{"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":"

About Competition

","metadata":{}},{"cell_type":"markdown","source":"

SIIM-FISABIO-RSNA COVID-19 Study Level Predictions

\n\n \n\n

About SIIM

\n\n

Society for Imaging Informatics in Medicine (SIIM) is the leading healthcare professional organization for those interested in the current and future use of informatics in medical imaging. The society's mission is to advance medical imaging informatics across the enterprise through education, research, and innovation in a multi-disciplinary community.

\n\n

This is my first Data Visualization Notebook. Kindly comment if there are any mistakes 🙂

\n\n

My Dataset

\n\n

My Data Preparation Notebook

 \n\n

YOLO v5 Image Level Training Notebook

 \n\n

My Training Notebook

 \nTo be updated","metadata":{}},{"cell_type":"markdown","source":"

Install Requirements

","metadata":{}},{"cell_type":"code","source":"!/opt/conda/bin/python3.7 -m pip install --upgrade pip\n! pip install -q efficientnet","metadata":{"execution":{"iopub.status.busy":"2021-05-28T10:27:57.986018Z","iopub.execute_input":"2021-05-28T10:27:57.986711Z","iopub.status.idle":"2021-05-28T10:28:28.563336Z","shell.execute_reply.started":"2021-05-28T10:27:57.9866Z","shell.execute_reply":"2021-05-28T10:28:28.562101Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

Importing Dependencies

","metadata":{}},{"cell_type":"code","source":"#-------------------\n# importing libraries\n#-------------------\nimport tensorflow as tf\nfrom tensorflow.keras import layers\nimport efficientnet.tfkeras as efn\nfrom sklearn.model_selection import KFold,StratifiedKFold\n\nfrom kaggle_datasets import KaggleDatasets\n\n\nimport pandas as pd\nimport numpy as np\n\nimport os\nimport shutil\nimport csv\n\nimport matplotlib.pyplot as plt\nimport PIL","metadata":{"execution":{"iopub.status.busy":"2021-05-28T10:28:28.565848Z","iopub.execute_input":"2021-05-28T10:28:28.566163Z","iopub.status.idle":"2021-05-28T10:28:34.354792Z","shell.execute_reply.started":"2021-05-28T10:28:28.566132Z","shell.execute_reply":"2021-05-28T10:28:34.353765Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

Checking TPU access

","metadata":{}},{"cell_type":"code","source":"AUTO = tf.data.experimental.AUTOTUNE\n\ntry:\n tpu = tf.distribute.cluster_resolver.TPUClusterResolver() \n print('Running on TPU ', tpu.master())\nexcept ValueError:\n tpu = None\n\nif tpu:\n tf.config.experimental_connect_to_cluster(tpu)\n tf.tpu.experimental.initialize_tpu_system(tpu)\n strategy = tf.distribute.experimental.TPUStrategy(tpu)\nelse:\n strategy = tf.distribute.get_strategy()\n\nprint(\"REPLICAS: \", strategy.num_replicas_in_sync)","metadata":{"execution":{"iopub.status.busy":"2021-05-28T10:28:34.356642Z","iopub.execute_input":"2021-05-28T10:28:34.357005Z","iopub.status.idle":"2021-05-28T10:28:40.729587Z","shell.execute_reply.started":"2021-05-28T10:28:34.35697Z","shell.execute_reply":"2021-05-28T10:28:40.728657Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"GCS_DS_PATH = KaggleDatasets().get_gcs_path(\"siim-covid19-dataset-256px-jpg\")\nprint(GCS_DS_PATH)\n\nTRAIN_PATH = GCS_DS_PATH + \"/256px/train/train/\"\nTEST_PATH = GCS_DS_PATH + \"/256px/test/test/\"\n\nNUM_CLASSES = 4\nHEIGHT,WIDTH = 256,256\nCHANNELS = 3\nBATCH_SIZE = 8 * strategy.num_replicas_in_sync\nSEED = 143\n\n\nclasses_dict = {\n \"Negative for Pneumonia\" : 0,\n \"Typical Appearance\" : 1,\n \"Indeterminate Appearance\" : 2,\n \"Atypical Appearance\" : 3\n}\n\nstudy_df = pd.read_csv(\"../input/siim-covid19-detection/train_study_level.csv\")\ntrain_df = pd.read_csv('../input/siim-covid19-dataset-256px-jpg/train.csv')\ntrain_df = train_df[[\"ImageInstanceUID\",\"StudyInstanceUID\",\"label_id\",\"study_label\"]]","metadata":{"execution":{"iopub.status.busy":"2021-05-28T10:28:40.731824Z","iopub.execute_input":"2021-05-28T10:28:40.732269Z","iopub.status.idle":"2021-05-28T10:28:41.126409Z","shell.execute_reply.started":"2021-05-28T10:28:40.732238Z","shell.execute_reply":"2021-05-28T10:28:41.125382Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df.head(3)","metadata":{"execution":{"iopub.status.busy":"2021-05-28T10:28:41.12756Z","iopub.execute_input":"2021-05-28T10:28:41.127833Z","iopub.status.idle":"2021-05-28T10:28:41.146818Z","shell.execute_reply.started":"2021-05-28T10:28:41.127807Z","shell.execute_reply":"2021-05-28T10:28:41.14568Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"study_df.head(3)","metadata":{"execution":{"iopub.status.busy":"2021-05-28T10:28:41.148227Z","iopub.execute_input":"2021-05-28T10:28:41.148661Z","iopub.status.idle":"2021-05-28T10:28:41.1602Z","shell.execute_reply.started":"2021-05-28T10:28:41.148588Z","shell.execute_reply":"2021-05-28T10:28:41.158997Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

