training new.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "medium-compensation",
   "metadata": {},
   "outputs": [],
   "source": [
    "#from functools import partial\n",
    "\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from tensorflow.keras.callbacks import EarlyStopping, TensorBoard\n",
    "from tensorflow.keras.optimizers import Adam\n",
    "\n",
    "#from nets import freeze_layers, get_model_from_name\n",
    "from utils.callbacks import (ExponentDecayScheduler, LossHistory,\n",
    "                             ModelCheckpoint)\n",
    "from utils.dataloader import ClsDatasets\n",
    "from utils.utils import get_classes\n",
    "#from utils.utils_fit import fit_one_epoch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "conditional-interest",
   "metadata": {},
   "outputs": [],
   "source": [
    "gpus = tf.config.experimental.list_physical_devices(device_type='GPU')\n",
    "for gpu in gpus:\n",
    "    tf.config.experimental.set_memory_growth(gpu, True)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "chubby-prefix",
   "metadata": {},
   "source": [
    "# 获取数据路径，并划分训练集和测试集数量"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "adapted-workshop",
   "metadata": {},
   "source": [
    "# 测试集的地址，我要把全部合在一起"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "supposed-grace",
   "metadata": {},
   "outputs": [],
   "source": [
    "annotation_path = \"path_txt//test_stable/train_val.txt\"\n",
    "\n",
    "with open(annotation_path, \"r\") as f:\n",
    "    lines_test = f.readlines()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "average-equation",
   "metadata": {},
   "outputs": [],
   "source": [
    "lines_test_0 = lines_test[:436]\n",
    "lines_test_1 = lines_test[436:556]\n",
    "lines_test_2 = lines_test[556:752]\n",
    "lines_test_3 = lines_test[752:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "aboriginal-booth",
   "metadata": {},
   "outputs": [],
   "source": [
    "np.random.seed(10101)\n",
    "np.random.shuffle(lines_test_0)\n",
    "np.random.shuffle(lines_test_1)\n",
    "np.random.shuffle(lines_test_2)\n",
    "np.random.shuffle(lines_test_3)\n",
    "np.random.seed(None)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "about-scroll",
   "metadata": {},
   "source": [
    "# 训练级的地址"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "precious-hypothesis",
   "metadata": {},
   "outputs": [],
   "source": [
    "classes_path = 'path_txt/cls_classes.txt'\n",
    "class_names, num_classes = get_classes(classes_path) #得到类别与种类个数"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "answering-garden",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['Retinopathy_grade_0',\n",
       " 'Retinopathy_grade_1',\n",
       " 'Retinopathy_grade_2',\n",
       " 'Retinopathy_grade_3']"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "class_names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "israeli-bailey",
   "metadata": {},
   "outputs": [],
   "source": [
    "annotation_path = \"path_txt/test_stable/train_val.txt\"\n",
    "val_split       = 0.2\n",
    "\n",
    "with open(annotation_path, \"r\") as f:\n",
    "    lines = f.readlines()\n",
    "    \n",
    "# np.random.seed(10101)\n",
    "# np.random.shuffle(lines)\n",
    "# np.random.seed(None)\n",
    "# num_val     = int(len(lines) * val_split)\n",
    "# num_train   = len(lines) - num_val"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "complimentary-burton",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "3338"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(lines)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "developmental-monitor",
   "metadata": {},
   "source": [
    "# 不同种类的地址"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "trying-insertion",
   "metadata": {},
   "outputs": [],
   "source": [
    "lines_0 = lines[:1050]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "governing-belief",
   "metadata": {},
   "outputs": [],
   "source": [
    "#lines_0[-1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "exact-jones",
   "metadata": {},
   "outputs": [],
   "source": [
    "lines_1 = lines[1050:1737]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "artistic-penguin",
   "metadata": {},
   "outputs": [],
   "source": [
    "#lines_1[-1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "boring-plaintiff",
   "metadata": {},
   "outputs": [],
   "source": [
    "lines_2 = lines[1737:2525]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "collaborative-condition",
   "metadata": {},
   "outputs": [],
   "source": [
    "#lines_2[-1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "further-electronics",
   "metadata": {},
   "outputs": [],
   "source": [
    "lines_3 = lines[2525:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "informative-amount",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "3338"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(lines_0) + len(lines_1) +len(lines_2) + len(lines_3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "defined-campbell",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "sixth-document",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "\n",
    "np.random.seed(10101)\n",
    "np.random.shuffle(lines_0)\n",
    "np.random.shuffle(lines_1)\n",
    "np.random.shuffle(lines_2)\n",
    "np.random.shuffle(lines_3)\n",
    "np.random.seed(None)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "absent-cookie",
   "metadata": {},
   "outputs": [],
   "source": [
    "#lines = [lines_0, lines_1, lines_2, lines_3]\n",
    "lines = [lines_0, lines_1, lines_2, lines_3]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "functional-manchester",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_path = []\n",
    "val_path = []\n",
    "\n",
    "for i in range(4):\n",
    "    line = lines[i]\n",
    "    N = len(line)\n",
    "    num_val     = int(N * val_split)\n",
    "    num_train   = N - num_val\n",
    "    for j in range(N):\n",
    "        if j+1<=num_val:\n",
    "            val_path.append(line[j])\n",
    "        else:\n",
    "            train_path.append(line[j])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "compact-brass",
   "metadata": {},
   "outputs": [],
   "source": [
    "num_train = len(train_path)\n",
    "num_val = len(val_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "sealed-philip",
   "metadata": {},
   "outputs": [],
   "source": [
    "np.random.seed(2022)\n",
    "np.random.shuffle(train_path)\n",
    "np.random.shuffle(val_path)\n",
    "np.random.seed(None)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "refined-leadership",
   "metadata": {},
   "outputs": [],
   "source": [
    "annotation_path = \"path_txt/test_stable/test.txt\"\n",
    "\n",
    "with open(annotation_path, \"r\") as f:\n",
