epoch[8] Batch [1249]\tSpeed: 187.706192 samples/sec\taccuracy=0.818800\tlr=0.010000\n", "Epoch[8] Batch [1299]\tSpeed: 189.938538 samples/sec\taccuracy=0.818894\tlr=0.010000\n", "Epoch[8] Batch [1349]\tSpeed: 191.409490 samples/sec\taccuracy=0.818426\tlr=0.010000\n", "Epoch[8] Batch [1399]\tSpeed: 190.879747 samples/sec\taccuracy=0.818304\tlr=0.010000\n", "Epoch[8] Batch [1449]\tSpeed: 191.949484 samples/sec\taccuracy=0.817974\tlr=0.010000\n", "Epoch[8] Batch [1499]\tSpeed: 190.851606 samples/sec\taccuracy=0.816792\tlr=0.010000\n", "Epoch[8] Batch [1549]\tSpeed: 189.997056 samples/sec\taccuracy=0.817016\tlr=0.010000\n", "Epoch[8] Batch [1599]\tSpeed: 189.462746 samples/sec\taccuracy=0.817617\tlr=0.010000\n", "Epoch[8] Batch [1649]\tSpeed: 191.730200 samples/sec\taccuracy=0.817462\tlr=0.010000\n", "Epoch[8] Batch [1699]\tSpeed: 190.952147 samples/sec\taccuracy=0.817243\tlr=0.010000\n", "Epoch[8] Batch [1749]\tSpeed: 192.185085 samples/sec\taccuracy=0.817679\tlr=0.010000\n", "Epoch[8] Batch [1799]\tSpeed: 191.927646 samples/sec\taccuracy=0.817951\tlr=0.010000\n", "Epoch[8] Batch [1849]\tSpeed: 198.813380 samples/sec\taccuracy=0.818243\tlr=0.010000\n", "Epoch[8] Batch [1899]\tSpeed: 220.399333 samples/sec\taccuracy=0.818355\tlr=0.010000\n", "Epoch[8] Batch [1949]\tSpeed: 220.728562 samples/sec\taccuracy=0.818269\tlr=0.010000\n", "Epoch[8] Batch [1999]\tSpeed: 220.839333 samples/sec\taccuracy=0.818281\tlr=0.010000\n", "Epoch[8] Batch [2049]\tSpeed: 224.614317 samples/sec\taccuracy=0.817744\tlr=0.010000\n", "Epoch[8] Batch [2099]\tSpeed: 222.903983 samples/sec\taccuracy=0.817351\tlr=0.010000\n", "Epoch[8] Batch [2149]\tSpeed: 221.828007 samples/sec\taccuracy=0.817093\tlr=0.010000\n", "Epoch[8] Batch [2199]\tSpeed: 223.570373 samples/sec\taccuracy=0.817187\tlr=0.010000\n", "Epoch[8] Batch [2249]\tSpeed: 196.745053 samples/sec\taccuracy=0.816833\tlr=0.010000\n", "Epoch[8] Batch [2299]\tSpeed: 187.222520 samples/sec\taccuracy=0.816793\tlr=0.010000\n", "Epoch[8] Batch [2349]\tSpeed: 187.801479 samples/sec\taccuracy=0.816622\tlr=0.010000\n", "Epoch[8] Batch [2399]\tSpeed: 204.504502 samples/sec\taccuracy=0.817057\tlr=0.010000\n", "Epoch[8] Batch [2449]\tSpeed: 201.459358 samples/sec\taccuracy=0.816811\tlr=0.010000\n", "Epoch[8] Batch [2499]\tSpeed: 189.966733 samples/sec\taccuracy=0.816525\tlr=0.010000\n", "Epoch[8] Batch [2549]\tSpeed: 190.527248 samples/sec\taccuracy=0.816691\tlr=0.010000\n", "Epoch[8] Batch [2599]\tSpeed: 188.612237 samples/sec\taccuracy=0.816635\tlr=0.010000\n", "Epoch[8] Batch [2649]\tSpeed: 188.641461 samples/sec\taccuracy=0.816439\tlr=0.010000\n", "Epoch[8] Batch [2699]\tSpeed: 188.897035 samples/sec\taccuracy=0.816667\tlr=0.010000\n", "Epoch[8] Batch [2749]\tSpeed: 190.547027 samples/sec\taccuracy=0.816705\tlr=0.010000\n", "Epoch[8] Batch [2799]\tSpeed: 192.014225 samples/sec\taccuracy=0.816540\tlr=0.010000\n", "[Epoch 8] training: accuracy=0.816345\n", "[Epoch 8] speed: 194 samples/sec\ttime cost: 230.793298\n", "[Epoch 8] validation: top1=0.953400 top5=0.999600\n", "Epoch[9] Batch [49]\tSpeed: 144.805320 samples/sec\taccuracy=0.802500\tlr=0.010000\n", "Epoch[9] Batch [99]\tSpeed: 177.308999 samples/sec\taccuracy=0.814375\tlr=0.010000\n", "Epoch[9] Batch [149]\tSpeed: 182.464156 samples/sec\taccuracy=0.817083\tlr=0.010000\n", "Epoch[9] Batch [199]\tSpeed: 178.034609 samples/sec\taccuracy=0.824688\tlr=0.010000\n", "Epoch[9] Batch [249]\tSpeed: 187.680091 samples/sec\taccuracy=0.820000\tlr=0.010000\n", "Epoch[9] Batch [299]\tSpeed: 190.231672 samples/sec\taccuracy=0.818750\tlr=0.010000\n", "Epoch[9] Batch [349]\tSpeed: 194.156291 samples/sec\taccuracy=0.818929\tlr=0.010000\n", "Epoch[9] Batch [399]\tSpeed: 190.671513 samples/sec\taccuracy=0.817969\tlr=0.010000\n", "Epoch[9] Batch [449]\tSpeed: 191.884040 samples/sec\taccuracy=0.817222\tlr=0.010000\n", "Epoch[9] Batch [499]\tSpeed: 159.686490 samples/sec\taccuracy=0.819125\tlr=0.010000\n", "Epoch[9] Batch [549]\tSpeed: 188.873196 samples/sec\taccuracy=0.819659\tlr=0.010000\n", "Epoch[9] Batch [599]\tSpeed: 187.625184 samples/sec\taccuracy=0.819375\tlr=0.010000\n", "Epoch[9] Batch [649]\tSpeed: 192.320573 samples/sec\taccuracy=0.819231\tlr=0.010000\n", "Epoch[9] Batch [699]\tSpeed: 185.925271 samples/sec\taccuracy=0.816786\tlr=0.010000\n", "Epoch[9] Batch [749]\tSpeed: 190.493609 samples/sec\taccuracy=0.817750\tlr=0.010000\n", "Epoch[9] Batch [799]\tSpeed: 188.270074 samples/sec\taccuracy=0.818594\tlr=0.010000\n", "Epoch[9] Batch [849]\tSpeed: 189.673782 samples/sec\taccuracy=0.820441\tlr=0.010000\n", "Epoch[9] Batch [899]\tSpeed: 183.686511 samples/sec\taccuracy=0.821944\tlr=0.010000\n", "Epoch[9] Batch [949]\tSpeed: 186.684691 samples/sec\taccuracy=0.822039\tlr=0.010000\n", "Epoch[9] Batch [999]\tSpeed: 189.331424 samples/sec\taccuracy=0.822688\tlr=0.010000\n", "Epoch[9] Batch [1049]\tSpeed: 191.179401 samples/sec\taccuracy=0.822321\tlr=0.010000\n", "Epoch[9] Batch [1099]\tSpeed: 187.458928 samples/sec\taccuracy=0.822784\tlr=0.010000\n", "Epoch[9] Batch [1149]\tSpeed: 186.827583 samples/sec\taccuracy=0.822609\tlr=0.010000\n", "Epoch[9] Batch [1199]\tSpeed: 187.082633 samples/sec\taccuracy=0.821719\tlr=0.010000\n", "Epoch[9] Batch [1249]\tSpeed: 190.074924 samples/sec\taccuracy=0.821350\tlr=0.010000\n", "Epoch[9] Batch [1299]\tSpeed: 190.427349 