Utility Functions for Data Preprocessing

","metadata":{}},{"cell_type":"code","source":"def process_img(filepath,label):\n image = tf.io.read_file(filepath)\n image = tf.image.decode_jpeg(image, channels=CHANNELS)\n image = tf.image.convert_image_dtype(image, tf.float32) \n #image = tf.image.resize(image, [HEIGHT,WIDTH])\n return image,label\n\n\ndef data_augment(image, label):\n p_spatial = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n p_rotate = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n p_pixel_1 = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n p_pixel_2 = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n p_pixel_3 = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n p_crop = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n \n # Flips\n image = tf.image.random_flip_left_right(image)\n image = tf.image.random_flip_up_down(image)\n if p_spatial > .75:\n image = tf.image.transpose(image)\n \n # Rotates\n if p_rotate > .75:\n image = tf.image.rot90(image, k=3) \n elif p_rotate > .5:\n image = tf.image.rot90(image, k=2) \n elif p_rotate > .25:\n image = tf.image.rot90(image, k=1) \n \n \n if p_pixel_1 >= .4:\n image = tf.image.random_saturation(image, lower=.7, upper=1.3)\n if p_pixel_2 >= .4:\n image = tf.image.random_contrast(image, lower=.8, upper=1.2)\n if p_pixel_3 >= .4:\n image = tf.image.random_brightness(image, max_delta=.1)\n \n \n if p_crop > .7:\n if p_crop > .9:\n image = tf.image.central_crop(image, central_fraction=.7)\n elif p_crop > .8:\n image = tf.image.central_crop(image, central_fraction=.8)\n else:\n image = tf.image.central_crop(image, central_fraction=.9)\n elif p_crop > .4:\n crop_size = tf.random.uniform([], int(HEIGHT*.8), HEIGHT, dtype=tf.int32)\n image = tf.image.random_crop(image, size=[crop_size, crop_size, CHANNELS])\n \n image = tf.image.resize(image, [HEIGHT,WIDTH])\n return image,label\n\ndef get_dataset(filenames,labels, training=True):\n dataset = tf.data.Dataset.from_tensor_slices((filenames,labels))\n dataset = dataset.map(process_img,num_parallel_calls=AUTO)\n dataset = dataset.map(data_augment,num_parallel_calls=AUTO)\n dataset = dataset.cache()\n dataset = dataset.repeat()\n if training:\n dataset = dataset.shuffle(2048)\n dataset = dataset.batch(BATCH_SIZE)\n dataset = dataset.prefetch(AUTO)\n return dataset","metadata":{"execution":{"iopub.status.busy":"2021-05-28T10:28:41.16198Z","iopub.execute_input":"2021-05-28T10:28:41.162404Z","iopub.status.idle":"2021-05-28T10:28:41.182195Z","shell.execute_reply.started":"2021-05-28T10:28:41.162359Z","shell.execute_reply":"2021-05-28T10:28:41.181165Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