    "    test_path = f.readlines()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "varying-modification",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "2672"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "num_train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "critical-greene",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "666"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "num_val"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "sustained-wilson",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "832"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(test_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "molecular-proxy",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "further-honor",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_path = []\n",
    "val_path = []\n",
    "\n",
    "for i in range(4):\n",
    "    line = lines[i]\n",
    "    N = len(line)\n",
    "    num_val     = int(N * val_split)\n",
    "    num_train   = N - num_val\n",
    "    for j in range(N):\n",
    "        if j+1<=num_val:\n",
    "            val_path.append(line[j])\n",
    "        else:\n",
    "            train_path.append(line[j])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "worthy-certificate",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "2506"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(train_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "written-ancient",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "832"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(val_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "characteristic-prior",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "832"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(test_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "complete-apparatus",
   "metadata": {},
   "outputs": [],
   "source": [
    "num_train = len(train_path)\n",
    "num_val = len(val_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "proud-deputy",
   "metadata": {},
   "outputs": [],
   "source": [
    "np.random.seed(2022)\n",
    "np.random.shuffle(train_path)\n",
    "np.random.shuffle(val_path)\n",
    "np.random.seed(None)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "shaped-accountability",
   "metadata": {},
   "source": [
    "# 模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "scheduled-pioneer",
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.layers import (Dense,Input,GlobalAveragePooling2D)\n",
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.keras import applications\n",
    "\n",
    "def resnet101(input_shape=None):\n",
    "    net = applications.ResNet101(include_top=False,input_shape=input_shape,weights=None)\n",
    "\n",
    "    return net\n",
    "\n",
    "def ResNet101(input_shape=[224,224,3],classes=4):\n",
    "    img_input = Input(input_shape)\n",
    "\n",
    "    x = resnet101(input_shape)(img_input)\n",
    "    x = GlobalAveragePooling2D()(x)\n",
    "    x = Dense(256)(x)\n",
    "    x = Dense(256)(x)\n",
    "    x = Dense(classes,activation='softmax', name='fc1000')(x)\n",
    "    x = Model(img_input, x, name='ResNet101')\n",
    "    return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "homeless-stations",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = ResNet101(input_shape=[224,224,3],classes=4)\n",
    "#model.summary()\n",
    "#model.load_weights('D:\\class_PR\\classification-tf2-main\\logs\\with_dict-ep015-loss0.060-val_loss0.509.h5')\n",
    "#model.load_weights(r'test_acc-0.87-resnet101.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "better-lebanon",
   "metadata": {},
   "outputs": [],
   "source": [
    "model.load_weights('resnet101_four.h5')\n",
    "#model.load_weights('D:\\class_PR\\ju\\weights\\\\test_acc-0.65-balance-resnet101.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "elect-valentine",
   "metadata": {},
   "outputs": [],
   "source": [
    "model.load_weights('D:\\class_PR\\ju\\\\weights\\\\cls_4\\\\4cls_0.88-ep010-loss0.098-val_loss0.244.h5')  #加载这个"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "located-solid",
   "metadata": {},
   "outputs": [],
   "source": [
    "for i in range(2):\n",
    "    model.layers[i].trainable = True"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "composite-semiconductor",
   "metadata": {},
   "source": [
    "# 回调函数"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "cubic-metallic",
   "metadata": {},
   "outputs": [],
   "source": [
    "logging         = TensorBoard(log_dir = 'logs/')\n",
    "checkpoint      = ModelCheckpoint('logs/ep{epoch:03d}-loss{loss:.3f}-val_loss{val_loss:.3f}.h5',\n",
    "                                    monitor='val_loss', save_weights_only=True, save_best_only=False, period=5)\n",
    "reduce_lr       = ExponentDecayScheduler(decay_rate = 0.95, verbose = 1)\n",
    "early_stopping  = EarlyStopping(monitor='val_loss', min_delta=0, patience=10, verbose=1)\n",
    "loss_history    = LossHistory('logs/')\n",
    "#tf.keras.callbacks.ModelCheckpoint()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "enormous-speaker",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 2672 samples, val on 666 samples, with batch size 16.\n"
     ]
    }
   ],
   "source": [
    "batch_size      = 16\n",
    "Lr              = 1e-4\n",
    "Init_Epoch      = 0\n",
    "Epoch           = 100\n",
    "input_shape     = [224, 224]\n",
    "num_workers     = 1\n",
    "\n",
    "epoch_step          = num_train // batch_size\n",
    "epoch_step_val      = num_val // batch_size\n",
    "\n",
    "if epoch_step == 0 or epoch_step_val == 0:\n",
    "    raise ValueError('数据集过小，无法进行训练，请扩充数据集。')\n",
    "            \n",
    "print('Train on {} samples, val on {} samples, with batch size {}.'.format(num_train, num_val, batch_size))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "extraordinary-match",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_dataloader    = ClsDatasets(train_path, input_shape, batch_size, num_classes, train = True)\n",
    "val_dataloader      = ClsDatasets(val_path, input_shape, batch_size, num_classes, train = False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "animated-eagle",
   "metadata": {},
   "source": [
    "# 得到tf的data train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "twenty-piece",
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train = []\n",
    "y_train =[]\n",
    "\n",
    "for x,y in train_dataloader:\n",
    "    x_train.append(x)\n",
    "    y_train.append(np.argmax(y,axis=1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "id": "angry-paint",
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train = np.asarray(x_train)\n",
    "y_train = np.asarray(y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "id": "burning-ozone",
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train = np.squeeze(x_train, 1)\n",
    "y_train = np.squeeze(y_train, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "id": "strong-hearts",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(224, 224, 3)"
      ]
     },
     "execution_count": 56,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x_train[0].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "id": "characteristic-flooring",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "()"
      ]
     },
