{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"

CutMix: Regularization Strategy to Train Strong Classifiers\nwith Localizable Features

\n\nThrough this notebook we will try to understand CutMix. \n\nLink of the paper \n* https://arxiv.org/pdf/1905.04899.pdf","metadata":{}},{"cell_type":"markdown","source":"# Import Library","metadata":{"_uuid":"fe76d1d1ded592430e7548feacfa38dc42f085d9"}},{"cell_type":"code","source":"# This Python 3 environment comes with many helpful analytics libraries installed\n# It is defined by the kaggle/python docker image: https://github.com/kaggle/docker-python\n# For example, here's several helpful packages to load in \n\nimport numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\nfrom tqdm.auto import tqdm\nfrom glob import glob\nimport time, gc\nimport cv2\n\nfrom tensorflow import keras\nimport matplotlib.image as mpimg\nfrom keras.preprocessing.image import ImageDataGenerator\nfrom keras.models import Model\nfrom keras.models import clone_model\nfrom keras.layers import Dense,Conv2D,Flatten,MaxPool2D,Dropout,BatchNormalization, Input\nfrom keras.optimizers import Adam, RMSprop\nfrom keras.callbacks import ReduceLROnPlateau\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import confusion_matrix\nimport PIL.Image as Image, PIL.ImageDraw as ImageDraw, PIL.ImageFont as ImageFont\nfrom matplotlib import pyplot as plt\nimport seaborn as sns","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:11:41.267139Z","iopub.execute_input":"2022-11-08T20:11:41.267383Z","iopub.status.idle":"2022-11-08T20:11:43.277798Z","shell.execute_reply.started":"2022-11-08T20:11:41.267331Z","shell.execute_reply":"2022-11-08T20:11:43.276846Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import numpy as np\nimport pandas as pd \nfrom keras.preprocessing.image import ImageDataGenerator, load_img\nfrom keras.utils import to_categorical\nfrom sklearn.model_selection import train_test_split\nimport matplotlib.pyplot as plt\nimport random\nimport os\nprint(os.listdir(\"../input\"))\n","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2022-11-08T20:11:43.282910Z","iopub.execute_input":"2022-11-08T20:11:43.285270Z","iopub.status.idle":"2022-11-08T20:11:43.297499Z","shell.execute_reply.started":"2022-11-08T20:11:43.285214Z","shell.execute_reply":"2022-11-08T20:11:43.296487Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import zipfile\n\n\nzip_files = ['test1', 'train']\n# Will unzip the files so that you can see them..\nfor zip_file in zip_files:\n with zipfile.ZipFile(\"../input/{}.zip\".format(zip_file),\"r\") as z:\n z.extractall(\".\")\n print(\"{} unzipped\".format(zip_file))","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:11:43.300553Z","iopub.execute_input":"2022-11-08T20:11:43.301595Z","iopub.status.idle":"2022-11-08T20:11:59.912424Z","shell.execute_reply.started":"2022-11-08T20:11:43.301541Z","shell.execute_reply":"2022-11-08T20:11:59.911500Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ls","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:11:59.913658Z","iopub.execute_input":"2022-11-08T20:11:59.913955Z","iopub.status.idle":"2022-11-08T20:12:00.861075Z","shell.execute_reply.started":"2022-11-08T20:11:59.913906Z","shell.execute_reply":"2022-11-08T20:12:00.860243Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"pwd","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:12:00.862636Z","iopub.execute_input":"2022-11-08T20:12:00.862971Z","iopub.status.idle":"2022-11-08T20:12:00.872158Z","shell.execute_reply.started":"2022-11-08T20:12:00.862907Z","shell.execute_reply":"2022-11-08T20:12:00.871384Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(os.listdir('../input'))","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:12:00.873932Z","iopub.execute_input":"2022-11-08T20:12:00.874516Z","iopub.status.idle":"2022-11-08T20:12:00.881155Z","shell.execute_reply.started":"2022-11-08T20:12:00.874466Z","shell.execute_reply":"2022-11-08T20:12:00.880203Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Define