Model Function

","metadata":{}},{"cell_type":"code","source":"def create_model():\n \n pretrained = efn.EfficientNetB4(include_top=False, weights='noisy-student',input_shape=[HEIGHT,WIDTH, 3])\n \n x = pretrained.output\n x = tf.keras.layers.GlobalAveragePooling2D() (x)\n outputs = tf.keras.layers.Dense(NUM_CLASSES,activation=\"softmax\", dtype='float32')(x)\n \n model = tf.keras.Model(pretrained.input, outputs)\n return model\n\nmodel = create_model()\n#model.summary()","metadata":{"execution":{"iopub.status.busy":"2021-05-28T10:28:41.184475Z","iopub.execute_input":"2021-05-28T10:28:41.184782Z","iopub.status.idle":"2021-05-28T10:28:52.820421Z","shell.execute_reply.started":"2021-05-28T10:28:41.184753Z","shell.execute_reply":"2021-05-28T10:28:52.819385Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#chxnet model\n\nwith strategy.scope():\n model = tf.keras.applications.DenseNet121(weights= \"imagenet\",\n include_top=False,\n input_shape=(HEIGHT,WIDTH,CHANNELS), pooling=\"avg\")\n predictions = tf.keras.layers.Dense(14, activation='sigmoid', name='predictions')(model.output)\n\n model = tf.keras.Model(inputs=model.input, outputs=predictions) \n model.load_weights(\"../input/pneumonia-classification-challenge/pretrained.h5\")\n model = tf.keras.Model(model.input, model.layers[-2].output) \n x = tf.keras.layers.Dense(512, activation = \"relu\")(model.output)\n x = tf.keras.layers.Dropout(0.3)(x)\n x = tf.keras.layers.Dense(128, activation = \"relu\")(x)\n x = tf.keras.layers.Dense(64, activation = \"relu\")(x)\n outputs = tf.keras.layers.Dense(NUM_CLASSES, activation = \"softmax\", dtype = tf.float32)(x)\n\n chxnet = tf.keras.Model(model.input,outputs)\n\n for layer in chxnet.layers[:-15]:\n layer.trainble = False","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

Compiling the Model

","metadata":{}},{"cell_type":"code","source":"import tensorflow_addons as tfa\n\ndef compile_model(model, lr=0.001):\n \n optimizer = tf.keras.optimizers.Adam(lr=lr)\n \n loss = tf.keras.losses.CategoricalCrossentropy()\n \n metrics = [\n tfa.metrics.F1Score(num_classes = NUM_CLASSES,average = \"macro\", name = \"f1_score\"),\n tf.keras.metrics.CategoricalAccuracy(name='acc')\n ]\n\n model.compile(optimizer=optimizer, loss=loss, metrics=metrics)\n\n return model","metadata":{"execution":{"iopub.status.busy":"2021-05-28T10:28:52.821914Z","iopub.execute_input":"2021-05-28T10:28:52.822209Z","iopub.status.idle":"2021-05-28T10:28:52.955686Z","shell.execute_reply.started":"2021-05-28T10:28:52.822181Z","shell.execute_reply":"2021-05-28T10:28:52.954595Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

Model Callbacks

","metadata":{}},{"cell_type":"code","source":"METRIC = \"val_acc\"\n\ndef create_callbacks(kfold,metric = METRIC):\n \n cpk_path = f'./best_model_{kfold}.h5'\n \n checkpoint = tf.keras.callbacks.ModelCheckpoint(\n filepath=cpk_path,\n monitor= metric,\n mode='max',\n save_best_only=True,\n verbose=1,\n )\n\n reducelr = tf.keras.callbacks.ReduceLROnPlateau(\n monitor= metric,\n mode='max',\n factor=0.1,\n patience=3,\n verbose=0\n )\n\n earlystop = tf.keras.callbacks.EarlyStopping(\n monitor= metric,\n mode='max',\n patience=10, \n verbose=1\n )\n \n callbacks = [checkpoint, reducelr, earlystop] \n \n return callbacks","metadata":{"execution":{"iopub.status.busy":"2021-05-28T10:28:52.956863Z","iopub.execute_input":"2021-05-28T10:28:52.957131Z","iopub.status.idle":"2021-05-28T10:28:52.964214Z","shell.execute_reply.started":"2021-05-28T10:28:52.957106Z","shell.execute_reply":"2021-05-28T10:28:52.96302Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