     "execution_count": 57,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_train[0].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "id": "appreciated-allowance",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data2 = tf.data.Dataset.from_tensor_slices((x_train, y_train))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "id": "listed-prescription",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<TensorSliceDataset shapes: ((224, 224, 3), ()), types: (tf.float32, tf.int64)>"
      ]
     },
     "execution_count": 76,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_data2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "id": "great-rouge",
   "metadata": {},
   "outputs": [],
   "source": [
    "AUTOTUNE = tf.data.experimental.AUTOTUNE\n",
    "\n",
    "train_data2 = train_data2.shuffle(buffer_size=num_train)#.repeat(-1)\n",
    "train_data2 = train_data2.batch(batch_size)\n",
    "train_data2 = train_data2.prefetch(buffer_size=AUTOTUNE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "id": "suitable-postcard",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<PrefetchDataset shapes: ((None, 224, 224, 3), (None,)), types: (tf.float32, tf.int64)>"
      ]
     },
     "execution_count": 78,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_data2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "conventional-aluminum",
   "metadata": {},
   "outputs": [],
   "source": [
    "#for x,y in val_dataloader:\n",
    "#    print(y)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "signed-combining",
   "metadata": {},
   "source": [
    "# 得到tf的data val"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "id": "million-channels",
   "metadata": {},
   "outputs": [],
   "source": [
    "x_val = []\n",
    "y_val =[]\n",
    "\n",
    "for x,y in val_dataloader:\n",
    "    x_val.append(x)\n",
    "    y_val.append(np.argmax(y,axis=1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "id": "front-surgeon",
   "metadata": {},
   "outputs": [],
   "source": [
    "x_val = np.asarray(x_val)\n",
    "y_val = np.asarray(y_val)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "id": "copyrighted-specialist",
   "metadata": {},
   "outputs": [],
   "source": [
    "x_val = np.squeeze(x_val, 1)\n",
    "y_val = np.squeeze(y_val, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "id": "dated-constant",
   "metadata": {},
   "outputs": [],
   "source": [
    "val_data1 = tf.data.Dataset.from_tensor_slices((x_val, y_val))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "id": "pointed-heritage",
   "metadata": {},
   "outputs": [],
   "source": [
    "val_data1 = val_data1.batch(batch_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "id": "driving-executive",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<BatchDataset shapes: ((None, 224, 224, 3), (None,)), types: (tf.float32, tf.int64)>"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "val_data1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "homeless-contractor",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "id": "studied-margin",
   "metadata": {},
   "outputs": [],
   "source": [
    "lr_schedule     = tf.keras.optimizers.schedules.ExponentialDecay(\n",
    "                initial_learning_rate=Lr, decay_steps=epoch_step, decay_rate=0.95, staircase=True\n",
    "            )\n",
    "optimizer       = tf.keras.optimizers.Adam(learning_rate=lr_schedule)\n",
    "\n",
    "\n",
    "model.compile(loss='categorical_crossentropy',optimizer=optimizer,metrics=[\"acc\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "id": "greek-virtue",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n"
     ]
    },
    {
     "ename": "ValueError",
     "evalue": "in user code:\n\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\engine\\training.py:806 train_function  *\n        return step_function(self, iterator)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\engine\\training.py:796 step_function  **\n        outputs = model.distribute_strategy.run(run_step, args=(data,))\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\distribute\\distribute_lib.py:1211 run\n        return self._extended.call_for_each_replica(fn, args=args, kwargs=kwargs)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\distribute\\distribute_lib.py:2585 call_for_each_replica\n        return self._call_for_each_replica(fn, args, kwargs)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\distribute\\distribute_lib.py:2945 _call_for_each_replica\n        return fn(*args, **kwargs)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\engine\\training.py:789 run_step  **\n        outputs = model.train_step(data)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\engine\\training.py:749 train_step\n        y, y_pred, sample_weight, regularization_losses=self.losses)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\engine\\compile_utils.py:204 __call__\n        loss_value = loss_obj(y_t, y_p, sample_weight=sw)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\losses.py:149 __call__\n        losses = ag_call(y_true, y_pred)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\losses.py:253 call  **\n        return ag_fn(y_true, y_pred, **self._fn_kwargs)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\util\\dispatch.py:201 wrapper\n        return target(*args, **kwargs)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\losses.py:1535 categorical_crossentropy\n        return K.categorical_crossentropy(y_true, y_pred, from_logits=from_logits)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\util\\dispatch.py:201 wrapper\n        return target(*args, **kwargs)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\backend.py:4687 categorical_crossentropy\n        target.shape.assert_is_compatible_with(output.shape)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\framework\\tensor_shape.py:1134 assert_is_compatible_with\n        raise ValueError(\"Shapes %s and %s are incompatible\" % (self, other))\n\n    ValueError: Shapes (None, 1) and (None, 4) are incompatible\n",
     "output_type": "error",
     "traceback": [
      "\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[1;31mValueError\u001b[0m                                Traceback (most recent call last)",
      "\u001b[1;32m<ipython-input-79-2a4912abc9fa>\u001b[0m in \u001b[0;36m<module>\u001b[1;34m\u001b[0m\n\u001b[0;32m      6\u001b[0m         \u001b[0muse_multiprocessing\u001b[0m \u001b[1;33m=\u001b[0m \u001b[1;32mTrue\u001b[0m \u001b[1;32mif\u001b[0m \u001b[0mnum_workers\u001b[0m \u001b[1;33m>\u001b[0m \u001b[1;36m1\u001b[0m \u001b[1;32melse\u001b[0m \u001b[1;32mFalse\u001b[0m\u001b[1;33m,\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m      7\u001b[0m         \u001b[0mworkers\u001b[0m             \u001b[1;33m=\u001b[0m \u001b[0mnum_workers\u001b[0m\u001b[1;33m,\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m----> 8\u001b[1;33m         \u001b[0mcallbacks\u001b[0m           \u001b[1;33m=\u001b[0m \u001b[1;33m[\u001b[0m\u001b[0mlogging\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mcheckpoint\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mearly_stopping\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mloss_history\u001b[0m\u001b[1;33m]\u001b[0m\u001b[1;33m,\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m      9\u001b[0m     \u001b[1;31m#class_weight= cls_wt_dict\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m     10\u001b[0m     )\n",
