{"cells":[{"metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true},"cell_type":"code","source":"!pip install -q tensorflow_addons\n!pip install -q efficientnet","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","trusted":true},"cell_type":"code","source":"import math, sys, re, os, gc\nimport tensorflow as tf\nimport tensorflow_addons as tfa\nimport numpy as np\n\nimport efficientnet.tfkeras as efn\n\nfrom tensorflow import keras\n\nfrom matplotlib import pyplot as plt\nprint(\"Tensorflow version \" + tf.__version__)\nAUTO = tf.data.experimental.AUTOTUNE\nfrom kaggle_datasets import KaggleDatasets\nfrom sklearn.metrics import f1_score, precision_score, recall_score, confusion_matrix","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"COMP_ENV_UNKNOWN = 0\nCOMP_ENV_CPU = 1\nCOMP_ENV_GPU = 2\nCOMP_ENV_TPU = 3\n\ndef get_computation_environment(verbose=1):\n    \"\"\" Detect computational hardware.\n        To use the selected distribution strategy:\n        with strategy.scope:\n            --- define your (Keras) model here ---\n        \n        For distributed computing, the batch size and learning rate need to be adjusted:\n        global_batch_size = BATCH_SIZE * strategy.num_replicas_in_sync # num replcas is 8 on a single TPU or N when runing on N GPUs.\n        learning_rate = LEARNING_RATE * strategy.num_replicas_in_sync\n    \"\"\"\n    calculator = COMP_ENV_UNKNOWN\n    try:\n        tpu = tf.distribute.cluster_resolver.TPUClusterResolver() # TPU detection\n    except ValueError:\n        tpu = None\n        gpus = tf.config.experimental.list_logical_devices(\"GPU\")\n    \n    # Select appropriate distribution strategy for hardware\n    if tpu:\n        calculator = COMP_ENV_TPU\n        tf.config.experimental_connect_to_cluster(tpu)\n        tf.tpu.experimental.initialize_tpu_system(tpu)\n        strategy = tf.distribute.experimental.TPUStrategy(tpu)\n    elif len(gpus) > 0:\n        calculator = COMP_ENV_GPU\n        strategy = tf.distribute.MirroredStrategy(gpus) # this works for 1 to multiple GPUs\n    else:\n        calculator = COMP_ENV_CPU\n        strategy = tf.distribute.get_strategy() # default strategy that works on CPU and single GPU\n        \n    \n    if verbose == 1 or verbose == 2:\n        if calculator == COMP_ENV_TPU:\n            print('Running on TPU ', tpu.master())\n        elif calculator == COMP_ENV_GPU:\n            print('Running on ', len(gpus), ' GPU(s) ')\n        elif calculator == COMP_ENV_CPU:\n            print('Running on CPU')\n        else:\n            print(\"Running on unknown environment\")\n        print(\"Number of accelerators: \", strategy.num_replicas_in_sync)\n\n    return calculator, strategy","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"calculator, strategy = get_computation_environment(verbose=2)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"GCS_DS_PATH = KaggleDatasets().get_gcs_path() # you can list the bucket with \"!gsutil ls $GCS_DS_PATH\"","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"SKIP_VALIDATION = False\nuse_half_splitted_dataset = False\nGLOBAL_VERBOSE = 2\n\nimg_size = 512\nlr_epochs_profile = {'RAMPUP':3, 'SUSTAIN':2, 'DECAY':35}\n# lr_epochs_profile = {'RAMPUP':1, 'SUSTAIN':1, 'DECAY':1}\n\nIMAGE_SIZE = [img_size, img_size] # at this size, a GPU will run out of memory. Use the TPU\n\nEPOCHS = lr_epochs_profile['RAMPUP'] + lr_epochs_profile['SUSTAIN'] + lr_epochs_profile['DECAY']\n\nBATCH_SIZE = 16 * strategy.num_replicas_in_sync\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}\nGCS_PATH = GCS_PATH_SELECT[IMAGE_SIZE[0]]\n\nTRAINING_FILENAMES = tf.io.gfile.glob(GCS_PATH + '/train/*.tfrec')\nVALIDATION_FILENAMES = 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\n\nCLASSES = ['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\n\nNUM_OF_CLASSES = len(CLASSES)\n\n#------------------------------\n# Custom LR schedule\n#------------------------------\nif calculator == COMP_ENV_TPU:\n    LR_MAX = 0.00005 * strategy.num_replicas_in_sync\nelse:\n    LR_MAX = 0.00005\n\nLR_MIN = LR_MAX / 10.0\nLR_START = LR_MAX / 3.0\nLR_START_AMPLITUDE = LR_MAX / 4.0\nLR_RAMPUP_EPOCHS = lr_epochs_profile['RAMPUP']\nLR_SUSTAIN_EPOCHS = lr_epochs_profile['SUSTAIN']\nLR_EXP_DECAY = .8\n\n#------------------------------\n\n@tf.function\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]))\n\nearly_stop_callback = keras.callbacks.EarlyStopping(\n    monitor='val_loss',\n    min_delta=0,\n    patience=8,\n    verbose=1,\n    mode='auto',\n    restore_best_weights=True)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# numpy and matplotlib defaults\nnp.set_printoptions(threshold=15, linewidth=80)\n\ndef batch_to_numpy_images_and_labels(data):\n    images, labels = data\n    numpy_images = images.numpy()\n    numpy_labels = labels.numpy()\n    if numpy_labels.dtype == object: # binary string in this case, these are image ID strings\n        numpy_labels = [None for _ in enumerate(numpy_images)]\n    # If no labels, only image IDs, return None for labels (this is the case for test data)\n    return numpy_images, numpy_labels\n\ndef title_from_label_and_target(label, correct_label):\n    if correct_label is None:\n        return CLASSES[label], True\n    correct = (label == correct_label)\n    return \"{} [{}{}{}]\".format(CLASSES[label], 'OK' if correct else 'NO', u\"\\u2192\" if not correct else '',\n                                CLASSES[correct_label] if not correct else ''), correct\n\ndef display_one_flower(image, title, subplot, red=False, titlesize=16):\n    plt.subplot(*subplot)\n    plt.axis('off')\n    plt.imshow(image)\n    if len(title) > 0:\n        plt.title(title, fontsize=int(titlesize) if not red else int(titlesize/1.2), color='red' if red else 'black', fontdict={'verticalalignment':'center'}, pad=int(titlesize/1.5))\n    return (subplot[0], subplot[1], subplot[2]+1)\n    \ndef display_batch_of_images(databatch, predictions=None):\n    \"\"\"This will work with:\n    display_batch_of_images(images)\n    display_batch_of_images(images, predictions)\n    display_batch_of_images((images, labels))\n    display_batch_of_images((images, labels), predictions)\n    \"\"\"\n    # data\n    images, labels = batch_to_numpy_images_and_labels(databatch)\n    if labels is None:\n        labels = [None for _ in enumerate(images)]\n        \n    # auto-squaring: this will drop data that does not fit into square or square-ish rectangle\n    rows = int(math.sqrt(len(images)))\n    cols = len(images)//rows\n        \n    # size and spacing\n    FIGSIZE = 13.0\n    SPACING = 0.1\n    subplot=(rows,cols,1)\n    if rows < cols:\n        plt.figure(figsize=(FIGSIZE,FIGSIZE/cols*rows))\n    else:\n        plt.figure(figsize=(FIGSIZE/rows*cols,FIGSIZE))\n    \n    # display\n    for i, (image, label) in enumerate(zip(images[:rows*cols], labels[:rows*cols])):\n        title = '' if label is None else CLASSES[np.argmax(label)]\n        correct = True\n        if predictions is not None:\n            title, correct = title_from_label_and_target(predictions[i], label)\n        dynamic_titlesize = FIGSIZE*SPACING/max(rows,cols)*40+3 # magic formula tested to work from 1x1 to 10x10 images\n        subplot = display_one_flower(image, title, subplot, not correct, titlesize=dynamic_titlesize)\n    \n    #layout\n    plt.tight_layout()\n    if label is None and predictions is None:\n        plt.subplots_adjust(wspace=0, hspace=0)\n    else:\n        plt.subplots_adjust(wspace=SPACING, hspace=SPACING)\n    plt.show()\n\ndef display_confusion_matrix(cmat, score, precision, recall):\n    plt.figure(figsize=(15,15))\n    ax = plt.gca()\n    ax.matshow(cmat, cmap='Reds')\n    ax.set_xticks(range(len(CLASSES)))\n    ax.set_xticklabels(CLASSES, fontdict={'fontsize': 7})\n    plt.setp(ax.get_xticklabels(), rotation=45, ha=\"left\", rotation_mode=\"anchor\")\n    ax.set_yticks(range(len(CLASSES)))\n    ax.set_yticklabels(CLASSES, fontdict={'fontsize': 7})\n    plt.setp(ax.get_yticklabels(), rotation=45, ha=\"right\", rotation_mode=\"anchor\")\n    titlestring = \"\"\n    if score is not None:\n        titlestring += 'f1 = {:.3f} '.format(score)\n    if precision is not None:\n        titlestring += '\\nprecision = {:.3f} '.format(precision)\n    