Training

","metadata":{}},{"cell_type":"code","source":"files_ls = tf.io.gfile.glob(TRAIN_PATH + \"*.jpg\" )\nfiles_df = pd.DataFrame(files_ls, columns = [\"filepath\"])\n\nlabels = np.zeros((len(files_ls),NUM_CLASSES))\ntmp_labels = np.zeros((len(files_ls)))\n\ndef get_id(filepath):\n tmp = filepath.split(\"/\")[-1]\n tmp = tmp.split(\".\")[0]\n tmp = tmp.split(\"_\")[-1]\n return tmp\n\nfor i in range(len(files_ls)):\n image_id = get_id(files_ls[i])\n label_id = train_df[train_df[\"ImageInstanceUID\"] == image_id][\"study_label\"]\n labels[i][label_id] = 1\n tmp_labels[i] = label_id\n \nprint(\"Labels shape: \",labels.shape)","metadata":{"execution":{"iopub.status.busy":"2021-05-28T10:32:15.024592Z","iopub.execute_input":"2021-05-28T10:32:15.024988Z","iopub.status.idle":"2021-05-28T10:32:16.191009Z","shell.execute_reply.started":"2021-05-28T10:32:15.024956Z","shell.execute_reply":"2021-05-28T10:32:16.19021Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"EPOCHS = 30\nVERBOSE = 1\nN_SPLITS = 5\n\nkfold = StratifiedKFold(n_splits=N_SPLITS, shuffle=True, random_state=SEED)\nhistory = {}\n\n\nfor fold,(tID,vID) in enumerate(kfold.split(files_ls,tmp_labels)):\n tFiles, tLabels = list(files_df.iloc[tID][\"filepath\"]) , labels[tID]\n vFiles, vLabels = list(files_df.iloc[vID][\"filepath\"]) , labels[vID]\n print(\"Number of Training Images: \",len(tID))\n print(\"Number of Validation Images: \",len(vID))\n \n STEPS_PER_EPOCH = len(tID)//BATCH_SIZE\n VALID_STEPS = len(vID)//BATCH_SIZE\n \n tf.keras.backend.clear_session()\n \n train_ds = get_dataset(tFiles,tLabels, training = True)\n val_ds = get_dataset(vFiles, vLabels, training = False)\n \n with strategy.scope():\n #model = create_model()\n model = chxnet\n model = compile_model(model, lr=0.0001)\n callbacks = create_callbacks(kfold = fold)\n \n print(\"------------------Fold - \",fold+1,\" --------------------------\")\n history[fold] = model.fit(\n train_ds,\n epochs=EPOCHS,\n callbacks=callbacks,\n validation_data = val_ds,\n verbose=VERBOSE,\n steps_per_epoch = STEPS_PER_EPOCH,\n validation_steps=VALID_STEPS\n )","metadata":{"execution":{"iopub.status.busy":"2021-05-28T10:32:23.889864Z","iopub.execute_input":"2021-05-28T10:32:23.890429Z","iopub.status.idle":"2021-05-28T10:53:38.886033Z","shell.execute_reply.started":"2021-05-28T10:32:23.890396Z","shell.execute_reply":"2021-05-28T10:53:38.884997Z"},"_kg_hide-output":true,"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

History Plotting

","metadata":{}},{"cell_type":"code","source":"plt.figure(figsize=(8*N_SPLITS,24))\n\nfor i in range(N_SPLITS):\n acc = history[i].history['acc']\n val_acc = history[i].history['val_acc']\n f1 = history[i].history['f1_score']\n val_f1 = history[i].history['val_f1_score']\n loss = history[i].history['loss']\n val_loss = history[i].history['val_loss']\n epochs_range = range(len(history[i].history['val_loss'])) \n \n plt.subplot(N_SPLITS, 3,i*3+1)\n plt.plot(epochs_range, acc, label='Training Accuracy')\n plt.plot(epochs_range, val_acc, label='Validation Accuracy')\n plt.legend(loc='lower right')\n plt.title(f'FOLD:{str(i)} Training and Validation Accuracy')\n \n plt.subplot(N_SPLITS, 3,i*3+2)\n plt.plot(epochs_range, f1, label='Training F1 score')\n plt.plot(epochs_range, val_f1, label='Validation F1 score')\n plt.legend(loc='lower right')\n plt.title(f'FOLD:{str(i)} Training and Validation F1 score')\n \n plt.subplot(N_SPLITS, 3, i*3+3)\n plt.plot(epochs_range, loss, label='Training Loss')\n plt.plot(epochs_range, val_loss, label='Validation Loss')\n plt.legend(loc='upper right')\n plt.title(f'FOLD:{str(i)} Training and Validation Loss')\n\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2021-05-28T10:53:38.890379Z","iopub.execute_input":"2021-05-28T10:53:38.89078Z","iopub.status.idle":"2021-05-28T10:53:41.459106Z","shell.execute_reply.started":"2021-05-28T10:53:38.890741Z","shell.execute_reply":"2021-05-28T10:53:41.458051Z"},"trusted":true},"execution_count":null,"outputs":[]}]}