      "\u001b[1;32md:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\engine\\training.py\u001b[0m in \u001b[0;36m_method_wrapper\u001b[1;34m(self, *args, **kwargs)\u001b[0m\n\u001b[0;32m    106\u001b[0m   \u001b[1;32mdef\u001b[0m \u001b[0m_method_wrapper\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mself\u001b[0m\u001b[1;33m,\u001b[0m \u001b[1;33m*\u001b[0m\u001b[0margs\u001b[0m\u001b[1;33m,\u001b[0m \u001b[1;33m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    107\u001b[0m     \u001b[1;32mif\u001b[0m \u001b[1;32mnot\u001b[0m \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_in_multi_worker_mode\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m:\u001b[0m  \u001b[1;31m# pylint: disable=protected-access\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m--> 108\u001b[1;33m       \u001b[1;32mreturn\u001b[0m \u001b[0mmethod\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mself\u001b[0m\u001b[1;33m,\u001b[0m \u001b[1;33m*\u001b[0m\u001b[0margs\u001b[0m\u001b[1;33m,\u001b[0m \u001b[1;33m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m    109\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    110\u001b[0m     \u001b[1;31m# Running inside `run_distribute_coordinator` already.\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32md:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\engine\\training.py\u001b[0m in \u001b[0;36mfit\u001b[1;34m(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_batch_size, validation_freq, max_queue_size, workers, use_multiprocessing)\u001b[0m\n\u001b[0;32m   1096\u001b[0m                 batch_size=batch_size):\n\u001b[0;32m   1097\u001b[0m               \u001b[0mcallbacks\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mon_train_batch_begin\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mstep\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m-> 1098\u001b[1;33m               \u001b[0mtmp_logs\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mtrain_function\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0miterator\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m   1099\u001b[0m               \u001b[1;32mif\u001b[0m \u001b[0mdata_handler\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mshould_sync\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1100\u001b[0m                 \u001b[0mcontext\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0masync_wait\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32md:\\miniconda\\lib\\site-packages\\tensorflow\\python\\eager\\def_function.py\u001b[0m in \u001b[0;36m__call__\u001b[1;34m(self, *args, **kwds)\u001b[0m\n\u001b[0;32m    778\u001b[0m       \u001b[1;32melse\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    779\u001b[0m         \u001b[0mcompiler\u001b[0m \u001b[1;33m=\u001b[0m \u001b[1;34m\"nonXla\"\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m--> 780\u001b[1;33m         \u001b[0mresult\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_call\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m*\u001b[0m\u001b[0margs\u001b[0m\u001b[1;33m,\u001b[0m \u001b[1;33m**\u001b[0m\u001b[0mkwds\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m    781\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    782\u001b[0m       \u001b[0mnew_tracing_count\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_get_tracing_count\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32md:\\miniconda\\lib\\site-packages\\tensorflow\\python\\eager\\def_function.py\u001b[0m in \u001b[0;36m_call\u001b[1;34m(self, *args, **kwds)\u001b[0m\n\u001b[0;32m    821\u001b[0m       \u001b[1;31m# This is the first call of __call__, so we have to initialize.\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    822\u001b[0m       \u001b[0minitializers\u001b[0m \u001b[1;33m=\u001b[0m \u001b[1;33m[\u001b[0m\u001b[1;33m]\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m--> 823\u001b[1;33m       \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_initialize\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0margs\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mkwds\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0madd_initializers_to\u001b[0m\u001b[1;33m=\u001b[0m\u001b[0minitializers\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m    824\u001b[0m     \u001b[1;32mfinally\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    825\u001b[0m       \u001b[1;31m# At this point we know that the initialization is complete (or less\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32md:\\miniconda\\lib\\site-packages\\tensorflow\\python\\eager\\def_function.py\u001b[0m in \u001b[0;36m_initialize\u001b[1;34m(self, args, kwds, add_initializers_to)\u001b[0m\n\u001b[0;32m    695\u001b[0m     self._concrete_stateful_fn = (\n\u001b[0;32m    696\u001b[0m         self._stateful_fn._get_concrete_function_internal_garbage_collected(  # pylint: disable=protected-access\n\u001b[1;32m--> 697\u001b[1;33m             *args, **kwds))\n\u001b[0m\u001b[0;32m    698\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    699\u001b[0m     \u001b[1;32mdef\u001b[0m \u001b[0minvalid_creator_scope\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m*\u001b[0m\u001b[0munused_args\u001b[0m\u001b[1;33m,\u001b[0m \u001b[1;33m**\u001b[0m\u001b[0munused_kwds\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32md:\\miniconda\\lib\\site-packages\\tensorflow\\python\\eager\\function.py\u001b[0m in \u001b[0;36m_get_concrete_function_internal_garbage_collected\u001b[1;34m(self, *args, **kwargs)\u001b[0m\n\u001b[0;32m   2853\u001b[0m       \u001b[0margs\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mkwargs\u001b[0m \u001b[1;33m=\u001b[0m \u001b[1;32mNone\u001b[0m\u001b[1;33m,\u001b[0m \u001b[1;32mNone\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   2854\u001b[0m     \u001b[1;32mwith\u001b[0m \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_lock\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m-> 2855\u001b[1;33m       \u001b[0mgraph_function\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0m_\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0m_\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_maybe_define_function\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0margs\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mkwargs\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m   2856\u001b[0m     \u001b[1;32mreturn\u001b[0m \u001b[0mgraph_function\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   2857\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32md:\\miniconda\\lib\\site-packages\\tensorflow\\python\\eager\\function.py\u001b[0m in \u001b[0;36m_maybe_define_function\u001b[1;34m(self, args, kwargs)\u001b[0m\n\u001b[0;32m   3211\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   3212\u001b[0m       \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_function_cache\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mmissed\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0madd\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mcall_context_key\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m-> 