{"cells":[{"metadata":{},"cell_type":"markdown","source":"**[Click here for the submission notebook](https://www.kaggle.com/xhlulu/ranzcr-efficientnet-submission)**\n\n**[Click here for updated discussions and results table](https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/204950)**\n\n**[Click here for GPU notebook instead](https://www.kaggle.com/xhlulu/ranzcr-efficientnet-gpu-starter-train-submit)**\n\nThis is the training notebook for the EfficientNet model trained on TPU using Keras. The best version achieves **LB 0.957** using a single EfficientNet B7."},{"metadata":{"trusted":true},"cell_type":"code","source":"!pip install git+https://github.com/keras-team/keras-applications.git -q","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true},"cell_type":"code","source":"import os\n\nimport numpy as np\nimport pandas as pd\nfrom kaggle_datasets import KaggleDatasets\nimport keras_applications as ka\nfrom sklearn.model_selection import train_test_split\nimport tensorflow as tf","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"## Set keras applications to tensorflow\n\nThis is needed since tensorflow 2.2.0 doesn't have the efficientnet, so we need to use `keras-applications` and inject tensorflow."},{"metadata":{"trusted":true},"cell_type":"code","source":"def set_to_tf(ka):\n    from tensorflow.keras import backend, layers, models, utils\n    ka._KERAS_BACKEND = backend\n    ka._KERAS_LAYERS = layers\n    ka._KERAS_MODELS = models\n    ka._KERAS_UTILS = utils","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"set_to_tf(ka)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"## Helper functions"},{"metadata":{},"cell_type":"markdown","source":"The following functions are defined below (unhide to see):\n```python\nauto_select_accelerator()\n\nbuild_decoder(with_labels=True, target_size=(256, 256), ext='jpg')\n\nbuild_augmenter(with_labels=True)\n\nbuild_dataset(paths, labels=None, bsize=32, cache=True,\n              decode_fn=None, augment_fn=None,\n              augment=True, repeat=True, shuffle=1024, \n              cache_dir=\"\")\n```"},{"metadata":{"_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","trusted":true,"_kg_hide-input":true},"cell_type":"code","source":"def auto_select_accelerator():\n    \"\"\"\n    Reference: \n        * https://www.kaggle.com/mgornergoogle/getting-started-with-100-flowers-on-tpu\n        * https://www.kaggle.com/xhlulu/ranzcr-efficientnet-tpu-training\n    \"\"\"\n    try:\n        tpu = tf.distribute.cluster_resolver.TPUClusterResolver()\n        tf.config.experimental_connect_to_cluster(tpu)\n        tf.tpu.experimental.initialize_tpu_system(tpu)\n        strategy = tf.distribute.experimental.TPUStrategy(tpu)\n        print(\"Running on TPU:\", tpu.master())\n    except ValueError:\n        strategy = tf.distribute.get_strategy()\n    print(f\"Running on {strategy.num_replicas_in_sync} replicas\")\n    \n    return strategy\n\n\ndef build_decoder(with_labels=True, target_size=(256, 256), ext='jpg'):\n    def decode(path):\n        file_bytes = tf.io.read_file(path)\n        if ext == 'png':\n            img = tf.image.decode_png(file_bytes, channels=3)\n        elif ext in ['jpg', 'jpeg']:\n            img = tf.image.decode_jpeg(file_bytes, channels=3)\n        else:\n            raise ValueError(\"Image extension not supported\")\n\n        img = tf.cast(img, tf.float32) / 255.0\n        img = tf.image.resize(img, target_size)\n\n        return img\n    \n    def decode_with_labels(path, label):\n        return decode(path), label\n    \n    return decode_with_labels if with_labels else decode\n\n\ndef build_augmenter(with_labels=True):\n    def augment(img):\n        img = tf.image.random_flip_left_right(img)\n        img = tf.image.random_flip_up_down(img)\n        return img\n    \n    def augment_with_labels(img, label):\n        return augment(img), label\n    \n    return augment_with_labels if with_labels else augment\n\n\ndef build_dataset(paths, labels=None, bsize=32, cache=True,\n                  decode_fn=None, augment_fn=None,\n                  augment=True, repeat=True, shuffle=1024, \n                  cache_dir=\"\"):\n    if cache_dir != \"\" and cache is True:\n        os.makedirs(cache_dir, exist_ok=True)\n    \n    if