samples/sec\taccuracy=0.821875\tlr=0.010000\n", "Epoch[9] Batch [1349]\tSpeed: 186.794811 samples/sec\taccuracy=0.821435\tlr=0.010000\n", "Epoch[9] Batch [1399]\tSpeed: 190.004512 samples/sec\taccuracy=0.819911\tlr=0.010000\n", "Epoch[9] Batch [1449]\tSpeed: 192.725739 samples/sec\taccuracy=0.820043\tlr=0.010000\n", "Epoch[9] Batch [1499]\tSpeed: 192.028587 samples/sec\taccuracy=0.819375\tlr=0.010000\n", "Epoch[9] Batch [1549]\tSpeed: 193.229878 samples/sec\taccuracy=0.819073\tlr=0.010000\n", "Epoch[9] Batch [1599]\tSpeed: 193.869136 samples/sec\taccuracy=0.819609\tlr=0.010000\n", "Epoch[9] Batch [1649]\tSpeed: 192.485807 samples/sec\taccuracy=0.819356\tlr=0.010000\n", "Epoch[9] Batch [1699]\tSpeed: 197.165354 samples/sec\taccuracy=0.819669\tlr=0.010000\n", "Epoch[9] Batch [1749]\tSpeed: 191.984199 samples/sec\taccuracy=0.819214\tlr=0.010000\n", "Epoch[9] Batch [1799]\tSpeed: 204.628806 samples/sec\taccuracy=0.819201\tlr=0.010000\n", "Epoch[9] Batch [1849]\tSpeed: 192.946664 samples/sec\taccuracy=0.818649\tlr=0.010000\n", "Epoch[9] Batch [1899]\tSpeed: 191.633271 samples/sec\taccuracy=0.819145\tlr=0.010000\n", "Epoch[9] Batch [1949]\tSpeed: 206.141657 samples/sec\taccuracy=0.819135\tlr=0.010000\n", "Epoch[9] Batch [1999]\tSpeed: 174.157578 samples/sec\taccuracy=0.819312\tlr=0.010000\n", "Epoch[9] Batch [2049]\tSpeed: 211.351244 samples/sec\taccuracy=0.819329\tlr=0.010000\n", "Epoch[9] Batch [2099]\tSpeed: 188.799602 samples/sec\taccuracy=0.819196\tlr=0.010000\n", "Epoch[9] Batch [2149]\tSpeed: 192.584960 samples/sec\taccuracy=0.819390\tlr=0.010000\n", "Epoch[9] Batch [2199]\tSpeed: 189.319620 samples/sec\taccuracy=0.819602\tlr=0.010000\n", "Epoch[9] Batch [2249]\tSpeed: 191.959740 samples/sec\taccuracy=0.819750\tlr=0.010000\n", "Epoch[9] Batch [2299]\tSpeed: 190.360315 samples/sec\taccuracy=0.819973\tlr=0.010000\n", "Epoch[9] Batch [2349]\tSpeed: 179.592479 samples/sec\taccuracy=0.819947\tlr=0.010000\n", "Epoch[9] Batch [2399]\tSpeed: 187.984224 samples/sec\taccuracy=0.819297\tlr=0.010000\n", "Epoch[9] Batch [2449]\tSpeed: 188.282973 samples/sec\taccuracy=0.819821\tlr=0.010000\n", "Epoch[9] Batch [2499]\tSpeed: 189.527543 samples/sec\taccuracy=0.820000\tlr=0.010000\n", "Epoch[9] Batch [2549]\tSpeed: 189.144218 samples/sec\taccuracy=0.820123\tlr=0.010000\n", "Epoch[9] Batch [2599]\tSpeed: 189.977586 samples/sec\taccuracy=0.820288\tlr=0.010000\n", "Epoch[9] Batch [2649]\tSpeed: 189.224547 samples/sec\taccuracy=0.820094\tlr=0.010000\n", "Epoch[9] Batch [2699]\tSpeed: 188.489312 samples/sec\taccuracy=0.820278\tlr=0.010000\n", "Epoch[9] Batch [2749]\tSpeed: 190.456047 samples/sec\taccuracy=0.820773\tlr=0.010000\n", "Epoch[9] Batch [2799]\tSpeed: 191.060539 samples/sec\taccuracy=0.820759\tlr=0.010000\n", "[Epoch 9] training: accuracy=0.820790\n", "[Epoch 9] speed: 187 samples/sec\ttime cost: 239.271153\n", "[Epoch 9] validation: top1=0.956600 top5=0.999600\n", "[Epoch 9] Current best top-1: 0.956600 vs previous 0.955600, saved to /hy-tmp/c148b04f/.trial_0/best_checkpoint.pkl\n", "Epoch[10] Batch [49]\tSpeed: 160.807183 samples/sec\taccuracy=0.851250\tlr=0.010000\n", "Epoch[10] Batch [99]\tSpeed: 188.393167 samples/sec\taccuracy=0.840000\tlr=0.010000\n", "Epoch[10] Batch [149]\tSpeed: 184.887623 samples/sec\taccuracy=0.832500\tlr=0.010000\n", "Epoch[10] Batch [199]\tSpeed: 192.557265 samples/sec\taccuracy=0.826250\tlr=0.010000\n", "Epoch[10] Batch [249]\tSpeed: 201.747413 samples/sec\taccuracy=0.831250\tlr=0.010000\n", "Epoch[10] Batch [299]\tSpeed: 201.615596 samples/sec\taccuracy=0.829375\tlr=0.010000\n", "Epoch[10] Batch [349]\tSpeed: 199.630456 samples/sec\taccuracy=0.829107\tlr=0.010000\n", "Epoch[10] Batch [399]\tSpeed: 199.965817 samples/sec\taccuracy=0.831406\tlr=0.010000\n", "Epoch[10] Batch [449]\tSpeed: 200.559076 samples/sec\taccuracy=0.832222\tlr=0.010000\n", "Epoch[10] Batch [499]\tSpeed: 202.638371 samples/sec\taccuracy=0.828750\tlr=0.010000\n", "Epoch[10] Batch [549]\tSpeed: 202.939187 samples/sec\taccuracy=0.827955\tlr=0.010000\n", "Epoch[10] Batch [599]\tSpeed: 203.804587 samples/sec\taccuracy=0.828229\tlr=0.010000\n", "Epoch[10] Batch [649]\tSpeed: 202.682387 samples/sec\taccuracy=0.827981\tlr=0.010000\n", "Epoch[10] Batch [699]\tSpeed: 200.745682 samples/sec\taccuracy=0.827143\tlr=0.010000\n", "Epoch[10] Batch [749]\tSpeed: 197.218001 samples/sec\taccuracy=0.825583\tlr=0.010000\n", "Epoch[10] Batch [799]\tSpeed: 232.845586 samples/sec\taccuracy=0.826406\tlr=0.010000\n", "Epoch[10] Batch [849]\tSpeed: 266.388104 samples/sec\taccuracy=0.827206\tlr=0.010000\n", "Epoch[10] Batch [899]\tSpeed: 260.420040 samples/sec\taccuracy=0.827431\tlr=0.010000\n", "Epoch[10] Batch [949]\tSpeed: 262.785072 samples/sec\taccuracy=0.827303\tlr=0.010000\n", "Epoch[10] Batch [999]\tSpeed: 260.782978 samples/sec\taccuracy=0.828750\tlr=0.010000\n", "Epoch[10] Batch [1049]\tSpeed: 218.120983 samples/sec\taccuracy=0.829524\tlr=0.010000\n", "Epoch[10] Batch [1099]\tSpeed: 177.596853 samples/sec\taccuracy=0.830000\tlr=0.010000\n", "Epoch[10] Batch [1149]\tSpeed: 164.407712 samples/sec\taccuracy=0.828804\tlr=0.010000\n", "Epoch[10] Batch [1199]\tSpeed: 200.243077 samples/sec\taccuracy=0.828698\tlr=0.010000\n", "Epoch[10] Batch [1249]\tSpeed: 190.320711 samples/sec\taccuracy=0.827850\tlr=0.010000\n", "Epoch[10] Batch [1299]\tSpeed: 188.320867 samples/sec\taccuracy=0.828173\tlr=0.010000\n", "Epoch[10] Batch [1349]\tSpeed: 200.954452 samples/sec\taccuracy=0.827917\tlr=0.010000\n", "Epoch[10] Batch [1399]\tSpeed: 184.532550 samples/sec\taccuracy=0.828571\tlr=0.010000\n", "Epoch[10] Batch [1449]\tSpeed: 189.404278 samples/sec\taccuracy=0.828966\tlr=0.010000\n", "Epoch[10] Batch [1499]\tSpeed: 188.937155 samples/sec\taccuracy=0.829667\tlr=0.010000\n", "Epoch[10] Batch [1549]\tSpeed: 193.651047 samples/sec\taccuracy=0.829435\tlr=0.010000\n", "Epoch[10] Batch [1599]\tSpeed: 191.185610 samples/sec\taccuracy=0.829648\tlr=0.010000\n", "Epoch[10] Batch [1649]\tSpeed: 185.288337 samples/sec\taccuracy=0.829470\tlr=0.010000\n", "Epoch[10] Batch [1699]\tSpeed: 157.863175 samples/sec\taccuracy=0.829632\tlr=0.010000\n", "Epoch[10] Batch [1749]\tSpeed: 191.939602 samples/sec\taccuracy=0.829536\tlr=0.010000\n", "Epoch[10] Batch [1799]\tSpeed: 187.531249 samples/sec\taccuracy=0.829826\tlr=0.010000\n", "Epoch[10] Batch [1849]\tSpeed: 187.214978 samples/sec\taccuracy=0.829966\tlr=0.010000\n", "Epoch[10] Batch [1899]\tSpeed: 191.817303 samples/sec\taccuracy=0.829671\tlr=0.010000\n", "Epoch[10] Batch [1949]\tSpeed: 190.701725 samples/sec\taccuracy=0.829968\tlr=0.010000\n", "Epoch[10] Batch [1999]\tSpeed: 191.346171 samples/sec\taccuracy=0.829656\tlr=0.010000\n", "Epoch[10] Batch [2049]\tSpeed: 192.698711 samples/sec\taccuracy=0.829512\tlr=0.010000\n", "Epoch[10] Batch [2099]\tSpeed: 189.332631 samples/sec\taccuracy=0.829911\tlr=0.010000\n", "Epoch[10] Batch [2149]\tSpeed: 190.315594 samples/sec\taccuracy=0.829826\tlr=0.010000\n", "Epoch[10] Batch [2199]\tSpeed: 191.986858 samples/sec\taccuracy=0.829659\tlr=0.010000\n", "Epoch[10] Batch [2249]\tSpeed: 190.147512 samples/sec\taccuracy=0.830333\tlr=0.010000\n", "Epoch[10] Batch [2299]\tSpeed: 191.356254 samples/sec\taccuracy=0.830245\tlr=0.010000\n", "Epoch[10] Batch [2349]\tSpeed: 185.961707 samples/sec\taccuracy=0.830638\tlr=0.010000\n", "Epoch[10] Batch [2399]\tSpeed: 182.480320 samples/sec\taccuracy=0.830573\tlr=0.010000\n", "Epoch[10] Batch [2449]\tSpeed: 186.099592 samples/sec\taccuracy=0.830102\tlr=0.010000\n", "Epoch[10] Batch [2499]\tSpeed: 190.880236 samples/sec\taccuracy=0.829300\tlr=0.010000\n", "Epoch[10] Batch [2549]\tSpeed: 189.452230 samples/sec\taccuracy=0.829191\tlr=0.010000\n", "Epoch[10] Batch [2599]\tSpeed: 189.131094 samples/sec\taccuracy=0.829135\tlr=0.010000\n", "Epoch[10] Batch [2649]\tSpeed: 196.527148 samples/sec\taccuracy=0.829009\tlr=0.010000\n", "Epoch[10] Batch [2699]\tSpeed: 192.030224 samples/sec\taccuracy=0.828866\tlr=0.010000\n", "Epoch[10] Batch [2749]\tSpeed: 191.745188 samples/sec\taccuracy=0.828568\tlr=0.010000\n", "Epoch[10] Batch [2799]\tSpeed: 197.073235 samples/sec\taccuracy=0.828393\tlr=0.010000\n", "[Epoch 10] training: accuracy=0.828503\n", "[Epoch 10] speed: 195 samples/sec\ttime cost: 230.565408\n", "[Epoch 10] validation: top1=0.958200 top5=0.999600\n", "[Epoch 10] Current best top-1: 0.958200 vs previous 0.956600, saved to /hy-tmp/c148b04f/.trial_0/best_checkpoint.pkl\n", "Epoch[11] Batch [49]\tSpeed: 192.052547 samples/sec\taccuracy=0.821250\tlr=0.010000\n", "Epoch[11] Batch [99]\tSpeed: 200.091428 samples/sec\taccuracy=0.830000\tlr=0.010000\n", "Epoch[11] Batch [149]\tSpeed: 201.270844 samples/sec\taccuracy=0.830417\tlr=0.010000\n", "Epoch[11] Batch [199]\tSpeed: 201.351777 samples/sec\taccuracy=0.831250\tlr=0.010000\n", "Epoch[11] Batch [249]\tSpeed: 201.310173 samples/sec\taccuracy=0.832750\tlr=0.010000\n", "Epoch[11] Batch [299]\tSpeed: 203.683334 samples/sec\taccuracy=0.831667\tlr=0.010000\n", "Epoch[11] Batch [349]\tSpeed: 159.587348 samples/sec\taccuracy=0.831786\tlr=0.010000\n", "Epoch[11] Batch [399]\tSpeed: 188.717309 samples/sec\taccuracy=0.832187\tlr=0.010000\n", "Epoch[11] Batch [449]\tSpeed: 189.028681 samples/sec\taccuracy=0.832222\tlr=0.010000\n", "Epoch[11] Batch [499]\tSpeed: 193.787782 samples/sec\taccuracy=0.831000\tlr=0.010000\n", "Epoch[11] Batch [549]\tSpeed: 194.389710 samples/sec\taccuracy=0.831364\tlr=0.010000\n", "Epoch[11] Batch [599]\tSpeed: 199.852516 samples/sec\taccuracy=0.830833\tlr=0.010000\n", "Epoch[11] Batch [649]\tSpeed: 194.018228 samples/sec\taccuracy=0.831538\tlr=0.010000\n", "Epoch[11] Batch [699]\tSpeed: 196.668310 samples/sec\taccuracy=0.832054\tlr=0.010000\n", "Epoch[11] Batch [749]\tSpeed: 202.567040 samples/sec\taccuracy=0.832667\tlr=0.010000\n", "Epoch[11] Batch [799]\tSpeed: 195.479480 samples/sec\taccuracy=0.832187\tlr=0.010000\n", "Epoch[11] Batch [849]\tSpeed: 188.019532 samples/sec\taccuracy=0.832426\tlr=0.010000\n", "Epoch[11] Batch [899]\tSpeed: 182.116829 samples/sec\taccuracy=0.832778\tlr=0.010000\n", "Epoch[11] Batch [949]\tSpeed: 189.595824 samples/sec\taccuracy=0.833289\tlr=0.010000\n", "Epoch[11] Batch [999]\tSpeed: 187.320841 samples/sec\taccuracy=0.833812\tlr=0.010000\n", "Epoch[11] Batch [1049]\tSpeed: 188.121592 samples/sec\taccuracy=0.832679\tlr=0.010000\n", "Epoch[11] Batch [1099]\tSpeed: 186.900491 samples/sec\taccuracy=0.832841\tlr=0.010000\n", "Epoch[11] Batch [1149]\tSpeed: 188.051450 samples/sec\taccuracy=0.832391\tlr=0.010000\n", "Epoch[11] Batch [1199]\tSpeed: 188.478587 samples/sec\taccuracy=0.831927\tlr=0.010000\n", "Epoch[11] Batch [1249]\tSpeed: 187.925139 