3213\u001b[1;33m       \u001b[0mgraph_function\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_create_graph_function\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0margs\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mkwargs\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m   3214\u001b[0m       \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_function_cache\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mprimary\u001b[0m\u001b[1;33m[\u001b[0m\u001b[0mcache_key\u001b[0m\u001b[1;33m]\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mgraph_function\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   3215\u001b[0m       \u001b[1;32mreturn\u001b[0m \u001b[0mgraph_function\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0margs\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mkwargs\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32md:\\miniconda\\lib\\site-packages\\tensorflow\\python\\eager\\function.py\u001b[0m in \u001b[0;36m_create_graph_function\u001b[1;34m(self, args, kwargs, override_flat_arg_shapes)\u001b[0m\n\u001b[0;32m   3073\u001b[0m             \u001b[0marg_names\u001b[0m\u001b[1;33m=\u001b[0m\u001b[0marg_names\u001b[0m\u001b[1;33m,\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   3074\u001b[0m             \u001b[0moverride_flat_arg_shapes\u001b[0m\u001b[1;33m=\u001b[0m\u001b[0moverride_flat_arg_shapes\u001b[0m\u001b[1;33m,\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m-> 3075\u001b[1;33m             capture_by_value=self._capture_by_value),\n\u001b[0m\u001b[0;32m   3076\u001b[0m         \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_function_attributes\u001b[0m\u001b[1;33m,\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   3077\u001b[0m         \u001b[0mfunction_spec\u001b[0m\u001b[1;33m=\u001b[0m\u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mfunction_spec\u001b[0m\u001b[1;33m,\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32md:\\miniconda\\lib\\site-packages\\tensorflow\\python\\framework\\func_graph.py\u001b[0m in \u001b[0;36mfunc_graph_from_py_func\u001b[1;34m(name, python_func, args, kwargs, signature, func_graph, autograph, autograph_options, add_control_dependencies, arg_names, op_return_value, collections, capture_by_value, override_flat_arg_shapes)\u001b[0m\n\u001b[0;32m    984\u001b[0m         \u001b[0m_\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0moriginal_func\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mtf_decorator\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0munwrap\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mpython_func\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    985\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m--> 986\u001b[1;33m       \u001b[0mfunc_outputs\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mpython_func\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m*\u001b[0m\u001b[0mfunc_args\u001b[0m\u001b[1;33m,\u001b[0m \u001b[1;33m**\u001b[0m\u001b[0mfunc_kwargs\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m    987\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    988\u001b[0m       \u001b[1;31m# invariant: `func_outputs` contains only Tensors, CompositeTensors,\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32md:\\miniconda\\lib\\site-packages\\tensorflow\\python\\eager\\def_function.py\u001b[0m in \u001b[0;36mwrapped_fn\u001b[1;34m(*args, **kwds)\u001b[0m\n\u001b[0;32m    598\u001b[0m         \u001b[1;31m# __wrapped__ allows AutoGraph to swap in a converted function. We give\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    599\u001b[0m         \u001b[1;31m# the function a weak reference to itself to avoid a reference cycle.\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m--> 600\u001b[1;33m         \u001b[1;32mreturn\u001b[0m \u001b[0mweak_wrapped_fn\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m__wrapped__\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m*\u001b[0m\u001b[0margs\u001b[0m\u001b[1;33m,\u001b[0m \u001b[1;33m**\u001b[0m\u001b[0mkwds\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m    601\u001b[0m     \u001b[0mweak_wrapped_fn\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mweakref\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mref\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mwrapped_fn\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    602\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32md:\\miniconda\\lib\\site-packages\\tensorflow\\python\\framework\\func_graph.py\u001b[0m in \u001b[0;36mwrapper\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m    971\u001b[0m           \u001b[1;32mexcept\u001b[0m \u001b[0mException\u001b[0m \u001b[1;32mas\u001b[0m \u001b[0me\u001b[0m\u001b[1;33m:\u001b[0m  \u001b[1;31m# pylint:disable=broad-except\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    972\u001b[0m             \u001b[1;32mif\u001b[0m \u001b[0mhasattr\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0me\u001b[0m\u001b[1;33m,\u001b[0m \u001b[1;34m\"ag_error_metadata\"\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m--> 973\u001b[1;33m               \u001b[1;32mraise\u001b[0m \u001b[0me\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mag_error_metadata\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mto_exception\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0me\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m    974\u001b[0m             \u001b[1;32melse\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    975\u001b[0m               \u001b[1;32mraise\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;31mValueError\u001b[0m: in user code:\n\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\engine\\training.py:806 train_function  *\n        return step_function(self, iterator)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\engine\\training.py:796 step_function  **\n        outputs = model.distribute_strategy.run(run_step, args=(data,))\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\distribute\\distribute_lib.py:1211 run\n        return self._extended.call_for_each_replica(fn, args=args, kwargs=kwargs)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\distribute\\distribute_lib.py:2585 call_for_each_replica\n        return self._call_for_each_replica(fn, args, kwargs)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\distribute\\distribute_lib.py:2945 _call_for_each_replica\n        return fn(*args, **kwargs)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\engine\\training.py:789 run_step  **\n        outputs = model.train_step(data)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\engine\\training.py:749 train_step\n        y, y_pred, sample_weight, regularization_losses=self.losses)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\engine\\compile_utils.py:204 __call__\n        loss_value = loss_obj(y_t, y_p, sample_weight=sw)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\losses.py:149 __call__\n        losses = ag_call(y_true, y_pred)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\losses.py:253 call  **\n        return ag_fn(y_true, y_pred, **self._fn_kwargs)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\util\\dispatch.py:201 wrapper\n        return target(*args, **kwargs)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\losses.py:1535 categorical_crossentropy\n        return K.categorical_crossentropy(y_true, y_pred, from_logits=from_logits)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\util\\dispatch.py:201 wrapper\n        return target(*args, **kwargs)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\keras\\backend.py:4687 categorical_crossentropy\n        target.shape.assert_is_compatible_with(output.shape)\n    d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\framework\\tensor_shape.py:1134 assert_is_compatible_with\n        raise ValueError(\"Shapes %s and %s are incompatible\" % (self, other))\n\n    ValueError: Shapes (None, 1) and (None, 4) are incompatible\n"