Constants","metadata":{}},{"cell_type":"code","source":"FAST_RUN = False\nIMAGE_WIDTH=128\nIMAGE_HEIGHT=128\nIMAGE_SIZE=(IMAGE_WIDTH, IMAGE_HEIGHT)\nIMAGE_CHANNELS=3","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:12:00.882636Z","iopub.execute_input":"2022-11-08T20:12:00.883203Z","iopub.status.idle":"2022-11-08T20:12:00.887838Z","shell.execute_reply.started":"2022-11-08T20:12:00.883154Z","shell.execute_reply":"2022-11-08T20:12:00.886843Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Prepare Traning Data","metadata":{"_uuid":"7335a579cc0268fba5d34d6f7558f33c187eedb3"}},{"cell_type":"code","source":"filenames = os.listdir(\"train\")\ncategories = []\nfor filename in filenames:\n category = filename.split('.')[0]\n if category == 'dog':\n categories.append(1)\n else:\n categories.append(0)\n\ndf = pd.DataFrame({\n 'filename': filenames,\n 'category': categories\n})","metadata":{"_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","execution":{"iopub.status.busy":"2022-11-08T20:12:00.889295Z","iopub.execute_input":"2022-11-08T20:12:00.889831Z","iopub.status.idle":"2022-11-08T20:12:00.931376Z","shell.execute_reply.started":"2022-11-08T20:12:00.889782Z","shell.execute_reply":"2022-11-08T20:12:00.930813Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df.head()","metadata":{"_uuid":"915bb9ba7063ab4d5c07c542419ae119003a5f98","execution":{"iopub.status.busy":"2022-11-08T20:12:00.932884Z","iopub.execute_input":"2022-11-08T20:12:00.933302Z","iopub.status.idle":"2022-11-08T20:12:00.955561Z","shell.execute_reply.started":"2022-11-08T20:12:00.933122Z","shell.execute_reply":"2022-11-08T20:12:00.954627Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df.tail()","metadata":{"_uuid":"72bf69e817f67f5a2eaff8561217e22077248553","execution":{"iopub.status.busy":"2022-11-08T20:12:00.956871Z","iopub.execute_input":"2022-11-08T20:12:00.957424Z","iopub.status.idle":"2022-11-08T20:12:00.971398Z","shell.execute_reply.started":"2022-11-08T20:12:00.957114Z","shell.execute_reply":"2022-11-08T20:12:00.970684Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### See Total In count","metadata":{"_uuid":"a999484fc35b73373fafe2253ae9db7ff46fdb90"}},{"cell_type":"code","source":"df['category'].value_counts().plot.bar()","metadata":{"_uuid":"fa26f0bc7a6d835a24989790b20f3c6f32946f45","execution":{"iopub.status.busy":"2022-11-08T20:12:00.972651Z","iopub.execute_input":"2022-11-08T20:12:00.973124Z","iopub.status.idle":"2022-11-08T20:12:01.226591Z","shell.execute_reply.started":"2022-11-08T20:12:00.972893Z","shell.execute_reply":"2022-11-08T20:12:01.225567Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"From our data we have 12000 cats and 12000 dogs","metadata":{"_uuid":"3a08da58107777a1dd05c4a4bf5c484484923cac"}},{"cell_type":"markdown","source":"# See sample image","metadata":{"_uuid":"400a293df3c8499059d9175f3915187074efd971"}},{"cell_type":"code","source":"sample = random.choice(filenames)\nimage = load_img(\"train/\"+sample)\nplt.imshow(image)","metadata":{"_uuid":"602b40f7353871cb161c60b5237f0da0096b2f47","execution":{"iopub.status.busy":"2022-11-08T20:12:01.228027Z","iopub.execute_input":"2022-11-08T20:12:01.228314Z","iopub.status.idle":"2022-11-08T20:12:01.620532Z","shell.execute_reply.started":"2022-11-08T20:12:01.228270Z","shell.execute_reply":"2022-11-08T20:12:01.619751Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Build Model\n\n","metadata":{"_uuid":"b244e6b7715a04fc6df92dd6dfa3d35c473ca600"}},{"cell_type":"markdown","source":"* **Input Layer**: It represent input image data. It will reshape image into single diminsion array. Example your image is 64x64 = 4096, it will convert to (4096,1) array.