samples/sec\taccuracy=0.830800\tlr=0.010000\n", "Epoch[11] Batch [1299]\tSpeed: 186.138368 samples/sec\taccuracy=0.831731\tlr=0.010000\n", "Epoch[11] Batch [1349]\tSpeed: 187.500954 samples/sec\taccuracy=0.832361\tlr=0.010000\n", "Epoch[11] Batch [1399]\tSpeed: 188.467599 samples/sec\taccuracy=0.831429\tlr=0.010000\n", "Epoch[11] Batch [1449]\tSpeed: 188.222085 samples/sec\taccuracy=0.832155\tlr=0.010000\n", "Epoch[11] Batch [1499]\tSpeed: 187.635603 samples/sec\taccuracy=0.832083\tlr=0.010000\n", "Epoch[11] Batch [1549]\tSpeed: 185.479652 samples/sec\taccuracy=0.831734\tlr=0.010000\n", "Epoch[11] Batch [1599]\tSpeed: 151.381577 samples/sec\taccuracy=0.832031\tlr=0.010000\n", "Epoch[11] Batch [1649]\tSpeed: 213.451712 samples/sec\taccuracy=0.831629\tlr=0.010000\n", "Epoch[11] Batch [1699]\tSpeed: 215.826406 samples/sec\taccuracy=0.831176\tlr=0.010000\n", "Epoch[11] Batch [1749]\tSpeed: 217.056611 samples/sec\taccuracy=0.831214\tlr=0.010000\n", "Epoch[11] Batch [1799]\tSpeed: 219.936415 samples/sec\taccuracy=0.831389\tlr=0.010000\n", "Epoch[11] Batch [1849]\tSpeed: 218.364367 samples/sec\taccuracy=0.831318\tlr=0.010000\n", "Epoch[11] Batch [1899]\tSpeed: 212.186268 samples/sec\taccuracy=0.831776\tlr=0.010000\n", "Epoch[11] Batch [1949]\tSpeed: 203.157098 samples/sec\taccuracy=0.831122\tlr=0.010000\n", "Epoch[11] Batch [1999]\tSpeed: 198.380237 samples/sec\taccuracy=0.831719\tlr=0.010000\n", "Epoch[11] Batch [2049]\tSpeed: 206.638711 samples/sec\taccuracy=0.831646\tlr=0.010000\n", "Epoch[11] Batch [2099]\tSpeed: 201.223179 samples/sec\taccuracy=0.831071\tlr=0.010000\n", "Epoch[11] Batch [2149]\tSpeed: 189.753360 samples/sec\taccuracy=0.830988\tlr=0.010000\n", "Epoch[11] Batch [2199]\tSpeed: 200.891710 samples/sec\taccuracy=0.830937\tlr=0.010000\n", "Epoch[11] Batch [2249]\tSpeed: 196.695287 samples/sec\taccuracy=0.831278\tlr=0.010000\n", "Epoch[11] Batch [2299]\tSpeed: 205.106521 samples/sec\taccuracy=0.831576\tlr=0.010000\n", "Epoch[11] Batch [2349]\tSpeed: 191.697821 samples/sec\taccuracy=0.831489\tlr=0.010000\n", "Epoch[11] Batch [2399]\tSpeed: 198.340708 samples/sec\taccuracy=0.831042\tlr=0.010000\n", "Epoch[11] Batch [2449]\tSpeed: 193.810930 samples/sec\taccuracy=0.830638\tlr=0.010000\n", "Epoch[11] Batch [2499]\tSpeed: 200.900093 samples/sec\taccuracy=0.830575\tlr=0.010000\n", "Epoch[11] Batch [2549]\tSpeed: 205.133467 samples/sec\taccuracy=0.830882\tlr=0.010000\n", "Epoch[11] Batch [2599]\tSpeed: 153.457007 samples/sec\taccuracy=0.830625\tlr=0.010000\n", "Epoch[11] Batch [2649]\tSpeed: 216.199510 samples/sec\taccuracy=0.829929\tlr=0.010000\n", "Epoch[11] Batch [2699]\tSpeed: 192.087916 samples/sec\taccuracy=0.830116\tlr=0.010000\n", "Epoch[11] Batch [2749]\tSpeed: 191.129406 samples/sec\taccuracy=0.830250\tlr=0.010000\n", "Epoch[11] Batch [2799]\tSpeed: 192.797364 samples/sec\taccuracy=0.829911\tlr=0.010000\n", "[Epoch 11] training: accuracy=0.829859\n", "[Epoch 11] speed: 193 samples/sec\ttime cost: 232.217432\n", "[Epoch 11] validation: top1=0.965600 top5=0.999400\n", "[Epoch 11] Current best top-1: 0.965600 vs previous 0.958200, saved to /hy-tmp/c148b04f/.trial_0/best_checkpoint.pkl\n", "Epoch[12] Batch [49]\tSpeed: 157.160988 samples/sec\taccuracy=0.865000\tlr=0.010000\n", "Epoch[12] Batch [99]\tSpeed: 185.171780 samples/sec\taccuracy=0.843750\tlr=0.010000\n", "Epoch[12] Batch [149]\tSpeed: 184.623494 samples/sec\taccuracy=0.842500\tlr=0.010000\n", "Epoch[12] Batch [199]\tSpeed: 186.996389 samples/sec\taccuracy=0.837500\tlr=0.010000\n", "Epoch[12] Batch [249]\tSpeed: 191.067371 samples/sec\taccuracy=0.835500\tlr=0.010000\n", "Epoch[12] Batch [299]\tSpeed: 168.654425 samples/sec\taccuracy=0.835208\tlr=0.010000\n", "Epoch[12] Batch [349]\tSpeed: 202.070032 samples/sec\taccuracy=0.838571\tlr=0.010000\n", "Epoch[12] Batch [399]\tSpeed: 186.345402 samples/sec\taccuracy=0.839219\tlr=0.010000\n", "Epoch[12] Batch [449]\tSpeed: 191.425302 samples/sec\taccuracy=0.839722\tlr=0.010000\n", "Epoch[12] Batch [499]\tSpeed: 189.566690 samples/sec\taccuracy=0.837500\tlr=0.010000\n", "Epoch[12] Batch [549]\tSpeed: 186.471214 samples/sec\taccuracy=0.834773\tlr=0.010000\n", "Epoch[12] Batch [599]\tSpeed: 193.961311 samples/sec\taccuracy=0.833646\tlr=0.010000\n", "Epoch[12] Batch [649]\tSpeed: 205.857280 samples/sec\taccuracy=0.834327\tlr=0.010000\n", "Epoch[12] Batch [699]\tSpeed: 206.150459 samples/sec\taccuracy=0.834196\tlr=0.010000\n", "Epoch[12] Batch [749]\tSpeed: 213.441108 samples/sec\taccuracy=0.833917\tlr=0.010000\n", "Epoch[12] Batch [799]\tSpeed: 215.731424 samples/sec\taccuracy=0.833516\tlr=0.010000\n", "Epoch[12] Batch [849]\tSpeed: 195.304453 samples/sec\taccuracy=0.832794\tlr=0.010000\n", "Epoch[12] Batch [899]\tSpeed: 192.458637 samples/sec\taccuracy=0.832778\tlr=0.010000\n", "Epoch[12] Batch [949]\tSpeed: 189.903397 samples/sec\taccuracy=0.834013\tlr=0.010000\n", "Epoch[12] Batch [999]\tSpeed: 191.008518 samples/sec\taccuracy=0.833875\tlr=0.010000\n", "Epoch[12] Batch [1049]\tSpeed: 191.002495 samples/sec\taccuracy=0.833690\tlr=0.010000\n", "Epoch[12] Batch [1099]\tSpeed: 189.251399 samples/sec\taccuracy=0.834148\tlr=0.010000\n", "Epoch[12] Batch [1149]\tSpeed: 190.941498 samples/sec\taccuracy=0.833424\tlr=0.010000\n", "Epoch[12] Batch [1199]\tSpeed: 190.958558 samples/sec\taccuracy=0.833385\tlr=0.010000\n", "Epoch[12] Batch [1249]\tSpeed: 190.154979 samples/sec\taccuracy=0.833150\tlr=0.010000\n", "Epoch[12] Batch [1299]\tSpeed: 207.673253 samples/sec\taccuracy=0.833221\tlr=0.010000\n", "Epoch[12] Batch [1349]\tSpeed: 165.727865 samples/sec\taccuracy=0.832917\tlr=0.010000\n", "Epoch[12] Batch [1399]\tSpeed: 217.777839 samples/sec\taccuracy=0.833571\tlr=0.010000\n", "Epoch[12] Batch [1449]\tSpeed: 195.220130 samples/sec\taccuracy=0.834052\tlr=0.010000\n", "Epoch[12] Batch [1499]\tSpeed: 201.873536 samples/sec\taccuracy=0.833875\tlr=0.010000\n", "Epoch[12] Batch [1549]\tSpeed: 200.028173 samples/sec\taccuracy=0.834960\tlr=0.010000\n", "Epoch[12] Batch [1599]\tSpeed: 197.385920 samples/sec\taccuracy=0.835664\tlr=0.010000\n", "Epoch[12] Batch [1649]\tSpeed: 199.029092 samples/sec\taccuracy=0.835379\tlr=0.010000\n", "Epoch[12] Batch [1699]\tSpeed: 205.430837 samples/sec\taccuracy=0.834853\tlr=0.010000\n", "Epoch[12] Batch [1749]\tSpeed: 205.107875 samples/sec\taccuracy=0.835036\tlr=0.010000\n", "Epoch[12] Batch [1799]\tSpeed: 205.967935 samples/sec\taccuracy=0.835208\tlr=0.010000\n", "Epoch[12] Batch [1849]\tSpeed: 203.936990 samples/sec\taccuracy=0.835068\tlr=0.010000\n", "Epoch[12] Batch [1899]\tSpeed: 190.703622 samples/sec\taccuracy=0.835033\tlr=0.010000\n", "Epoch[12] Batch [1949]\tSpeed: 189.029054 samples/sec\taccuracy=0.834744\tlr=0.010000\n", "Epoch[12] Batch [1999]\tSpeed: 194.136633 samples/sec\taccuracy=0.834375\tlr=0.010000\n", "Epoch[12] Batch [2049]\tSpeed: 199.560193 samples/sec\taccuracy=0.835030\tlr=0.010000\n", "Epoch[12] Batch [2099]\tSpeed: 200.300489 samples/sec\taccuracy=0.834851\tlr=0.010000\n", "Epoch[12] Batch [2149]\tSpeed: 200.202240 samples/sec\taccuracy=0.834942\tlr=0.010000\n", "Epoch[12] Batch [2199]\tSpeed: 201.949315 samples/sec\taccuracy=0.834574\tlr=0.010000\n", "Epoch[12] Batch [2249]\tSpeed: 188.361440 samples/sec\taccuracy=0.834972\tlr=0.010000\n", "Epoch[12] Batch [2299]\tSpeed: 185.679797 samples/sec\taccuracy=0.834565\tlr=0.010000\n", "Epoch[12] Batch [2349]\tSpeed: 186.617266 samples/sec\taccuracy=0.834388\tlr=0.010000\n", "Epoch[12] Batch [2399]\tSpeed: 193.648599 samples/sec\taccuracy=0.834036\tlr=0.010000\n", "Epoch[12] Batch [2449]\tSpeed: 202.019957 samples/sec\taccuracy=0.833954\tlr=0.010000\n", "Epoch[12] Batch [2499]\tSpeed: 201.916673 samples/sec\taccuracy=0.833475\tlr=0.010000\n", "Epoch[12] Batch [2549]\tSpeed: 187.960952 samples/sec\taccuracy=0.833578\tlr=0.010000\n", "Epoch[12] Batch [2599]\tSpeed: 186.992429 samples/sec\taccuracy=0.833005\tlr=0.010000\n", "Epoch[12] Batch [2649]\tSpeed: 188.880447 samples/sec\taccuracy=0.832995\tlr=0.010000\n", "Epoch[12] Batch [2699]\tSpeed: 187.285073 samples/sec\taccuracy=0.833449\tlr=0.010000\n", "Epoch[12] Batch [2749]\tSpeed: 193.142966 samples/sec\taccuracy=0.833341\tlr=0.010000\n", "Epoch[12] Batch [2799]\tSpeed: 203.258109 samples/sec\taccuracy=0.833527\tlr=0.010000\n", "[Epoch 12] training: accuracy=0.833348\n", "[Epoch 12] speed: 193 samples/sec\ttime cost: 232.508509\n", "[Epoch 12] validation: top1=0.959400 top5=0.999400\n", "Epoch[13] Batch [49]\tSpeed: 163.323943 samples/sec\taccuracy=0.847500\tlr=0.010000\n", "Epoch[13] Batch [99]\tSpeed: 189.008163 samples/sec\taccuracy=0.834375\tlr=0.010000\n", "Epoch[13] Batch [149]\tSpeed: 193.819483 samples/sec\taccuracy=0.842500\tlr=0.010000\n", "Epoch[13] Batch [199]\tSpeed: 158.509566 samples/sec\taccuracy=0.842187\tlr=0.010000\n", "Epoch[13] Batch [249]\tSpeed: 220.432012 samples/sec\taccuracy=0.842750\tlr=0.010000\n", "Epoch[13] Batch [299]\tSpeed: 218.581650 samples/sec\taccuracy=0.844375\tlr=0.010000\n", "Epoch[13] Batch [349]\tSpeed: 219.153927 samples/sec\taccuracy=0.846250\tlr=0.010000\n", "Epoch[13] Batch [399]\tSpeed: 218.479207 samples/sec\taccuracy=0.847812\tlr=0.010000\n", "Epoch[13] Batch [449]\tSpeed: 219.436225 samples/sec\taccuracy=0.847500\tlr=0.010000\n", "Epoch[13] Batch [499]\tSpeed: 171.233578 samples/sec\taccuracy=0.844000\tlr=0.010000\n", "Epoch[13] Batch [549]\tSpeed: 192.083628 samples/sec\taccuracy=0.844659\tlr=0.010000\n", "Epoch[13] Batch [599]\tSpeed: 189.390915 samples/sec\taccuracy=0.844167\tlr=0.010000\n", "Epoch[13] Batch [649]\tSpeed: 187.610760 samples/sec\taccuracy=0.841635\tlr=0.010000\n", "Epoch[13] Batch [699]\tSpeed: 187.397379 samples/sec\taccuracy=0.841964\tlr=0.010000\n", "Epoch[13] Batch [749]\tSpeed: 188.017341 samples/sec\taccuracy=0.841833\tlr=0.010000\n", "Epoch[13] Batch [799]\tSpeed: 203.993365 samples/sec\taccuracy=0.842578\tlr=0.010000\n", "Epoch[13] Batch [849]\tSpeed: 204.088270 samples/sec\taccuracy=0.842279\tlr=0.010000\n", "Epoch[13] Batch [899]\tSpeed: 187.661544 samples/sec\taccuracy=0.842083\tlr=0.010000\n", "Epoch[13] Batch [949]\tSpeed: 194.998198 samples/sec\taccuracy=0.842895\tlr=0.010000\n", "Epoch[13] Batch [999]\tSpeed: 206.713742 samples/sec\taccuracy=0.842938\tlr=0.010000\n", "Epoch[13] Batch [1049]\tSpeed: 207.940003 samples/sec\taccuracy=0.843036\tlr=0.010000\n", "Epoch[13] Batch [1099]\tSpeed: 191.553083 samples/sec\taccuracy=0.842045\tlr=0.010000\n", "Epoch[13] Batch [1149]\tSpeed: 190.053769 samples/sec\taccuracy=0.842174\tlr=0.010000\n", "Epoch[13] Batch [1199]\tSpeed: 192.723592 samples/sec\taccuracy=0.842917\tlr=0.010000\n", "Epoch[13] Batch [1249]\tSpeed: 192.362172 