     ]
    }
   ],
   "source": [
    "model.fit(train_data2,\n",
    "         validation_data     = val_data1,\n",
    "        validation_steps    = epoch_step_val,\n",
    "        epochs              = Epoch,\n",
    "        initial_epoch       = Init_Epoch,\n",
    "        use_multiprocessing = True if num_workers > 1 else False,\n",
    "        workers             = num_workers,\n",
    "        callbacks           = [logging, checkpoint, early_stopping, loss_history],\n",
    "    #class_weight= cls_wt_dict\n",
    "    )"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "understanding-format",
   "metadata": {},
   "source": [
    "# 类别的权重"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "alternate-theorem",
   "metadata": {},
   "outputs": [],
   "source": [
    "#cls_wt_dict = {0: 0.5490867579908676, 1: 1.9552845528455285, 2: 1.2208121827411167, 3: 1.178921568627451}\n",
    "#cls_wt_dict = {0: 0.8481735159817352, 1: 1.5101626016260163, 2: 0.9428934010152284, 3: 0.9105392156862745} #balance\n",
    "cls_wt_dict = {0: 0.91837899543379, 1: 1.0901084010840107, 2: 1.0209390862944163, 3: 0.9859068627450981}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "better-intelligence",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From <ipython-input-31-02dc97d35e21>:14: Model.fit_generator (from tensorflow.python.keras.engine.training) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use Model.fit, which supports generators.\n",
      "Epoch 1/100\n",
      "  1/167 [..............................] - ETA: 0s - loss: 1.8800 - categorical_accuracy: 0.2500WARNING:tensorflow:From d:\\miniconda\\lib\\site-packages\\tensorflow\\python\\ops\\summary_ops_v2.py:1277: stop (from tensorflow.python.eager.profiler) is deprecated and will be removed after 2020-07-01.\n",
      "Instructions for updating:\n",
      "use `tf.profiler.experimental.stop` instead.\n",
      "  2/167 [..............................] - ETA: 1:37 - loss: 1.7578 - categorical_accuracy: 0.3125WARNING:tensorflow:Callbacks method `on_train_batch_end` is slow compared to the batch time (batch time: 0.2554s vs `on_train_batch_end` time: 0.9177s). Check your callbacks.\n",
      "167/167 [==============================] - ETA: 0s - loss: 1.1767 - categorical_accuracy: 0.5228Setting learning rate to 9.499999760009814e-05.\n",
      "167/167 [==============================] - 144s 863ms/step - loss: 1.1767 - categorical_accuracy: 0.5228 - val_loss: 1.7318 - val_categorical_accuracy: 0.2088\n",
      "Epoch 2/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.8693 - categorical_accuracy: 0.6261Setting learning rate to 9.02499959920533e-05.\n",
      "167/167 [==============================] - 117s 700ms/step - loss: 0.8693 - categorical_accuracy: 0.6261 - val_loss: 1.4247 - val_categorical_accuracy: 0.3110\n",
      "Epoch 3/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.7768 - categorical_accuracy: 0.6692Setting learning rate to 8.573749619245064e-05.\n",
      "167/167 [==============================] - 120s 716ms/step - loss: 0.7768 - categorical_accuracy: 0.6692 - val_loss: 1.4623 - val_categorical_accuracy: 0.3628\n",
      "Epoch 4/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.7050 - categorical_accuracy: 0.6991Setting learning rate to 8.145062311086804e-05.\n",
      "167/167 [==============================] - 121s 726ms/step - loss: 0.7050 - categorical_accuracy: 0.6991 - val_loss: 1.7858 - val_categorical_accuracy: 0.4878\n",
      "Epoch 5/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.6326 - categorical_accuracy: 0.7463Setting learning rate to 7.737808919046074e-05.\n",
      "167/167 [==============================] - 123s 734ms/step - loss: 0.6326 - categorical_accuracy: 0.7463 - val_loss: 2.1923 - val_categorical_accuracy: 0.5229\n",
      "Epoch 6/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.5374 - categorical_accuracy: 0.7822Setting learning rate to 7.350918749580159e-05.\n",
      "167/167 [==============================] - 122s 733ms/step - loss: 0.5374 - categorical_accuracy: 0.7822 - val_loss: 0.9708 - val_categorical_accuracy: 0.7012\n",
      "Epoch 7/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.4546 - categorical_accuracy: 0.8211Setting learning rate to 6.983372950344346e-05.\n",
      "167/167 [==============================] - 123s 737ms/step - loss: 0.4546 - categorical_accuracy: 0.8211 - val_loss: 0.7090 - val_categorical_accuracy: 0.7393\n",
      "Epoch 8/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.3985 - categorical_accuracy: 0.8417Setting learning rate to 6.634204510191921e-05.\n",
      "167/167 [==============================] - 124s 742ms/step - loss: 0.3985 - categorical_accuracy: 0.8417 - val_loss: 0.5491 - val_categorical_accuracy: 0.8003\n",
      "Epoch 9/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.3443 - categorical_accuracy: 0.8746Setting learning rate to 6.302494111878331e-05.\n",
      "167/167 [==============================] - 124s 741ms/step - loss: 0.3443 - categorical_accuracy: 0.8746 - val_loss: 0.5350 - val_categorical_accuracy: 0.8308\n",
      "Epoch 10/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.3219 - categorical_accuracy: 0.8802Setting learning rate to 5.9873694408452134e-05.\n",
      "167/167 [==============================] - 125s 746ms/step - loss: 0.3219 - categorical_accuracy: 0.8802 - val_loss: 0.5508 - val_categorical_accuracy: 0.8293\n",
      "Epoch 11/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.2636 - categorical_accuracy: 0.9046Setting learning rate to 5.68800103792455e-05.\n",
      "167/167 [==============================] - 124s 744ms/step - loss: 0.2636 - categorical_accuracy: 0.9046 - val_loss: 0.5679 - val_categorical_accuracy: 0.8262\n",
      "Epoch 12/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.2233 - categorical_accuracy: 0.9113Setting learning rate to 5.4036009169067255e-05.\n",
      "167/167 [==============================] - 125s 746ms/step - loss: 0.2233 - categorical_accuracy: 0.9113 - val_loss: 0.7372 - val_categorical_accuracy: 0.8064\n",
      "Epoch 13/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.1991 - categorical_accuracy: 0.9229Setting learning rate to 5.133420836500591e-05.\n",
      "167/167 [==============================] - 125s 746ms/step - loss: 0.1991 - categorical_accuracy: 0.9229 - val_loss: 0.6015 - val_categorical_accuracy: 0.8445\n",
