{"cells":[{"metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true},"cell_type":"code","source":"import random\nimport re\nimport numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nfrom sklearn.metrics import f1_score, precision_score, recall_score, confusion_matrix\nimport tensorflow as tf\n# As TPUs require acces to the GCS path\nfrom kaggle_datasets import KaggleDatasets\n!pip install typeguard\n!pip install -q --no-deps tensorflow-addons~=0.7\nimport tensorflow_addons as tfa\nprint('Tensorflow version ' + tf.__version__)\nfrom sklearn.model_selection import KFold\nimport gc","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Configurations"},{"metadata":{"_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","trusted":true},"cell_type":"code","source":"AUTO = tf.data.experimental.AUTOTUNE\n\n# Cluster Resolver for Google Cloud TPUs.\ntpu = tf.distribute.cluster_resolver.TPUClusterResolver()\n# Connects to the given cluster.\ntf.config.experimental_connect_to_cluster(tpu)\n# Initialize the TPU devices.\ntf.tpu.experimental.initialize_tpu_system(tpu)\n# TPU distribution strategy implementation.\nstrategy = tf.distribute.experimental.TPUStrategy(tpu)\n\n# Configuration\nimage_size=512#192\nIMAGE_SIZE = [image_size, image_size]#[512, 512]\nEPOCHS = 50#20#50\nBATCH_SIZE = 16 * strategy.num_replicas_in_sync","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Data Directories"},{"metadata":{"trusted":true},"cell_type":"code","source":"# IMAGE_SIZE = [512, 512]","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# Data access\nGCS_DS_PATH = KaggleDatasets().get_gcs_path()\n\nGCS_PATH_SELECT = { # available image sizes\n    192: GCS_DS_PATH + '/tfrecords-jpeg-192x192',\n    224: GCS_DS_PATH + '/tfrecords-jpeg-224x224',\n    331: GCS_DS_PATH + '/tfrecords-jpeg-331x331',\n    512: GCS_DS_PATH + '/tfrecords-jpeg-512x512'\n}\n\nGCS_PATH = GCS_PATH_SELECT[IMAGE_SIZE[0]]\n\nTRAINING_FILENAMES = tf.io.gfile.glob(GCS_PATH + '/train/*.tfrec') + tf.io.gfile.glob(GCS_PATH + '/val/*.tfrec')\nTEST_FILENAMES = tf.io.gfile.glob(GCS_PATH + '/test/*.tfrec') # predictions on this dataset should be submitted for the competition","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Classes"},{"metadata":{"trusted":true},"cell_type":"code","source":"CLASSES = ['pink primrose',    'hard-leaved pocket orchid', 'canterbury bells', 'sweet pea',     'wild geranium',     'tiger lily',           'moon orchid',              'bird of paradise', 'monkshood',        'globe thistle',         # 00 - 09\n           'snapdragon',       \"colt's foot\",               'king protea',      'spear thistle', 'yellow iris',       'globe-flower',         'purple coneflower',        'peruvian lily',    'balloon flower',   'giant white arum lily', # 10 - 19\n           'fire lily',        'pincushion flower',         'fritillary',       'red ginger',    'grape hyacinth',    'corn poppy',           'prince of wales feathers', 'stemless gentian', 'artichoke',        'sweet william',         # 20 - 29\n           'carnation',        'garden phlox',              'love in the mist', 'cosmos',        'alpine sea holly',  'ruby-lipped cattleya', 'cape flower',              'great masterwort', 'siam tulip',       'lenten rose',           # 30 - 39\n           'barberton daisy',  'daffodil',                  'sword lily',       'poinsettia',    'bolero deep blue',  'wallflower',           'marigold',                 'buttercup',        'daisy',            'common dandelion',      # 40 - 49\n           'petunia',          'wild pansy',                'primula',          'sunflower',     'lilac hibiscus',    'bishop of llandaff',   'gaura',                    'geranium',         'orange dahlia',    'pink-yellow dahlia',    # 50 - 59\n           'cautleya spicata', 'japanese anemone',          'black-eyed susan', 'silverbush',    'californian poppy', 'osteospermum',         'spring crocus',            'iris',             'windflower',       'tree poppy',            # 60 - 69\n           'gazania',          'azalea',                    