decode_fn is None:\n        decode_fn = build_decoder(labels is not None)\n    \n    if augment_fn is None:\n        augment_fn = build_augmenter(labels is not None)\n    \n    AUTO = tf.data.experimental.AUTOTUNE\n    slices = paths if labels is None else (paths, labels)\n    \n    dset = tf.data.Dataset.from_tensor_slices(slices)\n    dset = dset.map(decode_fn, num_parallel_calls=AUTO)\n    dset = dset.cache(cache_dir) if cache else dset\n    dset = dset.map(augment_fn, num_parallel_calls=AUTO) if augment else dset\n    dset = dset.repeat() if repeat else dset\n    dset = dset.shuffle(shuffle) if shuffle else dset\n    dset = dset.batch(bsize).prefetch(AUTO)\n    \n    return dset","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"## Variables and configurations"},{"metadata":{"trusted":true},"cell_type":"code","source":"COMPETITION_NAME = \"ranzcr-clip-catheter-line-classification\"\nstrategy = auto_select_accelerator()\nBATCH_SIZE = strategy.num_replicas_in_sync * 16\nGCS_DS_PATH = KaggleDatasets().get_gcs_path(COMPETITION_NAME)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"## Preparing dataset"},{"metadata":{},"cell_type":"markdown","source":"### Loading and preprocess CSVs"},{"metadata":{"trusted":true},"cell_type":"code","source":"load_dir = f\"/kaggle/input/{COMPETITION_NAME}/\"\ndf = pd.read_csv(load_dir + 'train.csv')\n\n# paths = load_dir + \"train/\" + df['StudyInstanceUID'] + '.jpg'\npaths = GCS_DS_PATH + \"/train/\" + df['StudyInstanceUID'] + '.jpg'\n\nsub_df = pd.read_csv(load_dir + 'sample_submission.csv')\n\n# test_paths = load_dir + \"test/\" + sub_df['StudyInstanceUID'] + '.jpg'\ntest_paths = GCS_DS_PATH + \"/test/\" + sub_df['StudyInstanceUID'] + '.jpg'\n\n# Get the multi-labels\nlabel_cols = sub_df.columns[1:]\nlabels = df[label_cols].values","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# Train test split\n(\n    train_paths, valid_paths, \n    train_labels, valid_labels\n) = train_test_split(paths, labels, test_size=0.2, random_state=42)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# Build the tensorflow datasets\nIMSIZE = (224, 240, 260, 300, 380, 456, 528, 600)\n\ndecoder = build_decoder(with_labels=True, target_size=(IMSIZE[6], IMSIZE[6]))\ntest_decoder = build_decoder(with_labels=False, target_size=(IMSIZE[6], IMSIZE[6]))\n\ntrain_dataset = build_dataset(\n    train_paths, train_labels, bsize=BATCH_SIZE, decode_fn=decoder\n)\n\nvalid_dataset = build_dataset(\n    valid_paths, valid_labels, bsize=BATCH_SIZE, decode_fn=decoder,\n    repeat=False, shuffle=False, augment=False\n)\n\ntest_dataset = build_dataset(\n    test_paths, cache=False, bsize=BATCH_SIZE, decode_fn=test_decoder,\n    repeat=False, shuffle=False, augment=False\n)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"## Modeling"},{"metadata":{"trusted":true},"cell_type":"code","source":"n_labels = labels.shape[1]\n\nwith strategy.scope():\n    model = tf.keras.Sequential([\n        ka.efficientnet.EfficientNetB7(\n            input_shape=(IMSIZE[7], IMSIZE[7], 3),\n            weights='imagenet',\n            include_top=False),\n        tf.keras.layers.GlobalAveragePooling2D(),\n        tf.keras.layers.Dense(n_labels, activation='sigmoid')\n    ])\n    model.compile(\n        optimizer='adam',\n        loss='binary_crossentropy',\n        metrics=[tf.keras.metrics.AUC(multi_label=True)])\n    model.summary()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"steps_per_epoch = train_paths.shape[0] // BATCH_SIZE\ncheckpoint = tf.keras.callbacks.ModelCheckpoint(\n    'model.h5', save_best_only=True, monitor='val_auc', mode='max')\nlr_reducer = tf.keras.callbacks.ReduceLROnPlateau(\n    monitor=\"val_auc\", patience=3, min_lr=1e-6, mode='max')","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"history = model.fit(\n    train_dataset, \n    epochs=30,\n    verbose=2,\n    callbacks=[checkpoint, lr_reducer],\n    steps_per_epoch=steps_per_epoch,\n    validation_data=valid_dataset)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"## Save history"},{"metadata":{"trusted":true},"cell_type":"code","source":"hist_df = pd.DataFrame(history.history)\nhist_df.to_csv('history.csv')","execution_count":null,"outputs":[]}],"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":4,"nbformat_minor":4}