samples/sec\taccuracy=0.842100\tlr=0.010000\n", "Epoch[13] Batch [1299]\tSpeed: 189.091744 samples/sec\taccuracy=0.841971\tlr=0.010000\n", "Epoch[13] Batch [1349]\tSpeed: 188.636318 samples/sec\taccuracy=0.841667\tlr=0.010000\n", "Epoch[13] Batch [1399]\tSpeed: 189.092564 samples/sec\taccuracy=0.841786\tlr=0.010000\n", "Epoch[13] Batch [1449]\tSpeed: 189.598760 samples/sec\taccuracy=0.841767\tlr=0.010000\n", "Epoch[13] Batch [1499]\tSpeed: 191.700844 samples/sec\taccuracy=0.841333\tlr=0.010000\n", "Epoch[13] Batch [1549]\tSpeed: 186.958818 samples/sec\taccuracy=0.841411\tlr=0.010000\n", "Epoch[13] Batch [1599]\tSpeed: 183.448434 samples/sec\taccuracy=0.840781\tlr=0.010000\n", "Epoch[13] Batch [1649]\tSpeed: 185.156494 samples/sec\taccuracy=0.841212\tlr=0.010000\n", "Epoch[13] Batch [1699]\tSpeed: 188.577480 samples/sec\taccuracy=0.840919\tlr=0.010000\n", "Epoch[13] Batch [1749]\tSpeed: 199.747132 samples/sec\taccuracy=0.841036\tlr=0.010000\n", "Epoch[13] Batch [1799]\tSpeed: 192.287752 samples/sec\taccuracy=0.840903\tlr=0.010000\n", "Epoch[13] Batch [1849]\tSpeed: 191.173400 samples/sec\taccuracy=0.841824\tlr=0.010000\n", "Epoch[13] Batch [1899]\tSpeed: 187.401523 samples/sec\taccuracy=0.841480\tlr=0.010000\n", "Epoch[13] Batch [1949]\tSpeed: 208.698486 samples/sec\taccuracy=0.841346\tlr=0.010000\n", "Epoch[13] Batch [1999]\tSpeed: 214.900122 samples/sec\taccuracy=0.841344\tlr=0.010000\n", "Epoch[13] Batch [2049]\tSpeed: 183.910741 samples/sec\taccuracy=0.841555\tlr=0.010000\n", "Epoch[13] Batch [2099]\tSpeed: 207.458903 samples/sec\taccuracy=0.841101\tlr=0.010000\n", "Epoch[13] Batch [2149]\tSpeed: 227.248178 samples/sec\taccuracy=0.840814\tlr=0.010000\n", "Epoch[13] Batch [2199]\tSpeed: 239.005351 samples/sec\taccuracy=0.840795\tlr=0.010000\n", "Epoch[13] Batch [2249]\tSpeed: 239.264910 samples/sec\taccuracy=0.840667\tlr=0.010000\n", "Epoch[13] Batch [2299]\tSpeed: 239.295897 samples/sec\taccuracy=0.841060\tlr=0.010000\n", "Epoch[13] Batch [2349]\tSpeed: 221.741723 samples/sec\taccuracy=0.840851\tlr=0.010000\n", "Epoch[13] Batch [2399]\tSpeed: 218.594623 samples/sec\taccuracy=0.840964\tlr=0.010000\n", "Epoch[13] Batch [2449]\tSpeed: 214.676714 samples/sec\taccuracy=0.840995\tlr=0.010000\n", "Epoch[13] Batch [2499]\tSpeed: 218.722949 samples/sec\taccuracy=0.840700\tlr=0.010000\n", "Epoch[13] Batch [2549]\tSpeed: 232.782086 samples/sec\taccuracy=0.840417\tlr=0.010000\n", "Epoch[13] Batch [2599]\tSpeed: 229.453555 samples/sec\taccuracy=0.840216\tlr=0.010000\n", "Epoch[13] Batch [2649]\tSpeed: 213.674862 samples/sec\taccuracy=0.839906\tlr=0.010000\n", "Epoch[13] Batch [2699]\tSpeed: 188.872728 samples/sec\taccuracy=0.840162\tlr=0.010000\n", "Epoch[13] Batch [2749]\tSpeed: 190.395582 samples/sec\taccuracy=0.840364\tlr=0.010000\n", "Epoch[13] Batch [2799]\tSpeed: 189.636949 samples/sec\taccuracy=0.840536\tlr=0.010000\n", "[Epoch 13] training: accuracy=0.840483\n", "[Epoch 13] speed: 198 samples/sec\ttime cost: 226.305372\n", "[Epoch 13] validation: top1=0.956000 top5=0.999000\n", "Epoch[14] Batch [49]\tSpeed: 167.778972 samples/sec\taccuracy=0.840000\tlr=0.010000\n", "Epoch[14] Batch [99]\tSpeed: 189.725368 samples/sec\taccuracy=0.829375\tlr=0.010000\n", "Epoch[14] Batch [149]\tSpeed: 185.652891 samples/sec\taccuracy=0.840833\tlr=0.010000\n", "Epoch[14] Batch [199]\tSpeed: 183.953077 samples/sec\taccuracy=0.845625\tlr=0.010000\n", "Epoch[14] Batch [249]\tSpeed: 192.190920 samples/sec\taccuracy=0.842500\tlr=0.010000\n", "Epoch[14] Batch [299]\tSpeed: 194.979729 samples/sec\taccuracy=0.843958\tlr=0.010000\n", "Epoch[14] Batch [349]\tSpeed: 205.683684 samples/sec\taccuracy=0.840000\tlr=0.010000\n", "Epoch[14] Batch [399]\tSpeed: 205.594533 samples/sec\taccuracy=0.837500\tlr=0.010000\n", "Epoch[14] Batch [449]\tSpeed: 196.810092 samples/sec\taccuracy=0.836944\tlr=0.010000\n", "Epoch[14] Batch [499]\tSpeed: 190.075269 samples/sec\taccuracy=0.837875\tlr=0.010000\n", "Epoch[14] Batch [549]\tSpeed: 190.267959 samples/sec\taccuracy=0.837159\tlr=0.010000\n", "Epoch[14] Batch [599]\tSpeed: 200.328902 samples/sec\taccuracy=0.836458\tlr=0.010000\n", "Epoch[14] Batch [649]\tSpeed: 189.042366 samples/sec\taccuracy=0.838269\tlr=0.010000\n", "Epoch[14] Batch [699]\tSpeed: 189.987675 samples/sec\taccuracy=0.839196\tlr=0.010000\n", "Epoch[14] Batch [749]\tSpeed: 194.558948 samples/sec\taccuracy=0.839000\tlr=0.010000\n", "Epoch[14] Batch [799]\tSpeed: 189.491805 samples/sec\taccuracy=0.838359\tlr=0.010000\n", "Epoch[14] Batch [849]\tSpeed: 195.228944 samples/sec\taccuracy=0.837794\tlr=0.010000\n", "Epoch[14] Batch [899]\tSpeed: 183.395885 samples/sec\taccuracy=0.838819\tlr=0.010000\n", "Epoch[14] Batch [949]\tSpeed: 188.738136 samples/sec\taccuracy=0.839539\tlr=0.010000\n", "Epoch[14] Batch [999]\tSpeed: 186.068612 samples/sec\taccuracy=0.839938\tlr=0.010000\n", "Epoch[14] Batch [1049]\tSpeed: 190.308470 samples/sec\taccuracy=0.839048\tlr=0.010000\n", "Epoch[14] Batch [1099]\tSpeed: 189.602938 samples/sec\taccuracy=0.839261\tlr=0.010000\n", "Epoch[14] Batch [1149]\tSpeed: 189.546686 samples/sec\taccuracy=0.838533\tlr=0.010000\n", "Epoch[14] Batch [1199]\tSpeed: 191.981563 samples/sec\taccuracy=0.838698\tlr=0.010000\n", "Epoch[14] Batch [1249]\tSpeed: 192.317487 samples/sec\taccuracy=0.838300\tlr=0.010000\n", "Epoch[14] Batch [1299]\tSpeed: 195.719894 