{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.10.13","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"tpu1vmV38","dataSources":[{"sourceId":21154,"databundleVersionId":1243559,"sourceType":"competition"}],"dockerImageVersionId":30589,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"!pip install -q efficientnet","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2023-11-17T07:32:23.694474Z","iopub.execute_input":"2023-11-17T07:32:23.694727Z","iopub.status.idle":"2023-11-17T07:32:31.425997Z","shell.execute_reply.started":"2023-11-17T07:32:23.694699Z","shell.execute_reply":"2023-11-17T07:32:31.425026Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import math, re, os, random\nimport tensorflow as tf\nimport tensorflow_addons as tfa\nimport efficientnet.tfkeras as efn\nimport numpy as np\nfrom matplotlib import pyplot as plt\nfrom kaggle_datasets import KaggleDatasets\nfrom sklearn.metrics import f1_score, precision_score, recall_score, confusion_matrix\n\nprint(\"TF version \" + tf.__version__)","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:31.427685Z","iopub.execute_input":"2023-11-17T07:32:31.427945Z","iopub.status.idle":"2023-11-17T07:32:48.616576Z","shell.execute_reply.started":"2023-11-17T07:32:31.427919Z","shell.execute_reply":"2023-11-17T07:32:48.615563Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"AUTO = tf.data.experimental.AUTOTUNE\n\ntpu = tf.distribute.cluster_resolver.TPUClusterResolver()\ntf.config.experimental_connect_to_cluster(tpu)\ntf.tpu.experimental.initialize_tpu_system(tpu)\nstrategy = tf.distribute.experimental.TPUStrategy(tpu)\n\nprint(\"Number of accelerators: \", strategy.num_replicas_in_sync)","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:48.617687Z","iopub.execute_input":"2023-11-17T07:32:48.618191Z","iopub.status.idle":"2023-11-17T07:32:58.339122Z","shell.execute_reply.started":"2023-11-17T07:32:48.618146Z","shell.execute_reply":"2023-11-17T07:32:58.338159Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"IMAGE_SIZE = [331, 331] # 192, 224, 331, 512\nEPOCHS = 35\nBATCH_SIZE = 16* strategy.num_replicas_in_sync","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.341101Z","iopub.execute_input":"2023-11-17T07:32:58.341373Z","iopub.status.idle":"2023-11-17T07:32:58.345112Z","shell.execute_reply.started":"2023-11-17T07:32:58.341346Z","shell.execute_reply":"2023-11-17T07:32:58.344347Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"GCS_DS_PATH = KaggleDatasets().get_gcs_path('tpu-getting-started')\nGCS_DS_PATH_EXT = KaggleDatasets().get_gcs_path('tf-flower-photo-tfrec') # External dataset\n\nGCS_PATH_SELECT = { \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\nGCS_PATH_SELECT_EXT = {\n    192: '/tfrecords-jpeg-192x192',\n    224: '/tfrecords-jpeg-224x224',\n    331: '/tfrecords-jpeg-331x331',\n    512: '/tfrecords-jpeg-512x512'\n}\nGCS_PATH_EXT = GCS_PATH_SELECT_EXT[IMAGE_SIZE[0]]\n","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.346051Z","iopub.execute_input":"2023-11-17T07:32:58.346302Z","iopub.status.idle":"2023-11-17T07:32:58.363570Z","shell.execute_reply.started":"2023-11-17T07:32:58.346276Z","shell.execute_reply":"2023-11-17T07:32:58.362831Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"IMAGENET_FILES = tf.io.gfile.glob(GCS_DS_PATH_EXT + '/imagenet' + GCS_PATH_EXT + '/*.tfrec')\nINATURELIST_FILES = tf.io.gfile.glob(GCS_DS_PATH_EXT + '/inaturalist' + GCS_PATH_EXT + '/*.tfrec')\nOPENIMAGE_FILES = tf.io.gfile.glob(GCS_DS_PATH_EXT + '/openimage' + GCS_PATH_EXT + '/*.tfrec')\nOXFORD_FILES = tf.io.gfile.glob(GCS_DS_PATH_EXT + '/oxford_102' + GCS_PATH_EXT + '/*.tfrec')\nTENSORFLOW_FILES = tf.io.gfile.glob(GCS_DS_PATH_EXT + '/tf_flowers' + GCS_PATH_EXT + '/*.tfrec')","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.364431Z","iopub.execute_input":"2023-11-17T07:32:58.364642Z","iopub.status.idle":"2023-11-17T07:32:58.375461Z","shell.execute_reply.started":"2023-11-17T07:32:58.364619Z","shell.execute_reply":"2023-11-17T07:32:58.374684Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ADDITIONAL_TRAINING_FILENAMES = IMAGENET_FILES + INATURELIST_FILES + OPENIMAGE_FILES + OXFORD_FILES + TENSORFLOW_FILES  \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\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')\n\nTRAINING_FILENAMES = TRAINING_FILENAMES + ADDITIONAL_TRAINING_FILENAMES","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.376528Z","iopub.execute_input":"2023-11-17T07:32:58.376785Z","iopub.status.idle":"2023-11-17T07:32:58.413897Z","shell.execute_reply.started":"2023-11-17T07:32:58.376759Z","shell.execute_reply":"2023-11-17T07:32:58.413090Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Visualization functions\n","metadata":{}},{"cell_type":"code","source":"np.