'water lily',       'rose',          'thorn apple',       'morning glory',        'passion flower',           'lotus',            'toad lily',        'anthurium',             # 70 - 79\n           'frangipani',       'clematis',                  'hibiscus',         'columbine',     'desert-rose',       'tree mallow',          'magnolia',                 'cyclamen ',        'watercress',       'canna lily',            # 80 - 89\n           'hippeastrum ',     'bee balm',                  'pink quill',       'foxglove',      'bougainvillea',     'camellia',             'mallow',                   'mexican petunia',  'bromelia',         'blanket flower',        # 90 - 99\n           'trumpet creeper',  'blackberry lily',           'common tulip',     'wild rose']                                                                                                                                               # 100 - 102","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Custom LR scheduler"},{"metadata":{"trusted":true},"cell_type":"code","source":"# Learning rate schedule for TPU, GPU and CPU.\n# Using an LR ramp up because fine-tuning a pre-trained model.\n# Starting with a high LR would break the pre-trained weights.\n\nLR_START = 0.00001\nLR_MAX = 0.00005 * strategy.num_replicas_in_sync\nLR_MIN = 0.00001\nLR_RAMPUP_EPOCHS = 5\nLR_SUSTAIN_EPOCHS = 0\nLR_EXP_DECAY = .8\n\ndef lrfn(epoch):\n    if epoch < LR_RAMPUP_EPOCHS:\n        lr = (LR_MAX - LR_START) / LR_RAMPUP_EPOCHS * epoch + LR_START\n    elif epoch < LR_RAMPUP_EPOCHS + LR_SUSTAIN_EPOCHS:\n        lr = LR_MAX\n    else:\n        lr = (LR_MAX - LR_MIN) * LR_EXP_DECAY**(epoch - LR_RAMPUP_EPOCHS - LR_SUSTAIN_EPOCHS) + LR_MIN\n    return lr\n    \nlr_callback = tf.keras.callbacks.LearningRateScheduler(lrfn, verbose = True)\n\nrng = [i for i in range(EPOCHS)]\ny = [lrfn(x) for x in rng]\nplt.plot(rng, y)\nprint(\"Learning rate schedule: {:.3g} to {:.3g} to {:.3g}\".format(y[0], max(y), y[-1]))","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Dataset Functions"},{"metadata":{"trusted":true},"cell_type":"code","source":"def decode_image(image_data):\n    image = tf.image.decode_jpeg(image_data, channels=3)\n    image = tf.cast(image, tf.float32) / 255.0  # convert image to floats in [0, 1] range\n    image = tf.reshape(image, [*IMAGE_SIZE, 3]) # explicit size needed for TPU\n    return image\n\ndef read_labeled_tfrecord(example):\n    LABELED_TFREC_FORMAT = {\n        \"image\": tf.io.FixedLenFeature([], tf.string), # tf.string means bytestring\n        \"class\": tf.io.FixedLenFeature([], tf.int64),  # shape [] means single element\n    }\n    example = tf.io.parse_single_example(example, LABELED_TFREC_FORMAT)\n    image = decode_image(example['image'])\n    label = tf.cast(example['class'], tf.int32)\n    return image, label # returns a dataset of (image, label) pairs\n\ndef read_unlabeled_tfrecord(example):\n    UNLABELED_TFREC_FORMAT = {\n        \"image\": tf.io.FixedLenFeature([], tf.string), # tf.string means bytestring\n        \"id\": tf.io.FixedLenFeature([], tf.string),  # shape [] means single element\n        # class is missing, this competitions's challenge is to predict flower classes for the test dataset\n    }\n    example = tf.io.parse_single_example(example, UNLABELED_TFREC_FORMAT)\n    image = decode_image(example['image'])\n    idnum = example['id']\n    return image, idnum # returns a dataset of image(s)\n\ndef load_dataset(filenames, labeled = True, ordered = False):\n    # Read from TFRecords. For optimal performance, reading from multiple files at once and\n    # Diregarding data order. Order does not matter since we will be shuffling the data anyway\n    \n    ignore_order = tf.data.Options()\n    if not ordered:\n        ignore_order.experimental_deterministic = False # disable order, increase speed\n        \n    dataset = tf.data.TFRecordDataset(filenames, num_parallel_reads = AUTO) # automatically interleaves reads from multiple files\n    dataset = dataset.with_options(ignore_order) # use data as soon as it streams in, rather than in its original order\n    dataset = dataset.map(read_labeled_tfrecord if labeled else read_unlabeled_tfrecord, num_parallel_calls = AUTO) # returns a dataset of (image, label) pairs if labeled = True or (image, id) pair if labeld = False\n    return dataset\n\ndef data_augment(image, label):\n    # data augmentation. Thanks to the dataset.prefetch(AUTO) statement in the next function (below),\n    # this happens essentially for free on TPU. Data pipeline code is executed on the \"CPU\" part\n    # of the TPU while the TPU itself is computing gradients.