samples/sec\taccuracy=0.838125\tlr=0.010000\n", "Epoch[14] Batch [1349]\tSpeed: 190.049808 samples/sec\taccuracy=0.838704\tlr=0.010000\n", "Epoch[14] Batch [1399]\tSpeed: 192.028565 samples/sec\taccuracy=0.837634\tlr=0.010000\n", "Epoch[14] Batch [1449]\tSpeed: 200.702492 samples/sec\taccuracy=0.838147\tlr=0.010000\n", "Epoch[14] Batch [1499]\tSpeed: 215.862242 samples/sec\taccuracy=0.837792\tlr=0.010000\n", "Epoch[14] Batch [1549]\tSpeed: 208.062469 samples/sec\taccuracy=0.837661\tlr=0.010000\n", "Epoch[14] Batch [1599]\tSpeed: 188.957403 samples/sec\taccuracy=0.837383\tlr=0.010000\n", "Epoch[14] Batch [1649]\tSpeed: 186.531202 samples/sec\taccuracy=0.837083\tlr=0.010000\n", "Epoch[14] Batch [1699]\tSpeed: 187.496135 samples/sec\taccuracy=0.837426\tlr=0.010000\n", "Epoch[14] Batch [1749]\tSpeed: 186.238023 samples/sec\taccuracy=0.837893\tlr=0.010000\n", "Epoch[14] Batch [1799]\tSpeed: 192.013532 samples/sec\taccuracy=0.837917\tlr=0.010000\n", "Epoch[14] Batch [1849]\tSpeed: 202.584149 samples/sec\taccuracy=0.838176\tlr=0.010000\n", "Epoch[14] Batch [1899]\tSpeed: 204.241962 samples/sec\taccuracy=0.838322\tlr=0.010000\n", "Epoch[14] Batch [1949]\tSpeed: 204.374262 samples/sec\taccuracy=0.838590\tlr=0.010000\n", "Epoch[14] Batch [1999]\tSpeed: 202.309228 samples/sec\taccuracy=0.838156\tlr=0.010000\n", "Epoch[14] Batch [2049]\tSpeed: 189.708087 samples/sec\taccuracy=0.838293\tlr=0.010000\n", "Epoch[14] Batch [2099]\tSpeed: 186.813864 samples/sec\taccuracy=0.838333\tlr=0.010000\n", "Epoch[14] Batch [2149]\tSpeed: 189.971401 samples/sec\taccuracy=0.838721\tlr=0.010000\n", "Epoch[14] Batch [2199]\tSpeed: 190.714732 samples/sec\taccuracy=0.838977\tlr=0.010000\n", "Epoch[14] Batch [2249]\tSpeed: 192.449773 samples/sec\taccuracy=0.839250\tlr=0.010000\n", "Epoch[14] Batch [2299]\tSpeed: 189.536343 samples/sec\taccuracy=0.838913\tlr=0.010000\n", "Epoch[14] Batch [2349]\tSpeed: 188.600533 samples/sec\taccuracy=0.838963\tlr=0.010000\n", "Epoch[14] Batch [2399]\tSpeed: 187.414628 samples/sec\taccuracy=0.838984\tlr=0.010000\n", "Epoch[14] Batch [2449]\tSpeed: 193.161390 samples/sec\taccuracy=0.838520\tlr=0.010000\n", "Epoch[14] Batch [2499]\tSpeed: 204.176405 samples/sec\taccuracy=0.838450\tlr=0.010000\n", "Epoch[14] Batch [2549]\tSpeed: 204.404752 samples/sec\taccuracy=0.838676\tlr=0.010000\n", "Epoch[14] Batch [2599]\tSpeed: 205.100478 samples/sec\taccuracy=0.838654\tlr=0.010000\n", "Epoch[14] Batch [2649]\tSpeed: 198.978790 samples/sec\taccuracy=0.838561\tlr=0.010000\n", "Epoch[14] Batch [2699]\tSpeed: 204.000682 samples/sec\taccuracy=0.838843\tlr=0.010000\n", "Epoch[14] Batch [2749]\tSpeed: 205.230465 samples/sec\taccuracy=0.839205\tlr=0.010000\n", "Epoch[14] Batch [2799]\tSpeed: 190.405716 samples/sec\taccuracy=0.838973\tlr=0.010000\n", "[Epoch 14] training: accuracy=0.839060\n", "[Epoch 14] speed: 193 samples/sec\ttime cost: 232.667396\n", "[Epoch 14] validation: top1=0.958600 top5=0.999600\n", "Applying the state from the best checkpoint...\n", "Finished, total runtime is 3651.02 s\n", "{ 'best_config': { 'batch_size': 16,\n", " 'dist_ip_addrs': None,\n", " 'early_stop_baseline': -inf,\n", " 'early_stop_max_value': inf,\n", " 'early_stop_patience': 10,\n", " 'epochs': 15,\n", " 'final_fit': False,\n", " 'gpus': [0],\n", " 'lr': 0.01,\n", " 'model': 'resnet50',\n", " 'ngpus_per_trial': 8,\n", " 'nthreads_per_trial': 128,\n", " 'num_workers': 16,\n", " 'searcher': 'random',\n", " 'seed': 435,\n", " 'time_limits': 7200},\n", " 'total_time': 3462.3342187404633,\n", " 'train_acc': 0.8390602773826458,\n", " 'valid_acc': 0.9656}\n" ] } ], "source": [ "predictor = ImagePredictor().fit(train_data=train_data)\n", "predictions = predictor.predict(test_data)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Top-1 train acc: 0.839, val acc: 0.966\n" ] } ], "source": [ "fit_result = predictor.fit_summary()\n", "print('Top-1 train acc: %.3f, val acc: %.3f' %(fit_result['train_acc'], fit_result['valid_acc']))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "[Epoch 11] validation: top1=0.961800 top5=0.999400\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Top-1 test acc: 0.962\n" ] } ], "source": [ "test_acc = predictor.evaluate(test_data)\n", "print('Top-1 test acc: %.3f' % test_acc['top1'])" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "filename = 'predictor.ag'\n", "predictor.save(filename)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "以下是sota算法vision transformers的实践。" ] }, { "cell_type": "code", "execution_count": 2, "metadata": {}, "outputs": [], "source": [ "!pip intsall vit_pytorch\n", "import torch.nn as nn\n", "import torch\n", "from vit_pytorch.cait import CaiT\n", "# baseline CaiT models pretrained on ImageNet\n", "v = CaiT(\n", " image_size = 32,\n", " patch_size = 32,\n", " num_classes = 10,\n", " dim = 1024,\n", " depth = 12, # depth of transformer for patch to patch attention only\n", " cls_depth = 2, # depth of cross attention of CLS tokens to patch\n", " heads = 16,\n", " mlp_dim = 2048,\n", " dropout = 0.1,\n", " emb_dropout = 0.1,\n", " layer_dropout = 0.05 # randomly dropout 5% of the layers\n", ")\n", "\n", "class Patch_embeded(nn.Module):\n", " def __init__(self, image_size=224, patch_size=16, embed_dim=768, in_channel=3):\n", " super().