{"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":"#### Thanks to https://www.kaggle.com/xhlulu/ranzcr-efficientnet-tpu-training","metadata":{}},{"cell_type":"code","source":"!pip install efficientnet -q","metadata":{"execution":{"iopub.status.busy":"2021-07-17T13:19:37.513338Z","iopub.execute_input":"2021-07-17T13:19:37.514005Z","iopub.status.idle":"2021-07-17T13:19:47.496494Z","shell.execute_reply.started":"2021-07-17T13:19:37.513897Z","shell.execute_reply":"2021-07-17T13:19:47.494888Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import os\n\nimport efficientnet.tfkeras as efn\nimport numpy as np\nimport pandas as pd\nfrom kaggle_datasets import KaggleDatasets\nfrom sklearn.model_selection import train_test_split\nimport tensorflow as tf\nfrom sklearn.model_selection import GroupKFold\nimport shutil\nimport distutils\nfrom distutils import dir_util","metadata":{"execution":{"iopub.status.busy":"2021-07-17T13:19:47.498408Z","iopub.execute_input":"2021-07-17T13:19:47.498755Z","iopub.status.idle":"2021-07-17T13:19:55.853802Z","shell.execute_reply.started":"2021-07-17T13:19:47.498723Z","shell.execute_reply":"2021-07-17T13:19:55.852359Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"d1 = pd.read_csv(\"../input/label-180/psudolabel (2).csv\")\nd2 = pd.read_csv(\"../input/siim-covid19-detection/train_study_level.csv\")","metadata":{"execution":{"iopub.status.busy":"2021-07-17T13:19:55.856461Z","iopub.execute_input":"2021-07-17T13:19:55.856811Z","iopub.status.idle":"2021-07-17T13:19:55.899601Z","shell.execute_reply.started":"2021-07-17T13:19:55.856775Z","shell.execute_reply":"2021-07-17T13:19:55.897937Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"d3 = pd.DataFrame([], columns=['id', 'Negative for Pneumonia' ,'Typical Appearance' ,'Indeterminate Appearance','Atypical Appearance'])\nd1 = d1.rename(columns={'id':'id', 'negative':'Negative for Pneumonia' ,'typical':'Typical Appearance' ,'indeterminate':'Indeterminate Appearance','atypical':'Atypical Appearance'})","metadata":{"execution":{"iopub.status.busy":"2021-07-17T13:19:55.901320Z","iopub.execute_input":"2021-07-17T13:19:55.901656Z","iopub.status.idle":"2021-07-17T13:19:55.919489Z","shell.execute_reply.started":"2021-07-17T13:19:55.901625Z","shell.execute_reply":"2021-07-17T13:19:55.918017Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"d1['Negative for Pneumonia'] = np.where(d1['Negative for Pneumonia'] > .95, 1, 0)\nd1['Typical Appearance'] = np.where(d1['Typical Appearance'] > 0.95, 1, 0)\nd1['Indeterminate Appearance'] = np.where(d1['Indeterminate Appearance'] > 0.95, 1, 0)\nd1['Atypical Appearance'] = np.where(d1['Atypical Appearance'] > 0.95, 1, 0)","metadata":{"execution":{"iopub.status.busy":"2021-07-17T13:19:55.921012Z","iopub.execute_input":"2021-07-17T13:19:55.921412Z","iopub.status.idle":"2021-07-17T13:19:55.949824Z","shell.execute_reply.started":"2021-07-17T13:19:55.921378Z","shell.execute_reply":"2021-07-17T13:19:55.948851Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"d3 =pd.concat(\n    [d1,d2],\n    axis=0,\n    join=\"outer\",\n    ignore_index=True,\n    keys=None,\n    levels=None,\n    names=None,\n    verify_integrity=False,\n    copy=True,\n)","metadata":{"execution":{"iopub.status.busy":"2021-07-17T13:19:55.951289Z","iopub.execute_input":"2021-07-17T13:19:55.951619Z","iopub.status.idle":"2021-07-17T13:19:55.960579Z","shell.execute_reply.started":"2021-07-17T13:19:55.951587Z","shell.execute_reply":"2021-07-17T13:19:55.959436Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def auto_select_accelerator():\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\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        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=128, 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","metadata":{"execution":{"iopub.status.busy":"2021-07-17T13:19:55.962137Z","iopub.execute_input":"2021-07-17T13:19:55.962564Z","iopub.status.idle":"2021-07-17T13:19:55.983280Z","shell.execute_reply.started":"2021-07-17T13:19:55.962532Z","shell.execute_reply":"2021-07-17T13:19:55.982182Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"COMPETITION_NAME = \"pseudo-images\"\nstrategy = auto_select_accelerator()\nBATCH_SIZE = strategy.num_replicas_in_sync * 16\nGCS_DS_PATH = KaggleDatasets().get_gcs_path(COMPETITION_NAME)","metadata":{"execution":{"iopub.status.busy":"2021-07-17T13:19:55.986530Z","iopub.execute_input":"2021-07-17T13:19:55.987103Z","iopub.status.idle":"2021-07-17T13:20:02.011753Z","shell.execute_reply.started":"2021-07-17T13:19:55.987057Z","shell.execute_reply":"2021-07-17T13:20:02.010313Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#load_dir = f\"/kaggle/input/{COMPETITION_NAME}/\"\ndf = d3\nlabel_cols = df.columns[1:5]\n","metadata":{"execution":{"iopub.status.busy":"2021-07-17T13:20:02.014075Z","iopub.execute_input":"2021-07-17T13:20:02.014600Z","iopub.status.idle":"2021-07-17T13:20:02.020664Z","shell.execute_reply.started":"2021-07-17T13:20:02.014556Z","shell.execute_reply":"2021-07-17T13:20:02.019028Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"gkf  = GroupKFold(n_splits = 5)\ndf['fold'] = -1\nfor fold, (train_idx, val_idx) in enumerate(gkf.split(df, groups = df.id.tolist())):\n    df.loc[val_idx, 'fold'] = fold","metadata":{"execution":{"iopub.status.busy":"2021-07-17T13:20:02.022796Z","iopub.execute_input":"2021-07-17T13:20:02.023169Z","iopub.status.idle":"2021-07-17T13:20:02.086088Z","shell.execute_reply.started":"2021-07-17T13:20:02.023135Z","shell.execute_reply":"2021-07-17T13:20:02.084977Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"for i in range(5):\n    \n    valid_paths = GCS_DS_PATH + '/' + df[df['fold'] == i]['id'] + '.png' #\"/train/\"\n    train_paths = GCS_DS_PATH + '/' + df[df['fold'] != i]['id'] + '.png' #\"/train/\" \n    valid_labels = df[df['fold'] == i][label_cols].values\n    train_labels = df[df['fold'] != i][label_cols].values\n\n    IMSIZE = (224, 240, 260, 300, 380, 456, 528, 600)\n    IMS = 7\n\n    decoder = build_decoder(with_labels=True, target_size=(IMSIZE[IMS], IMSIZE[IMS]), ext='png')\n    test_decoder = build_decoder(with_labels=False, target_size=(IMSIZE[IMS], IMSIZE[IMS]),ext='png')\n\n    train_dataset = build_dataset(\n        train_paths, train_labels, bsize=BATCH_SIZE, decode_fn=decoder\n    )\n\n    valid_dataset = build_dataset(\n        valid_paths, valid_labels, bsize=BATCH_SIZE, decode_fn=decoder,\n        repeat=False, shuffle=False, augment=False\n    )\n\n    try:\n        n_labels = train_labels.shape[1]\n    except:\n        n_labels = 1\n\n    with strategy.scope():\n        model = tf.keras.Sequential([\n            efn.EfficientNetB7(\n                input_shape=(IMSIZE[IMS], IMSIZE[IMS], 3),\n                weights='imagenet',\n                include_top=False),\n            tf.keras.layers.GlobalAveragePooling2D(),\n            tf.keras.layers.Dense(n_labels, activation='softmax')\n        ])\n        model.compile(\n            optimizer=tf.keras.optimizers.Adam(),\n            loss='categorical_crossentropy',\n            metrics=[tf.keras.metrics.AUC(multi_label=True)])\n\n        model.summary()\n\n\n    steps_per_epoch = train_paths.shape[0] // BATCH_SIZE\n    checkpoint = tf.keras.callbacks.ModelCheckpoint(\n        f'model{i}.h5', save_best_only=True, monitor='val_loss', mode='min')\n    lr_reducer = tf.keras.callbacks.ReduceLROnPlateau(\n        monitor=\"val_loss\", patience=3, min_lr=1e-6, mode='min')\n\n    history = model.fit(\n        train_dataset, \n        epochs=40,\n        verbose=1,\n        callbacks=[checkpoint, lr_reducer],\n        steps_per_epoch=steps_per_epoch,\n        validation_data=valid_dataset)\n\n    hist_df = pd.DataFrame(history.history)\n    hist_df.to_csv(f'history{i}.csv')","metadata":{"execution":{"iopub.status.busy":"2021-07-17T13:20:02.087747Z","iopub.execute_input":"2021-07-17T13:20:02.088109Z","iopub.status.idle":"2021-07-17T16:30:07.776601Z","shell.execute_reply.started":"2021-07-17T13:20:02.088076Z","shell.execute_reply":"2021-07-17T16:30:07.771097Z"},"trusted":true},"execution_count":null,"outputs":[]}]}