\n    image = tf.image.random_flip_left_right(image)\n    return image, label   \n\ndef get_training_dataset(dataset):\n    dataset = dataset.map(data_augment, num_parallel_calls=AUTO)\n    dataset = dataset.repeat() # the training dataset must repeat for several epochs\n    dataset = dataset.shuffle(2048)\n    dataset = dataset.batch(BATCH_SIZE)\n    dataset = dataset.prefetch(AUTO) # prefetch next batch while training (autotune prefetch buffer size)\n    return dataset\n\ndef get_validation_dataset(dataset):\n    dataset = dataset.batch(BATCH_SIZE)\n    dataset = dataset.cache()\n    dataset = dataset.prefetch(AUTO) # prefetch next batch while training (autotune prefetch buffer size)\n    return dataset\n\ndef get_test_dataset(ordered=False):\n    dataset = load_dataset(TEST_FILENAMES, labeled=False, ordered=ordered)\n    dataset = dataset.batch(BATCH_SIZE)\n    dataset = dataset.prefetch(AUTO) # prefetch next batch while training (autotune prefetch buffer size)\n    return dataset\n\ndef count_data_items(filenames):\n    # the number of data items is written in the name of the .tfrec files, i.e. flowers00-230.tfrec = 230 data items\n    n = [int(re.compile(r\"-([0-9]*)\\.\").search(filename).group(1)) for filename in filenames]\n    return np.sum(n)\n\nNUM_TRAINING_IMAGES = count_data_items(TRAINING_FILENAMES) * 0.8\n# use validation data for training\nNUM_VALIDATION_IMAGES = count_data_items(TRAINING_FILENAMES) * 0.2\nNUM_TEST_IMAGES = count_data_items(TEST_FILENAMES)\nSTEPS_PER_EPOCH = NUM_TRAINING_IMAGES // BATCH_SIZE\n\nprint('Dataset: {} training images, {} validation images, {} unlabeled test images'.format(NUM_TRAINING_IMAGES, NUM_VALIDATION_IMAGES, NUM_TEST_IMAGES))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"\n\n!pip install -q efficientnet\nimport efficientnet.tfkeras as efn\n","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"from tensorflow.keras.applications import DenseNet201,DenseNet121\n\ndef get_model():\n    \n    with strategy.scope():\n#         rnet = DenseNet121( #DenseNet201\n#             input_shape=(IMAGE_SIZE[0], IMAGE_SIZE[1], 3),\n#             weights='imagenet',\n#             include_top=False\n#         )\n        enet = efn.EfficientNetB7( \n        input_shape=(image_size, image_size, 3),\n        weights='imagenet',\n        include_top=False\n    )\n\n        # trainable rnet\n        enet.trainable = True\n\n        model = tf.keras.Sequential([\n            enet,\n            tf.keras.layers.GlobalAveragePooling2D(),\n            tf.keras.layers.Dense(len(CLASSES), activation='softmax')\n        ])\n\n    model.compile(\n        optimizer='adam',\n        loss = 'sparse_categorical_crossentropy',\n        metrics=['sparse_categorical_accuracy']\n    )\n\n    return model\n\ndef train_cross_validate(folds = 5):\n    histories = []\n    models = []\n    early_stopping = tf.keras.callbacks.EarlyStopping(monitor = 'val_loss', patience = 4)\n    kfold = KFold(folds, shuffle = True, random_state = 42)\n    oof_labels = []\n    oof_predictions = []\n    test_ds = get_test_dataset(ordered=True) # since we are splitting the dataset and iterating separately on images and ids, order matters.\n    test_images_ds = test_ds.map(lambda image, idnum: image)\n    probabilities=0.\n    for fold, (trn_ind, val_ind) in enumerate(kfold.split(TRAINING_FILENAMES)):\n        if fold in [0,1,3,4]:continue\n        print('fold {} start....'