__init__()\n", " self.patch_size = patch_size\n", " self.embed_dim = embed_dim\n", " self.in_channel = in_channel\n", " self.image_size = image_size\n", " self.num_patches = (self.image_size // self.patch_size) ** 2\n", " self.proj = nn.Conv2d(self.in_channel, self.embed_dim, kernel_size=self.patch_size, stride=self.patch_size)\n", "\n", " self.cls_token = nn.Parameter(torch.zeros(1, 1, self.embed_dim))\n", " self.pos_embed = nn.Parameter(torch.zeros(1, self.num_patches+1, self.embed_dim))\n", " \n", " def forward(self, x):\n", " # x: [B, C, H, W]\n", "\n", " x = self.proj(x)\n", " x = x.flatten(2) # [B, embed_dim, num_patches]\n", " x = x.transpose(1, 2) # [B, num_patches, embed_dim]\n", "\n", " cls_token = self.cls_token.expand(x.shape[0], -1 ,-1)\n", " x = torch.cat((x,cls_token), dim=1) #[B, num_patches+1, embed_dim]\n", " pos_embed = self.pos_embed\n", "\n", " out = x + pos_embed #[B, num_patches+1, embed_dim]\n", " return out\n", "\n", "class attention(nn.Module):\n", " def __init__(self, dim, num_heads = 8, qkv_bias = False):\n", " super().__init__()\n", "\n", " self.num_heads = num_heads\n", " prehead_dim = dim // self.num_heads\n", " self.scale = prehead_dim ** -0.5\n", "\n", " self.qkv = nn.Linear(dim, dim*3, bias=qkv_bias)\n", " self.proj = nn.Linear(dim, dim)\n", "\n", " def forward(self, x):\n", " # x: [B, num_patches+1, embed_dim]\n", " B, num_patches, total_dim = x.shape\n", "\n", " qkv = self.qkv(x) #[B, num_patches+1, 3*embed_dim]\n", "\n", " qkv = qkv.reshape(B, num_patches, 3, self.num_heads, total_dim // self.num_heads) #[B, num_patches+1, 3, num_heads, prehead_dim]\n", "\n", " qkv = qkv.permute(2, 0, 3, 1, 4) #[3, B, num_heads, num_patches+1, prehead_dim]\n", " q, k, v = qkv[0], qkv[1], qkv[2] #[B, num_heads, num_patches+1, prehead_dim]\n", "\n", " atten = (q @ k.transpose(-2,-1)) * self.scale #[B, num_heads, num_patches+1, num_patches+1]\n", " atten = atten.softmax(dim=-1)\n", " atten = atten @ v ## [B, num_heads, num_patches+1, prehead_dim]\n", " atten = atten.transpose(1,2) #[B, num_patches+1, num_heads, prehead_dim]\n", " atten = atten.reshape(B, num_patches, total_dim) #[B, num_patches+1, embed_dim]\n", "\n", " out = self.proj(atten)\n", "\n", " return out\n", "\n", "class MLP(nn.Module):\n", " def __init__(self, in_dim, hidden_dim=None, out_dim=None):\n", " super().__init__()\n", " self.fc1 = nn.Linear(in_dim, hidden_dim)\n", " self.actlayer = nn.GELU()\n", " self.fc2 = nn.Linear(hidden_dim, out_dim)\n", "\n", " def forward(self, x):\n", " x = self.fc1(x) #[B, num_patches+1, hidden_dim]\n", " x = self.actlayer(x)\n", " x = self.fc2(x) #[B, num_patches+1, out_dim]\n", " x = self.actlayer(x)\n", "\n", " return x\n", "\n", "class Encoder_block(nn.Module):\n", " def __init__(self,\n", " dim,\n", " num_heads, \n", " mlp_ration=4,\n", " qkv_bias = False,\n", "\n", " ):\n", " super().__init__()\n", "\n", " self.normlayer = nn.LayerNorm(dim)\n", " self.atten = attention(dim, num_heads, qkv_bias=qkv_bias)\n", " self.hidden_dim = int(dim*mlp_ration)\n", " self.mlp = MLP(in_dim = dim, hidden_dim = self.hidden_dim, out_dim=dim)\n", " \n", " def forward(self, x):\n", " x = x + self.atten(self.normlayer(x))\n", " x = x + self.mlp(self.normlayer(x))\n", "\n", " return x\n", "\n", "class Vision_Model(nn.Module):\n", " def __init__(self,\n", " in_channel = 3,\n", " dim=768,\n", " num_heads = 12,\n", " image_size = 224,\n", " patch_szie = 16,\n", " num_classes = 10,\n", " depth = 12,\n", " qkv_bias = True\n", " ):\n", " super().__init__()\n", " self.image_size = image_size\n", " self.patch_size = patch_szie\n", " self.patch_embed = Patch_embeded(image_size= self.image_size, patch_size=self.patch_size, embed_dim=dim, in_channel=in_channel)\n", " self.depth = depth\n", " self.norm = nn.LayerNorm(dim)\n", " self.encoder = nn.Sequential(*[\n", " Encoder_block(dim=dim, num_heads= num_heads, mlp_ration=4, qkv_bias=qkv_bias) for i in range(depth)\n", " ])\n", "\n", " self.head = nn.Linear(dim, num_classes)\n", "\n", " def forward(self,x):\n", "\n", " x = self.patch_embed(x)\n", " x = self.encoder(x)\n", " x = self.norm(x)\n", " x = self.head(x)\n", " return x" ] }, { "cell_type": "code", "execution_count": 48, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "10000张测试集中的准确率为: 99.4%\n" ] } ], "source": [ "import torchvision as tv\n", "import numpy as np\n", "\n", "testset = tv.datasets.ImageFolder(root='./data/cifar/test',transform=tv.transforms.ToTensor())\n", "\n", "\n", "correct = 0 # 预测正确的图片数\n", "total = 0 # 总共的图片数\n", "\n", "\n", "# 由于测试的时候不需要求导，可以暂时关闭autograd，提高速度，节约内存\n", "images = []\n", "labels = []\n", "with torch.no_grad():\n", " for data in testset:\n", " image, label = data\n", " image = image\n", " label = label\n", " images.append(image)\n", " labels.append(label)\n", " total += 1\n", " # 把images转成tensor\n", " images = torch.stack(images)\n", " outputs = v(images)\n", "\n", "for i in range(len(outputs)):\n", " predicted = np.argmax(outputs[i])\n", " if predicted == labels[i]:\n", " correct += 1\n", "\n", "print('10000张测试集中的准确率为: %d %%' % (100 * correct / total))" ] } ], "metadata": { "interpreter": { "hash": "c65094271e18768ba853ee00da8ef0989016e0be5b6a976cebf700525936202a" }, "kernelspec": { "display_name": "Python 3.8.12 ('torch')", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.8.12" } }, "nbformat": 4, "nbformat_minor": 4 }