      "Epoch 14/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.1477 - categorical_accuracy: 0.9454Setting learning rate to 4.876749881077558e-05.\n",
      "167/167 [==============================] - 124s 745ms/step - loss: 0.1477 - categorical_accuracy: 0.9454 - val_loss: 0.5881 - val_categorical_accuracy: 0.8567\n",
      "Epoch 15/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.1358 - categorical_accuracy: 0.9517Setting learning rate to 4.63291238702368e-05.\n",
      "167/167 [==============================] - 125s 750ms/step - loss: 0.1358 - categorical_accuracy: 0.9517 - val_loss: 0.6671 - val_categorical_accuracy: 0.8384\n",
      "Epoch 16/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.1161 - categorical_accuracy: 0.9562Setting learning rate to 4.4012669059156905e-05.\n",
      "167/167 [==============================] - 125s 748ms/step - loss: 0.1161 - categorical_accuracy: 0.9562 - val_loss: 0.7587 - val_categorical_accuracy: 0.8415\n",
      "Epoch 17/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.1141 - categorical_accuracy: 0.9603Setting learning rate to 4.181203439657111e-05.\n",
      "167/167 [==============================] - 125s 751ms/step - loss: 0.1141 - categorical_accuracy: 0.9603 - val_loss: 0.6908 - val_categorical_accuracy: 0.8232\n",
      "Epoch 18/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.1034 - categorical_accuracy: 0.9644Setting learning rate to 3.9721430948702614e-05.\n",
      "167/167 [==============================] - 126s 752ms/step - loss: 0.1034 - categorical_accuracy: 0.9644 - val_loss: 0.6978 - val_categorical_accuracy: 0.8354\n",
      "Epoch 19/100\n",
      "167/167 [==============================] - ETA: 0s - loss: 0.0969 - categorical_accuracy: 0.9663Setting learning rate to 3.773536009248346e-05.\n",
      "167/167 [==============================] - 125s 748ms/step - loss: 0.0969 - categorical_accuracy: 0.9663 - val_loss: 0.6387 - val_categorical_accuracy: 0.8399\n",
      "Epoch 00019: early stopping\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x14565526148>"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.compile(loss = 'categorical_crossentropy',\n",
    "                    optimizer = Adam(lr = Lr),\n",
    "                    metrics = ['categorical_accuracy'])\n",
    "            \n",
    "model.fit_generator(\n",
    "    generator           = train_dataloader,\n",
    "    steps_per_epoch     = epoch_step,\n",
    "    validation_data     = val_dataloader,\n",
    "    validation_steps    = epoch_step_val,\n",
    "    epochs              = Epoch,\n",
    "    initial_epoch       = Init_Epoch,\n",
    "    use_multiprocessing = True if num_workers > 1 else False,\n",
    "    workers             = num_workers,\n",
    "    callbacks           = [logging, checkpoint, reduce_lr, early_stopping, loss_history]\n",
    "    #class_weight= cls_wt_dict\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "actual-arabic",
   "metadata": {},
   "source": [
    "# 下面是画混淆矩阵用的"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "acknowledged-citizenship",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import confusion_matrix\n",
    "from sklearn.utils.multiclass import unique_labels\n",
    "from matplotlib import pyplot as plt\n",
    "%matplotlib inline\n",
    "\n",
    "# borrowed from scikit learn  这个是为了画混淆矩阵\n",
    "def plot_confusion_matrix(y_true, y_pred, classes,\n",
    "                          normalize=False,\n",
    "                          title=None,\n",
    "                          cmap=plt.cm.Blues):\n",
    "    \"\"\"\n",
    "    This function prints and plots the confusion matrix.\n",
    "    Normalization can be applied by setting `normalize=True`.\n",
    "    \"\"\"\n",
    "    if not title:\n",
    "        if normalize:\n",
    "            title = 'Normalized confusion matrix'\n",
    "        else:\n",
    "            title = 'Confusion matrix, without normalization'\n",
    "\n",
    "    # Compute confusion matrix\n",
    "    cm = confusion_matrix(y_true, y_pred)\n",
    "    # Only use the labels that appear in the data\n",
    "    classes = classes[unique_labels(y_true, y_pred)]\n",
    "    if normalize:\n",
    "        cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n",
    "        print(\"Normalized confusion matrix\")\n",
    "    else:\n",
    "        print('Confusion matrix, without normalization')\n",
    "\n",
    "    print(cm)\n",
    "\n",
    "    fig, ax = plt.subplots()\n",
    "    im = ax.imshow(cm, interpolation='nearest', cmap=cmap)\n",
    "    ax.figure.colorbar(im, ax=ax)\n",
    "    # We want to show all ticks...\n",
    "    ax.set(xticks=np.arange(cm.shape[1]),\n",
    "           yticks=np.arange(cm.shape[0]),\n",
    "           # ... and label them with the respective list entries\n",
    "           xticklabels=classes, yticklabels=classes,\n",
    "           title=title,\n",
    "           ylabel='True label',\n",
    "           xlabel='Predicted label')\n",
    "\n",
    "    # Rotate the tick labels and set their alignment.\n",
    "    plt.setp(ax.get_xticklabels(), rotation=45, ha=\"right\",\n",
    "             rotation_mode=\"anchor\")\n",
    "\n",
    "    # Loop over data dimensions and create text annotations.\n",
    "    fmt = '.2f' if normalize else 'd'\n",
    "    thresh = cm.max() / 2.\n",
    "    for i in range(cm.shape[0]):\n",
    "        for j in range(cm.shape[1]):\n",
    "            ax.text(j, i, format(cm[i, j], fmt),\n",
    "                    ha=\"center\", va=\"center\",\n",
    "                    color=\"white\" if cm[i, j] > thresh else \"black\")\n",
    "    fig.tight_layout()\n",
    "    return ax"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "mental-controversy",
   "metadata": {},
   "source": [
    "# 合在一起后的测试"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "thrown-china",
   "metadata": {},
   "outputs": [],
   "source": [
    "annotation_path = \"path_txt/test_stable/test.txt\"\n",
    "\n",
    "with open(annotation_path, \"r\") as f:\n",
    "    test_path = f.readlines()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "practical-therapist",
   "metadata": {},
   "outputs": [],
   "source": [
    "val_for_test = ClsDatasets(test_path, input_shape, 1, num_classes, train = False)\n",
    "\n",
    "x_val_for_test = []\n",
    "y_val_for_test =[]\n",
    "\n",
    "for x,y in val_for_test:\n",
    "    x_val_for_test.append(x)\n",
    "    y_val_for_test.append(y)\n",
    "\n",
    "x_val_for_test = np.asarray(x_val_for_test)\n",
    "y_val_for_test = np.asarray(y_val_for_test)\n",
    "\n",
    "x_val_for_test = np.squeeze(x_val_for_test, 1)\n",
    "y_val_for_test = np.squeeze(y_val_for_test, 1)\n",
    "\n",
    "y_val_for_test_pred = model.predict(x_val_for_test)\n",
    "\n",
    "Y_pred_val = np.argmax(y_val_for_test_pred,axis=1)\n",
    "Y_actual_val = np.argmax(y_val_for_test,axis=1)\n",
    "\n",
    "#plot_confusion_matrix(Y_actual_val, Y_pred_val, np.array(class_names))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "black-january",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Confusion matrix, without normalization\n",