.format(fold))\n        train_dataset = load_dataset(list(pd.DataFrame({'TRAINING_FILENAMES': TRAINING_FILENAMES}).loc[trn_ind]['TRAINING_FILENAMES']), labeled = True)\n        val_dataset = load_dataset(list(pd.DataFrame({'TRAINING_FILENAMES': TRAINING_FILENAMES}).loc[val_ind]['TRAINING_FILENAMES']), labeled = True, ordered = True)\n        checkpoint_name = f'model_fold_{fold + 1}' + '.h5'\n        model_checkpoint = tf.keras.callbacks.ModelCheckpoint(checkpoint_name, save_best_only = True, save_weights_only = True)\n        model = get_model()\n        history = model.fit(\n        get_training_dataset(train_dataset), \n        steps_per_epoch = STEPS_PER_EPOCH,\n        epochs = EPOCHS,\n        callbacks = [lr_callback, early_stopping, model_checkpoint],\n        validation_data = get_validation_dataset(val_dataset)\n        )\n        print('Load best weights for model prediction')\n        model.load_weights(checkpoint_name)\n#         models.append(model)\n        histories.append(history)\n        print('Get validation images for predicting ')\n#         validation_dataset = get_validation_dataset(val_dataset)\n#         images_ds = validation_dataset.map(lambda image, label: image)\n#         labels_ds = validation_dataset.map(lambda image, label: label).unbatch()\n#         NUM_FOLD_VALIDATION_IMAGES = count_data_items(list(pd.DataFrame({'TRAINING_FILENAMES': TRAINING_FILENAMES}).loc[val_ind]['TRAINING_FILENAMES']))\n#         print(f'We have { NUM_FOLD_VALIDATION_IMAGES} validation images for fold {fold + 1}')\n#         val_correct_labels = list(next(iter(labels_ds.batch(NUM_FOLD_VALIDATION_IMAGES))).numpy()) # get everything\n#         val_probabilities = model.predict(images_ds)\n#         val_predictions = list(np.argmax(val_probabilities, axis=-1))\n        print('Saving validation labels and prediction for oof calculation')\n        probabilities += model.predict(test_images_ds)\n    \n#         oof_labels.extend(val_correct_labels)\n#         oof_predictions.extend(val_predictions)\n        del model_checkpoint,model,train_dataset,val_dataset\n        gc.collect()\n        tf.keras.backend.clear_session()\n        \n#     score = f1_score(np.array(oof_labels), np.array(oof_predictions), labels = range(len(CLASSES)), average='macro')\n#     print(f'Our oof f1_score is {score}')\n    \n    return histories, probabilities\n\ndef train_and_predict(folds = 5):\n    test_ds = get_test_dataset(ordered=True) # since we are splitting the dataset and iterating separately on images and ids, order matters.\n    test_images_ds = test_ds.map(lambda image, idnum: image)\n    print('Start training 5 models')\n    histories, probabilities = train_cross_validate(folds = folds)\n    print('Computing predictions...')\n    # get the mean probability of the folds models\n#     probabilities = np.average([models[i].predict(test_images_ds) for i in range(folds)], axis = 0)\n    #for oom bug ,we fix folds=2\n    folds=1\n    probabilities=probabilities/folds\n    predictions = np.argmax(probabilities, axis=-1)\n    print('Generating submission.csv file...')\n    test_ids_ds = test_ds.map(lambda image, idnum: idnum).unbatch()\n    print('predictions shape',predictions.shape)\n    print('probabilities shape',probabilities.shape)\n    test_ids = next(iter(test_ids_ds.batch(NUM_TEST_IMAGES))).numpy().astype('U') # all in one batch\n    print('test_ids shape',test_ids.shape)\n    np.savetxt('submission.csv', np.rec.fromarrays([test_ids, predictions]), fmt=['%s', '%d'], delimiter=',', header='id,label', comments='')\n#     np.savetxt('submission_props.csv', np.rec.fromarrays([test_ids, probabilities]), fmt=['%s', '%d'], delimiter=',', header='id,label', comments='')\n#     with open('probs.npz',mode='wb') as fout:\n    np.savez_compressed('probs.npz', a=probabilities)\n    return histories\n    \n# run train and predict\nhistories = train_and_predict(folds = 5)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"import numpy as np\nhelp(np.savez_compressed)\n","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"","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":1}