\n* **Conv Layer**: This layer will extract features from image.\n* **Pooling Layer**: This layerreduce the spatial volume of input image after convolution.\n* **Fully Connected Layer**: It connect the network from a layer to another layer\n* **Output Layer**: It is the predicted values layer. ","metadata":{}},{"cell_type":"code","source":"from keras.models import Sequential\nfrom keras.layers import Conv2D, MaxPooling2D, Dropout, Flatten, Dense, Activation, BatchNormalization\n\nmodel = Sequential()\n\nmodel.add(Conv2D(32, (3, 3), activation='relu', input_shape=(IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_CHANNELS)))\nmodel.add(BatchNormalization())\nmodel.add(MaxPooling2D(pool_size=(2, 2)))\nmodel.add(Dropout(0.25))\n\nmodel.add(Conv2D(64, (3, 3), activation='relu'))\nmodel.add(BatchNormalization())\nmodel.add(MaxPooling2D(pool_size=(2, 2)))\nmodel.add(Dropout(0.25))\n\nmodel.add(Conv2D(128, (3, 3), activation='relu'))\nmodel.add(BatchNormalization())\nmodel.add(MaxPooling2D(pool_size=(2, 2)))\nmodel.add(Dropout(0.25))\n\nmodel.add(Flatten())\nmodel.add(Dense(512, activation='relu'))\nmodel.add(BatchNormalization())\nmodel.add(Dropout(0.5))\nmodel.add(Dense(2, activation='softmax')) # 2 because we have cat and dog classes\n\nmodel.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])\n\nmodel.summary()","metadata":{"_uuid":"8c9f833c1441b657c779844912d0b8028218d454","execution":{"iopub.status.busy":"2022-11-08T20:12:01.621486Z","iopub.execute_input":"2022-11-08T20:12:01.621716Z","iopub.status.idle":"2022-11-08T20:12:04.030015Z","shell.execute_reply.started":"2022-11-08T20:12:01.621673Z","shell.execute_reply":"2022-11-08T20:12:04.027864Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Callbacks","metadata":{"_uuid":"bd496f6c65888a969be3703135b0b03a8a1190c8"}},{"cell_type":"code","source":"from keras.callbacks import EarlyStopping, ReduceLROnPlateau","metadata":{"_uuid":"9aa032f0f6da539d23918890d2d131cc3aac8c7a","execution":{"iopub.status.busy":"2022-11-08T20:12:04.031225Z","iopub.execute_input":"2022-11-08T20:12:04.031519Z","iopub.status.idle":"2022-11-08T20:12:04.039639Z","shell.execute_reply.started":"2022-11-08T20:12:04.031471Z","shell.execute_reply":"2022-11-08T20:12:04.038578Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"**Early Stop**\n\nTo prevent over fitting we will stop the learning after 10 epochs and val_loss value not decreased","metadata":{"_uuid":"76c9ba4fb7f930c96b2c3e0d6b68ed9fa6a4227b"}},{"cell_type":"code","source":"earlystop = EarlyStopping(patience=10)","metadata":{"_uuid":"3421c5ec428da6c0d8cc1184179a9caff1e01d1c","execution":{"iopub.status.busy":"2022-11-08T20:12:04.040916Z","iopub.execute_input":"2022-11-08T20:12:04.041182Z","iopub.status.idle":"2022-11-08T20:12:04.046913Z","shell.execute_reply.started":"2022-11-08T20:12:04.041135Z","shell.execute_reply":"2022-11-08T20:12:04.046108Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"**Learning Rate Reduction**\n\nWe will reduce the learning rate when then accuracy not increase for 2 steps","metadata":{"_uuid":"51d3fe52e911286433cedf6e47332948a253361e"}},{"cell_type":"code","source":"learning_rate_reduction = ReduceLROnPlateau(monitor='val_acc', \n patience=2, \n verbose=1, \n factor=0.5, \n min_lr=0.00001)","metadata":{"_uuid":"8010a5661ad8924d2db24af0f3c00b1593b38901","execution":{"iopub.status.busy":"2022-11-08T20:12:04.048003Z","iopub.execute_input":"2022-11-08T20:12:04.048247Z","iopub.status.idle":"2022-11-08T20:12:04.056738Z","shell.execute_reply.started":"2022-11-08T20:12:04.048203Z","shell.execute_reply":"2022-11-08T20:12:04.055981Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"callbacks = [earlystop, learning_rate_reduction]","metadata":{"_uuid":"a79cc604199469789f183096d863f7248e5f6aab","execution":{"iopub.status.busy":"2022-11-08T20:12:04.057838Z","iopub.execute_input":"2022-11-08T20:12:04.058137Z","iopub.status.idle":"2022-11-08T20:12:04.066462Z","shell.execute_reply.started":"2022-11-08T20:12:04.058090Z","shell.execute_reply":"2022-11-08T20:12:04.065419Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Prepare data","metadata":{}},{"cell_type":"markdown","source":"Because we will use image genaretor `with class_mode=\"categorical\"`. We need to convert column category into string. Then imagenerator will convert it one-hot encoding which is good for our classification. \n\nSo we will convert 1 to dog and 0 to cat","metadata":{}},{"cell_type":"code","source":"df[\"category\"] = df[\"category\"].replace({0: 'cat', 1: 'dog'}) ","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:12:04.067527Z","iopub.execute_input":"2022-11-08T20:12:04.067752Z","iopub.status.idle":"2022-11-08T20:12:04.080790Z","shell.execute_reply.started":"2022-11-08T20:12:04.067695Z","shell.execute_reply":"2022-11-08T20:12:04.080082Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df, validate_df = train_test_split(df, test_size=0.20, random_state=42)\ntrain_df = train_df.reset_index(drop=True)\nvalidate_df = validate_df.reset_index(drop=True)","metadata":{"_uuid":"4eeb7af8dcf02c4ef5ca744c8305c51a2f5cedef","execution":{"iopub.status.busy":"2022-11-08T20:12:04.083873Z","iopub.execute_input":"2022-11-08T20:12:04.084332Z","iopub.status.idle":"2022-11-08T20:12:04.096668Z","shell.execute_reply.started":"2022-11-08T20:12:04.084057Z","shell.execute_reply":"2022-11-08T20:12:04.095719Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df['category'].value_counts().plot.bar()","metadata":{"_uuid":"b84836337441705eda9d2e655665ffa14d9feead","execution":{"iopub.status.busy":"2022-11-08T20:12:04.099669Z","iopub.execute_input":"2022-11-08T20:12:04.099964Z","iopub.status.idle":"2022-11-08T20:12:04.332315Z","shell.execute_reply.started":"2022-11-08T20:12:04.099887Z","shell.execute_reply":"2022-11-08T20:12:04.331406Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"validate_df['category'].value_counts().plot.bar()","metadata":{"_uuid":"19cf03f9a3c39532d6e2d06bd30be49a5afd9d57","execution":{"iopub.status.busy":"2022-11-08T20:12:04.333388Z","iopub.execute_input":"2022-11-08T20:12:04.333796Z","iopub.status.idle":"2022-11-08T20:12:04.582151Z","shell.execute_reply.started":"2022-11-08T20:12:04.333747Z","shell.execute_reply":"2022-11-08T20:12:04.581198Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"total_train = train_df.shape[0]\ntotal_validate = validate_df.shape[0]\nbatch_size=15","metadata":{"_uuid":"ae3dec0361f0443132d0309d3b883ee80070cf9f","execution":{"iopub.status.busy":"2022-11-08T20:12:04.583353Z","iopub.execute_input":"2022-11-08T20:12:04.583836Z","iopub.status.idle":"2022-11-08T20:12:04.587875Z","shell.execute_reply.started":"2022-11-08T20:12:04.583786Z","shell.execute_reply":"2022-11-08T20:12:04.587072Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class