if recall is not None:\n        titlestring += '\\nrecall = {:.3f} '.format(recall)\n    if len(titlestring) > 0:\n        ax.text(101, 1, titlestring, fontdict={'fontsize': 18, 'horizontalalignment':'right', 'verticalalignment':'top', 'color':'#804040'})\n    plt.show()\n    \ndef display_training_curves(training, validation, title, subplot):\n    if subplot%10==1: # set up the subplots on the first call\n        plt.subplots(figsize=(10,10), facecolor='#F0F0F0')\n        plt.tight_layout()\n    ax = plt.subplot(subplot)\n    ax.set_facecolor('#F8F8F8')\n    ax.plot(training)\n    ax.plot(validation)\n    ax.set_title('model '+ title)\n    ax.set_ylabel(title)\n    #ax.set_ylim(0.28,1.05)\n    ax.set_xlabel('epoch')\n    ax.legend(['train', 'valid.'])","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def show_training_curves(training_history):\n    display_training_curves(training_history.history['loss'], training_history.history['val_loss'], 'loss', 211)\n    display_training_curves(training_history.history['sparse_categorical_accuracy'], training_history.history['sparse_categorical_accuracy'], 'accuracy', 212)\n\ndef get_cmat_score_precision_recall__one_hot(models, num_of_classes, cmdataset, num_of_images):\n    images_ds = cmdataset.map(lambda image, label: image)\n    labels_ds = cmdataset.map(lambda image, label: label).unbatch()\n    cm_correct_labels = np.array(list(map(np.argmax, next(iter(labels_ds.batch(num_of_images))).numpy()))) # get everything as one batch\n    \n    cm_probabilities = np.zeros(num_of_classes, dtype=float)\n    for model in models:\n        probabilities = model.predict(images_ds)\n        cm_probabilities = cm_probabilities + probabilities\n        \n    cm_predictions = np.argmax(cm_probabilities, axis=-1)\n\n    cmat = confusion_matrix(cm_correct_labels, cm_predictions, labels=range(num_of_classes))\n    score = f1_score(cm_correct_labels, cm_predictions, labels=range(num_of_classes), average='macro')\n    precision = precision_score(cm_correct_labels, cm_predictions, labels=range(num_of_classes), average='macro')\n    recall = recall_score(cm_correct_labels, cm_predictions, labels=range(num_of_classes), average='macro')\n    cmat = (cmat.T / cmat.sum(axis=1)).T # normalized\n    return cmat, score, precision, recall\n\ndef get_cmat_score_precision_recall(models, alphas, num_of_classes, cmdataset, num_of_images, verbose=0):\n    images_ds = cmdataset.map(lambda image, label: image)\n    labels_ds = cmdataset.map(lambda image, label: label).unbatch()\n    cm_correct_labels = next(iter(labels_ds.batch(num_of_images))).numpy() # get everything as one batch\n    \n    cm_probabilities = np.zeros(num_of_classes, dtype=float)\n    for i in range(len(models)):\n        model = models[i]\n        alpha = alphas[i]\n        probabilities = alpha * model.predict(images_ds)\n        cm_probabilities = cm_probabilities + probabilities\n        \n    cm_predictions = np.argmax(cm_probabilities, axis=-1)\n    \n    cmat = confusion_matrix(cm_correct_labels, cm_predictions, labels=range(num_of_classes))\n    score = f1_score(cm_correct_labels, cm_predictions, labels=range(num_of_classes), average='macro')\n    precision = precision_score(cm_correct_labels, cm_predictions, labels=range(num_of_classes), average='macro')\n    recall = recall_score(cm_correct_labels, cm_predictions, labels=range(num_of_classes), average='macro')\n    cmat = (cmat.T / cmat.sum(axis=1)).T # normalized\n    return cmat, score, precision, recall\n\ndef show_confusion_matrix(models, num_of_classes, cmdataset, num_of_images, verbose=0):\n    cmat, score, precision, recall = get_cmat_score_precision_recall(models)\n    print('f1 score: {:.3f}, precision: {:.3f}, recall: {:.3f}'.format(score, precision, recall))\n    display_confusion_matrix(cmat, score, precision, recall)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def 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\ndef 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, one_hot=False):\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    if one_hot:\n        label = tf.one_hot(label, depth=len(CLASSES))\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 force_image_sizes(dataset, image_size):\n    # explicit size needed for TPU\n    reshape_images = lambda image, label: (tf.reshape(image, [*image_size, 3]), label)\n    dataset = dataset.map(reshape_images, num_parallel_calls=AUTO)\n    return dataset\n\ndef load_dataset(filenames, labeled=True, ordered=False):\n    # Read from TFRecords. For optimal performance, reading from multiple files at once and\n    # disregarding 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) # uses 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)\n    # returns a dataset of (image, label) pairs if labeled=True or (image, id) pairs if labeled=False\n    return dataset\n\ndef data_augment(image, one_hot_class):\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    image = tf.image.random_saturation(image, 0.8, 1)\n#     image = tf.image.random_jpeg_quality(image, 80, 100)\n    image = tf.image.random_brightness(image, 0.1)\n    image = tf.image.random_contrast(image, 0.8, 1)\n    return image, one_hot_class\n\ndef tf_tut_augment(image, label):\n    image = tf.image.random_flip_left_right(image)\n    image = tf.image.random_flip_up_down(image)\n    \n    image = tf.image.random_saturation(image, 0.8, 1)\n    \n    image = tf.image.random_brightness(image, max_delta=0.3)\n    image = tf.image.random_contrast(image, 0.8, 1)\n    \n    resize_factor = (img_size // 5) * 6\n    image = tf.image.resize_with_crop_or_pad(image, resize_factor, resize_factor)\n    image = tf.image.random_crop(image, size=[*IMAGE_SIZE, 3])\n    \n    tf.image.random_jpeg_quality(image, 20, 100)\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(ordered=False):\n    dataset = load_dataset(VALIDATION_FILENAMES, labeled=True, ordered=ordered)\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 calc_datasets_parameters(training_filenames, validation_filenames, test_filenames, batch_size):\n    NUM_TRAINING_IMAGES = count_data_items(training_filenames)\n    NUM_VALIDATION_IMAGES = count_data_items(validation_filenames)\n    NUM_TEST_IMAGES = count_data_items(test_filenames)\n    STEPS_PER_EPOCH = NUM_TRAINING_IMAGES // batch_size\n    print('BASE Dataset: {} training images, {} validation images, {} unlabeled test images'.format(NUM_TRAINING_IMAGES, NUM_VALIDATION_IMAGES, NUM_TEST_IMAGES))\n    return NUM_TRAINING_IMAGES, NUM_VALIDATION_IMAGES, NUM_TEST_IMAGES, STEPS_PER_EPOCH\n\ndef get_raw_half_splitted_training_datasets(training_filenames):\n    half_of_images = count_data_items(training_filenames) // 2\n    dataset = load_dataset(training_filenames, labeled=True)\n    dataset.shuffle(2048)\n    ds_1 = dataset.take(half_of_images)\n    ds_2 = dataset.skip(half_of_images)\n    print(\"Splitted on 2 datasets with {} images\".format(half_of_images))\n    return ds_1, ds_2\n\ndef prepare_weights(trainingdataset, num_of_classes, calculator, one_hot=False, verbose=0):\n    counters = np.zeros(num_of_classes, dtype=float)\n    total_samples = 0.0\n    \n    for batch_data in trainingdataset:\n        _, labels = batch_data\n        if one_hot:\n            np.add(counters, labels.numpy(), out=counters)\n        else:\n            counters[labels.numpy()] += 1\n        total_samples += 1\n\n    weights = total_samples / (num_of_classes * counters)\n    weights = weights / weights.min()\n    \n    if calculator == COMP_ENV_TPU:\n        classweights = weights.tolist()\n    else:\n        classweights = dict(enumerate(weights))\n    \n    if verbose == 1 or verbose ==2:\n        print(\"Weihgting. Counters, min : {},  max : {}\".format(counters.min(),counters.max()))\n        \n    if verbose ==2:\n        print('Ordered counts:')\n        print(np.sort(counters))\n        print('Weights:')\n        print(classweights)\n    return classweights","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":" def get_model_EfficientNetB7():\n        return efn.EfficientNetB7(weights='noisy-student', include_top=False, input_shape=[*IMAGE_SIZE, 3])\n\ndef get_model_DenseNet201():\n        return