      "[[199  49  13   1]\n",
      " [ 10 158   2   1]\n",
      " [  9  17 162   8]\n",
      " [  2   3  16 182]]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<AxesSubplot:title={'center':'Confusion matrix, without normalization'}, xlabel='Predicted label', ylabel='True label'>"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_confusion_matrix(Y_actual_val, Y_pred_val, np.array(class_names))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "sixth-weather",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "baking-undergraduate",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "id": "supreme-oxide",
   "metadata": {},
   "source": [
    "# 获得测试集"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "id": "processed-forum",
   "metadata": {},
   "outputs": [],
   "source": [
    "import glob\n",
    "\n",
    "img_path = glob.glob('D:\\\\class_PR\\\\dataset\\\\test\\\\*\\\\*.jpg')\n",
    "#img_path = glob.glob('D:\\\\class_PR\\\\preporessing_test_data\\\\*\\\\*.jpg')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "id": "emerging-canon",
   "metadata": {},
   "outputs": [],
   "source": [
    "#img_path"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "id": "royal-trust",
   "metadata": {},
   "outputs": [],
   "source": [
    "label_name = [name.split('\\\\')[-2] for name in img_path]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "id": "distributed-heavy",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "952"
      ]
     },
     "execution_count": 100,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(label_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "id": "facial-demonstration",
   "metadata": {},
   "outputs": [],
   "source": [
    "label_to_index = dict((name, i) for (i, name) in enumerate(class_names))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "id": "roman-adelaide",
   "metadata": {},
   "outputs": [],
   "source": [
    "#label_to_index"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "id": "latest-restoration",
   "metadata": {},
   "outputs": [],
   "source": [
    "all_labels = [label_to_index.get(la) for la in label_name]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "id": "billion-salvation",
   "metadata": {},
   "outputs": [],
   "source": [
    "from PIL import Image\n",
    "\n",
    "# def preprocess_image(image_path, desired_size=224):\n",
    "#     im = Image.open(image_path)\n",
    "#     im = im.resize((desired_size, )*2, resample=Image.BILINEAR)\n",
    "#     return im\n",
    "\n",
    "def preprocess_image(image_path, size):\n",
    "    image = Image.open(image_path)\n",
    "    \n",
    "    iw, ih  = image.size\n",
    "    w, h    = size\n",
    "\n",
    "    scale   = min(w/iw, h/ih)\n",
    "    nw      = int(iw*scale)\n",
    "    nh      = int(ih*scale)\n",
    "\n",
    "    image   = image.resize((nw,nh), Image.BICUBIC)\n",
    "    new_image = Image.new('RGB', size, (128,128,128))\n",
    "    new_image.paste(image, ((w-nw)//2, (h-nh)//2))\n",
    "\n",
    "    return new_image"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "id": "turned-situation",
   "metadata": {},
   "outputs": [],
   "source": [
    "N = len(img_path)\n",
    "x_test = np.empty((N, 224, 224, 3), dtype=np.float32)\n",
    "\n",
    "for i, path in enumerate(img_path):\n",
    "    x_test[i, :, :, :] = preprocess_image(path,[224,224])\n",
    "    \n",
    "#x_test = x_test/255.0\n",
    "x_test /= 127.5\n",
    "x_test -= 1."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "id": "literary-watson",
   "metadata": {},
   "outputs": [],
   "source": [
    "y_test = tf.one_hot(all_labels,num_classes)\n",
    "\n",
    "y_test = np.asarray(y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "id": "distinct-compound",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([1., 0., 0., 0.], dtype=float32)"
      ]
     },
     "execution_count": 107,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_test[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "id": "intensive-greene",
   "metadata": {},
   "outputs": [],
   "source": [
    "y_pred = model.predict(x_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "id": "functional-patio",
   "metadata": {},
   "outputs": [],
   "source": [
    "weight_4class = np.asarray([0.91837899543379, 1.0901084010840107, 1.0209390862944163, 0.9859068627450981], np.float32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "id": "unique-string",
   "metadata": {},
   "outputs": [],
   "source": [
    "y_pred = y_pred * weight_4class"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "id": "experienced-serial",
   "metadata": {},
   "outputs": [],
   "source": [
    "#y_pred[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 116,
   "id": "molecular-listing",
   "metadata": {},
   "outputs": [],
   "source": [
    "Y_pred_hot = np.argmax(y_pred,axis=1)\n",
    "Y_actual_hot = np.argmax(y_test,axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 110,
   "id": "widespread-diary",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "3"
      ]
     },
     "execution_count": 110,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Y_actual_hot[-1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "id": "amber-variance",
   "metadata": {},
   "outputs": [],
   "source": [
    "#Y_pred_hot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "id": "external-combine",
   "metadata": {},
   "outputs": [],
   "source": [
    "#Y_actual_hot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "id": "dying-guinea",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Confusion matrix, without normalization\n",
      "[[342  25  62   7]\n",
      " [ 91   3  23   3]\n",
      " [ 43   6 111  36]\n",
      " [  6   1  33 160]]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<AxesSubplot:title={'center':'Confusion matrix, without normalization'}, xlabel='Predicted label', ylabel='True label'>"
      ]
     },
     "execution_count": 117,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_confusion_matrix(Y_actual_hot, Y_pred_hot, np.array(class_names))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "id": "happy-forge",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12/12 [==============================] - 4s 309ms/step - loss: 0.5119 - categorical_accuracy: 0.8750\n",
      "Test loss: 0.51\n",
      "Test accuracy: 87.5%\n"
     ]
    }
   ],
   "source": [
    "loss, accuracy = model.evaluate(x_test, y_test) #每个batch的\n",
    "print(f\"Test loss: {round(loss, 2)}\")\n",
    "print(f\"Test accuracy: {round(accuracy * 100, 2)}%\")\n",
    "#print(f\"Test top 5 accuracy: {round(top_5_accuracy * 100, 2)}%\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "id": "photographic-hazard",
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save_weights('test_acc-0.63-balance-resnet101.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "integrated-friendship",
   "metadata": {},
   "outputs": [],
   "source": [
    "model.load_weights('logs/ep053-loss0.012-val_loss0.045.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "helpful-router",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}