CutMixImageDataGenerator():\n def __init__(self, generator1, generator2, img_size, batch_size):\n self.batch_index = 0\n self.samples = generator1.samples\n self.class_indices = generator1.class_indices\n self.generator1 = generator1\n self.generator2 = generator2\n self.img_size = img_size\n self.batch_size = batch_size\n\n def reset_index(self): # Ordering Reset (If Shuffle is True, Shuffle Again)\n self.generator1._set_index_array()\n self.generator2._set_index_array()\n\n def reset(self):\n self.batch_index = 0\n self.generator1.reset()\n self.generator2.reset()\n self.reset_index()\n\n def get_steps_per_epoch(self):\n quotient, remainder = divmod(self.samples, self.batch_size)\n return (quotient + 1) if remainder else quotient\n \n def __len__(self):\n self.get_steps_per_epoch()\n \n# def __iter__(self):\n# return self\n\n def __next__(self):\n if self.batch_index == 0: self.reset()\n\n crt_idx = self.batch_index * self.batch_size\n if self.samples > crt_idx + self.batch_size:\n self.batch_index += 1\n else: # If current index over number of samples\n self.batch_index = 0\n\n reshape_size = self.batch_size\n last_step_start_idx = (self.get_steps_per_epoch()-1) * self.batch_size\n if crt_idx == last_step_start_idx:\n reshape_size = self.samples - last_step_start_idx\n \n X_1, y_1 = self.generator1.next()\n X_2, y_2 = self.generator2.next()\n reshape_size = min(X_1.shape[0], X_2.shape[0])\n X_1 = X_1[:reshape_size]\n y_1 = y_1[:reshape_size]\n \n X_2 = X_2[:reshape_size]\n y_2 = y_2[:reshape_size]\n\n if X_1.shape[0] != X_2.shape[0]:\n print(X_1.shape,X_2.shape, reshape_size)\n \n cut_ratio = np.random.beta(a=1, b=1, size=reshape_size)\n cut_ratio = np.clip(cut_ratio, 0.2, 0.8)\n label_ratio = cut_ratio.reshape(reshape_size, 1)\n cut_img = X_2\n\n X = X_1\n for i in range(reshape_size):\n cut_size = int((self.img_size-1) * cut_ratio[i])\n y1 = random.randint(0, (self.img_size-1) - cut_size)\n x1 = random.randint(0, (self.img_size-1) - cut_size)\n #print(y1, 'this is y1')\n y2 = y1 + cut_size\n x2 = x1 + cut_size\n cut_arr = cut_img[i][y1:y2, x1:x2]\n cutmix_img = X_1[i]\n cutmix_img[y1:y2, x1:x2] = cut_arr\n X[i] = cutmix_img \n y = y_1 * (1 - (label_ratio ** 2)) + y_2 * (label_ratio ** 2)\n\n return X, y","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:12:04.589217Z","iopub.execute_input":"2022-11-08T20:12:04.589766Z","iopub.status.idle":"2022-11-08T20:12:04.614605Z","shell.execute_reply.started":"2022-11-08T20:12:04.589703Z","shell.execute_reply":"2022-11-08T20:12:04.613403Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y1= np.array([1,0])\ny2 = np.array([1,0])\nlr = 0.56038959\ny1*(1-(lr**2)) + y2*(lr**2)","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:12:04.616257Z","iopub.execute_input":"2022-11-08T20:12:04.616786Z","iopub.status.idle":"2022-11-08T20:12:04.624996Z","shell.execute_reply.started":"2022-11-08T20:12:04.616701Z","shell.execute_reply":"2022-11-08T20:12:04.623794Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df.head(2)","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:12:04.626563Z","iopub.execute_input":"2022-11-08T20:12:04.627278Z","iopub.status.idle":"2022-11-08T20:12:04.645682Z","shell.execute_reply.started":"2022-11-08T20:12:04.627228Z","shell.execute_reply":"2022-11-08T20:12:04.644843Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"validate_df.head(2)","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:13:55.269633Z","iopub.execute_input":"2022-11-08T20:13:55.269984Z","iopub.status.idle":"2022-11-08T20:13:55.285880Z","shell.execute_reply.started":"2022-11-08T20:13:55.269929Z","shell.execute_reply":"2022-11-08T20:13:55.285006Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"image_path = f\"train/\"\n#test_path = f\"../input/input-{self.locker['comp_name']}/\" + \"test_images/\"\ntrain_datagen = ImageDataGenerator(rescale=1.0 / 255)\nvalid_datagen = ImageDataGenerator(rescale=1.0 / 255) \n \n# train_datagen1 = train_datagen.flow_from_dataframe(\n# dataframe=xtrain,\n# directory=image_path,\n# target_size=(128, 128), # images are resized to (28,28)\n# x_col='filename', #self.locker[\"id_name\"],\n# y_col='category', #self.locker[\"target_name\"],\n# batch_size=32,\n# seed=42,\n# shuffle=True, # Required for cutmix\n# class_mode=\"categorical\", # \"binary\"\n# )\ntrain_dataset = train_datagen.flow_from_dataframe(\n dataframe=train_df,\n directory=image_path,\n target_size=(128, 128), # images are resized to (28,28)\n x_col='filename', #self.locker[\"id_name\"],\n y_col='category', #self.locker[\"target_name\"],\n batch_size=32,\n seed=42,\n shuffle=True, # Required for cutmix\n class_mode=\"categorical\", # \"binary\"\n)\n# train_dataset = CutMixImageDataGenerator(\n# generator1= train_datagen1, #train_generator1,\n# generator2= train_datagen1, #train_generator2,\n# img_size=128,\n# batch_size=32,\n# )\n\nvalid_dataset = valid_datagen.flow_from_dataframe(\n dataframe=validate_df,\n directory=image_path,\n target_size=(128, 128), # images are resized to (28,28)\n x_col='filename', #self.locker[\"id_name\"],\n y_col='category', #self.locker[\"target_name\"],\n batch_size=32,\n seed=42,\n shuffle=True,\n class_mode=\"categorical\", # \"binary\"\n )","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:13:56.259326Z","iopub.execute_input":"2022-11-08T20:13:56.260093Z","iopub.status.idle":"2022-11-08T20:13:56.572028Z","shell.execute_reply.started":"2022-11-08T20:13:56.259631Z","shell.execute_reply":"2022-11-08T20:13:56.571229Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"epochs = 50\n \nhistory = model.fit_generator(\n train_dataset,\n steps_per_epoch= 8, #train_dataset.get_steps_per_epoch(),\n epochs=epochs,\n verbose=2,\n validation_data=valid_dataset,\n validation_steps=8,\n callbacks = callbacks\n)","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:13:57.418438Z","iopub.execute_input":"2022-11-08T20:13:57.418712Z","iopub.status.idle":"2022-11-08T20:14:34.016026Z","shell.execute_reply.started":"2022-11-08T20:13:57.418658Z","shell.execute_reply":"2022-11-08T20:14:34.015095Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model.evaluate_generator(\n generator=valid_dataset, steps=1\n) # 1 to make it perfectly divisible","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:14:34.017295Z","iopub.execute_input":"2022-11-08T20:14:34.017573Z","iopub.status.idle":"2022-11-08T20:14:34.210784Z","shell.execute_reply.started":"2022-11-08T20:14:34.017524Z","shell.execute_reply":"2022-11-08T20:14:34.210027Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"fig = plt.figure(figsize=(40, 40))\nfig.subplots_adjust(hspace=0.1, wspace=0.1)\nrow = 6\ncol = 6\na,b = next(train_dataset)\nfor i,(im1, l1) in enumerate(zip(a,b)):\n plot_num = i+1\n ax = fig.add_subplot(row, col, plot_num, xticks=[], yticks=[], label=f\"Level {l1}\")\n ax.set_title(f\"Level {l1}\")\n ax.imshow(im1)\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:14:34.214376Z","iopub.execute_input":"2022-11-08T20:14:34.214914Z","iopub.status.idle":"2022-11-08T20:14:38.242770Z","shell.execute_reply.started":"2022-11-08T20:14:34.214770Z","shell.execute_reply":"2022-11-08T20:14:38.242041Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Save