keras.applications.DenseNet201(weights='imagenet', include_top=False, input_shape=[*IMAGE_SIZE, 3])\n\ndef get_model_Xception():\n        return keras.applications.Xception(weights='imagenet', include_top=False, input_shape=[*IMAGE_SIZE, 3])\n\ndef get_model_ResNet50V2():\n        return keras.applications.ResNet50V2(weights='imagenet', include_top=False, input_shape=[*IMAGE_SIZE, 3])","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def create_model(pretrained_model, num_of_classes):\n    pretrained_model.trainable = True\n\n    model = tf.keras.Sequential([\n        pretrained_model,\n        keras.layers.GlobalAveragePooling2D(),\n        keras.layers.Dense(NUM_OF_CLASSES, activation=None, name='Output_features'),   # , kernel_regularizer=keras.regularizers.l2(0.05)\n        keras.layers.Softmax()\n    ])\n    return model\n\ndef create_Ranger_optimizer(lr, min_lr, total_steps):\n    radam = tfa.optimizers.RectifiedAdam(\n        lr=lr,\n        total_steps=total_steps,\n        warmup_proportion=0.1,\n        min_lr=min_lr,\n    )\n    ranger = tfa.optimizers.Lookahead(radam, sync_period=6, slow_step_size=0.5)\n    return ranger\n\n\ndef create_model_1(num_of_classes):\n    model = create_model(get_model_EfficientNetB7(), num_of_classes)\n    \n    model.compile(\n        optimizer = keras.optimizers.Adam(),\n        loss = keras.losses.SparseCategoricalCrossentropy(from_logits=False),\n        metrics=['sparse_categorical_accuracy']\n    )\n    return model\n\ndef create_model_2(num_of_classes):\n    model = create_model(get_model_DenseNet201(), num_of_classes)\n    \n    model.compile(\n        optimizer = keras.optimizers.Adam(),\n        loss = keras.losses.SparseCategoricalCrossentropy(from_logits=False),\n        metrics=['sparse_categorical_accuracy']\n    )\n    return model\n","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Prepare Dataset","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"G_NUM_TRAINING_IMAGES, G_NUM_VALIDATION_IMAGES, G_NUM_TEST_IMAGES, G_STEPS_PER_EPOCH = calc_datasets_parameters(\n    TRAINING_FILENAMES, VALIDATION_FILENAMES, TEST_FILENAMES, BATCH_SIZE)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"if use_half_splitted_dataset:\n    raw_training_ds_1, raw_training_ds_2 = get_raw_half_splitted_training_datasets(TRAINING_FILENAMES)\n    steps_per_epoch = G_STEPS_PER_EPOCH // 2\n\n    classweights_1 = prepare_weights(raw_training_ds_1, NUM_OF_CLASSES, calculator, one_hot=False, verbose=0)\n    classweights_2 = prepare_weights(raw_training_ds_2, NUM_OF_CLASSES, calculator, one_hot=False, verbose=0)\n\n    training_dataset_1 = get_training_dataset(raw_training_ds_1)\n    training_dataset_2 = get_training_dataset(raw_training_ds_2)\nelse:\n    raw_training_dataset = load_dataset(TRAINING_FILENAMES, labeled=True)\n    steps_per_epoch = G_STEPS_PER_EPOCH\n\n    classweights_1 = prepare_weights(raw_training_dataset, NUM_OF_CLASSES, calculator, one_hot=False, verbose=0)\n    classweights_2 = classweights_1\n\n    training_dataset_1 = get_training_dataset(raw_training_dataset)\n    training_dataset_2 = training_dataset_1","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"### Validation dataset","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"validation_dataset = get_validation_dataset(ordered=True)\ncm_dataset = get_validation_dataset(ordered=True)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# TRAIN","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"print(\"Image size : {}\".format(IMAGE_SIZE))\nprint(\"EPOCHS : {}\".format(EPOCHS))\nprint(\"Steps per epoch : {}\".format(steps_per_epoch))\n","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# numpy and matplcm_datasetotlib defaults\nnp.set_printoptions(threshold=sys.maxsize, linewidth=80)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"gc.collect()\nwith strategy.scope():\n    model_1 = create_model_1(NUM_OF_CLASSES)\nmodel_1.summary()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"history_1 = model_1.fit(\n    training_dataset_1,\n    validation_data=validation_dataset,\n    steps_per_epoch=int(steps_per_epoch),\n    epochs=EPOCHS,\n    class_weight = classweights_1,\n    callbacks=[lr_callback, early_stop_callback],\n    verbose=GLOBAL_VERBOSE,\n)\n\n# model_1.save('model_1.h5')","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# show_training_curves(history_1)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"gc.collect()\nwith strategy.scope():\n    model_2 = create_model_2(NUM_OF_CLASSES)\nmodel_2.summary()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"history_2 = model_2.fit(\n    training_dataset_2,\n    validation_data=validation_dataset,\n    steps_per_epoch=int(steps_per_epoch),\n    epochs=EPOCHS,\n    class_weight = classweights_1,\n    callbacks=[lr_callback, early_stop_callback],\n    verbose=GLOBAL_VERBOSE,\n)\n\n# model_2.save('model_2__DenseNet201.h5')","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# keras.utils.plot_model(model_1, 'model.png', show_shapes=True)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# keras.utils.plot_model(model_2, 'model.png', show_shapes=True)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Finding best alpha","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"if not SKIP_VALIDATION:\n    cmdataset = get_validation_dataset(ordered=True) # since we are splitting the dataset and iterating separately on images and labels, order matters.\n    images_ds = cmdataset.map(lambda image, label: image)\n    labels_ds = cmdataset.map(lambda image, label: label).unbatch()\n    cm_correct_labels = next(iter(labels_ds.batch(G_NUM_VALIDATION_IMAGES))).numpy() # get everything as one batch\n    m1 = model_1.predict(images_ds)\n    m2 = model_2.predict(images_ds)\n    scores = []\n    N = 100\n    for alpha in np.linspace(0, 1, N):\n        cm_probabilities = alpha * m1 + (1 - alpha) * m2\n        cm_predictions = np.argmax(cm_probabilities, axis=-1)\n        scores.append(f1_score(cm_correct_labels, cm_predictions, labels=range(len(CLASSES)), average='macro'))\n        \n    plt.plot(scores)\n    best_alpha = np.argmax(scores) / N\n    cm_probabilities = best_alpha * m1 + (1 - best_alpha) * m2\n    cm_predictions = np.argmax(cm_probabilities, axis=-1)\nelse:\n    best_alpha = 0.55\n\nprint(best_alpha)\n\nalpha_1 = best_alpha\nalpha_2 = 1.0 - alpha_1","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"alphas = [alpha_1, alpha_2]\ncmat, score_1, precision_1, recall_1 = get_cmat_score_precision_recall([model_1], [1.0], NUM_OF_CLASSES, cm_dataset, G_NUM_VALIDATION_IMAGES)\ncmat, score_2, precision_2, recall_2 = get_cmat_score_precision_recall([model_2], [1.0], NUM_OF_CLASSES, cm_dataset, G_NUM_VALIDATION_IMAGES)\ncmat, score_12, precision_12, recall_12 = get_cmat_score_precision_recall([model_1, model_2], alphas, NUM_OF_CLASSES, cm_dataset, G_NUM_VALIDATION_IMAGES)\n\nprint(\"MODEL___1  F1 : {}, precision : {}, Recall : {}\".format(score_1, precision_1, recall_1))\nprint(\"MODEL___2  F1 : {}, precision : {}, Recall : {}\".format(score_2, precision_2, recall_2))\nprint(\"MODEL_1+2  F1 : {}, precision : {}, Recall : {}\".format(score_12, precision_12, recall_12))\n","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Predictions","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"print('Computing predictions...')\ntest_ds = get_test_dataset(ordered=True) # since we are splitting the dataset and iterating separately on images and ids, order matters.\ntest_images_ds = test_ds.map(lambda image, idnum: image)\n\nprobabilities_1 = alpha_1 * model_1.predict(test_images_ds)\nprobabilities_2 = alpha_2 * model_2.predict(test_images_ds)\n\nprobabilities = probabilities_1 + probabilities_2\n\npredictions = np.argmax(probabilities, axis=-1)\n\nprint('Generating submission.csv file...')\ntest_ids_ds = test_ds.map(lambda image, idnum: idnum).unbatch()\ntest_ids = next(iter(test_ids_ds.batch(G_NUM_TEST_IMAGES))).numpy().astype('U') # all in one batch\nnp.savetxt('submission.csv', np.rec.fromarrays([test_ids, predictions]), fmt=['%s', '%d'], delimiter=',', header='id,label', comments='')\n!head submission.csv","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# STOP","execution_count":null},{"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":4}