Model","metadata":{"_uuid":"aa1fbc4081ae0de2993188b2bf658a0be5bc0687"}},{"cell_type":"code","source":"model.save_weights(\"model.h5\")","metadata":{"_uuid":"67575a4decdaf79a915d23151626b784ffa82642","execution":{"iopub.status.busy":"2022-11-08T20:14:38.243788Z","iopub.execute_input":"2022-11-08T20:14:38.244155Z","iopub.status.idle":"2022-11-08T20:14:38.466869Z","shell.execute_reply.started":"2022-11-08T20:14:38.244113Z","shell.execute_reply":"2022-11-08T20:14:38.466028Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Virtualize Training","metadata":{"_uuid":"1b76c0a9040bc0babf0a453e567e41e22f8a1e0e"}},{"cell_type":"code","source":"fig, (ax1, ax2) = plt.subplots(2, 1, figsize=(12, 12))\nax1.plot(history.history['loss'], color='b', label=\"Training loss\")\nax1.plot(history.history['val_loss'], color='r', label=\"validation loss\")\nax1.set_xticks(np.arange(1, epochs, 1))\nax1.set_yticks(np.arange(0, 1, 0.1))\n\nax2.plot(history.history['acc'], color='b', label=\"Training accuracy\")\nax2.plot(history.history['val_acc'], color='r',label=\"Validation accuracy\")\nax2.set_xticks(np.arange(1, epochs, 1))\n\nlegend = plt.legend(loc='best', shadow=True)\nplt.tight_layout()\nplt.show()","metadata":{"_uuid":"79055f2dc3e2abb47bea758e0464c86ca42ab431","execution":{"iopub.status.busy":"2022-11-08T20:14:38.468608Z","iopub.execute_input":"2022-11-08T20:14:38.469148Z","iopub.status.idle":"2022-11-08T20:14:39.305424Z","shell.execute_reply.started":"2022-11-08T20:14:38.469094Z","shell.execute_reply":"2022-11-08T20:14:39.304671Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Prepare Testing Data","metadata":{"_uuid":"764dc66e4b2bc558f3a0f90b80bb802f5b3d45a8"}},{"cell_type":"code","source":"test_filenames = os.listdir(\"test1\")\ntest_df = pd.DataFrame({\n 'filename': test_filenames\n})\nnb_samples = test_df.shape[0]","metadata":{"_uuid":"c35e70d3e1e4834dbbf840fa0ea08c049bfcd915","execution":{"iopub.status.busy":"2022-11-08T20:12:04.927179Z","iopub.status.idle":"2022-11-08T20:12:04.927924Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"nb_samples","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:12:04.928925Z","iopub.status.idle":"2022-11-08T20:12:04.929946Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Create Testing Generator","metadata":{"_uuid":"291bc3996dce8d05e174b27d64f03996d3e8038e"}},{"cell_type":"code","source":"image_path","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:12:04.931184Z","iopub.status.idle":"2022-11-08T20:12:04.932103Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_df.head(2)","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:12:04.933052Z","iopub.status.idle":"2022-11-08T20:12:04.933854Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_datagen = ImageDataGenerator(rescale=1./255)\nbatch_size = 32\n\ntest_dataset = test_datagen.flow_from_dataframe(\n dataframe=test_df,\n directory='test1/',\n target_size=(128, 128), # images are resized to (28,28)\n x_col='filename', #self.locker[\"id_name\"],\n y_col=None, #self.locker[\"target_name\"],\n batch_size=batch_size,\n seed=42,\n shuffle=False,\n class_mode=None, # \"binary\"\n )\n","metadata":{"_uuid":"52249ec1c35fb1be3adef386be245de3794e55aa","execution":{"iopub.status.busy":"2022-11-08T20:12:04.941059Z","iopub.status.idle":"2022-11-08T20:12:04.941704Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Predict","metadata":{"_uuid":"2fa580afca2931ec5ce374e732d8c1789d03f2ed"}},{"cell_type":"code","source":"predict = model.predict_generator(test_dataset, steps=np.ceil(nb_samples/batch_size))","metadata":{"_uuid":"4782eb23fa7d003f0e2415d995894017edb2d896","execution":{"iopub.status.busy":"2022-11-08T20:12:04.942776Z","iopub.status.idle":"2022-11-08T20:12:04.943721Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"For categoral classication the prediction will come with probability of each category. So we will pick the category that have the highest probability with numpy average max","metadata":{}},{"cell_type":"code","source":"predict.shape, test_df.shape","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:12:04.944745Z","iopub.status.idle":"2022-11-08T20:12:04.945571Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_df['category'] = np.argmax(predict, axis=-1)","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:12:04.946458Z","iopub.status.idle":"2022-11-08T20:12:04.947446Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"We will convert the predict category back into our generator classes by using `train_generator.class_indices`. It is the classes that image generator map while converting data into computer vision","metadata":{}},{"cell_type":"code","source":"label_map = dict((v,k) for k,v in train_dataset.class_indices.items())\ntest_df['category'] = test_df['category'].replace(label_map)","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:12:04.948249Z","iopub.status.idle":"2022-11-08T20:12:04.948994Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"From our prepare data part. We map data with `{1: 'dog', 0: 'cat'}`. Now we will map the result back to dog is 1 and cat is 0","metadata":{}},{"cell_type":"code","source":"test_df['category'] = test_df['category'].replace({ 'dog': 1, 'cat': 0 })","metadata":{"execution":{"iopub.status.busy":"2022-11-08T20:12:04.949893Z","iopub.status.idle":"2022-11-08T20:12:04.950629Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Virtaulize Result","metadata":{"_uuid":"b00add65fe765529e637c3a9904d710bb7eff1d8"}},{"cell_type":"code","source":"test_df['category'].value_counts().plot.bar()","metadata":{"_uuid":"d0bf6dd5ff344092fa0121f70bdd60fa5a40e29c","execution":{"iopub.status.busy":"2022-11-08T20:12:04.951599Z","iopub.status.idle":"2022-11-08T20:12:04.952543Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### See predicted result with images","metadata":{"_uuid":"ce72a83f80d6e012b12b82c8ee3365d671a3b307"}},{"cell_type":"code","source":"sample_test = test_df.head(18)\nsample_test.head()\nplt.figure(figsize=(12, 24))\nfor index, row in sample_test.iterrows():\n filename = row['filename']\n category = row['category']\n img = load_img(\"test1/\"+filename, target_size=IMAGE_SIZE)\n plt.subplot(6, 3, index+1)\n plt.imshow(img)\n plt.xlabel(filename + '(' + \"{}\".format(category) + ')' )\nplt.tight_layout()\nplt.show()","metadata":{"_uuid":"98b41dc83075e6297137fb45bf703c313dd4ae28","execution":{"iopub.status.busy":"2022-11-08T20:12:04.953463Z","iopub.status.idle":"2022-11-08T20:12:04.954409Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Submission","metadata":{"_uuid":"d1ca25943e73aa20a37f9fb8670ee430caeaaf1f"}},{"cell_type":"code","source":"submission_df = test_df.copy()\nsubmission_df['id'] = submission_df['filename'].str.split('.').str[0]\nsubmission_df['label'] = submission_df['category']\nsubmission_df.drop(['filename', 'category'], axis=1, inplace=True)\nsubmission_df.to_csv('submission_without_cutmix.csv', index=False)","metadata":{"_uuid":"cce9f3e2ffff0693d79d84590ed71fbbca7c3c7c","execution":{"iopub.status.busy":"2022-11-08T20:12:04.955326Z","iopub.status.idle":"2022-11-08T20:12:04.956156Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}