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    \n    if numpy_labels.dtype == object: # binary string in this case, these are image ID strings\n        # If no labels, i.e Test data return None for labels\n        numpy_labels = [None for _ in enumerate(numpy_images)]\n    \n    return numpy_images, numpy_labels","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.414826Z","iopub.execute_input":"2023-11-17T07:32:58.415079Z","iopub.status.idle":"2023-11-17T07:32:58.420135Z","shell.execute_reply.started":"2023-11-17T07:32:58.415053Z","shell.execute_reply":"2023-11-17T07:32:58.419344Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def title_from_label_and_target(label, correct_label):\n    if correct_label is None:\n        return CLASSES[label], True \n    \n    correct = (label == correct_label)\n    \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","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.421145Z","iopub.execute_input":"2023-11-17T07:32:58.421427Z","iopub.status.idle":"2023-11-17T07:32:58.431812Z","shell.execute_reply.started":"2023-11-17T07:32:58.421401Z","shell.execute_reply":"2023-11-17T07:32:58.431062Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def 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)","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.434822Z","iopub.execute_input":"2023-11-17T07:32:58.435093Z","iopub.status.idle":"2023-11-17T07:32:58.442735Z","shell.execute_reply.started":"2023-11-17T07:32:58.435068Z","shell.execute_reply":"2023-11-17T07:32:58.442036Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def 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[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","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.443579Z","iopub.execute_input":"2023-11-17T07:32:58.443801Z","iopub.status.idle":"2023-11-17T07:32:58.457817Z","shell.execute_reply.started":"2023-11-17T07:32:58.443780Z","shell.execute_reply":"2023-11-17T07:32:58.457037Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"\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    ","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.458720Z","iopub.execute_input":"2023-11-17T07:32:58.458943Z","iopub.status.idle":"2023-11-17T07:32:58.468618Z","shell.execute_reply.started":"2023-11-17T07:32:58.458914Z","shell.execute_reply":"2023-11-17T07:32:58.467874Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def 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.'])","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.469435Z","iopub.execute_input":"2023-11-17T07:32:58.469643Z","iopub.status.idle":"2023-11-17T07:32:58.482317Z","shell.execute_reply.started":"2023-11-17T07:32:58.469621Z","shell.execute_reply":"2023-11-17T07:32:58.481615Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Random Blockout (Erasing) Augmentation\nRandom Erasing is a data augmentation method for training the convolutional neural network (CNN), which randomly selects a rectangle region in an image and erases its pixels with random values. In this process, training images with various levels of occlusion are generated, which reduces the risk of over-fitting and makes the model robust to occlusion.","metadata":{}},{"cell_type":"code","source":"def random_erasing(img, sl=0.1, sh=0.2, rl=0.4, p=0.3):\n    h = tf.shape(img)[0]\n    w = tf.shape(img)[1]\n    c = tf.shape(img)[2]\n    origin_area = tf.cast(h*w, tf.float32)\n\n    e_size_l = tf.cast(tf.round(tf.sqrt(origin_area * sl * rl)), tf.int32)\n    e_size_h = tf.cast(tf.round(tf.sqrt(origin_area * sh / rl)), tf.int32)\n\n    e_height_h = tf.minimum(e_size_h, h)\n    e_width_h = tf.minimum(e_size_h, w)\n\n    erase_height = tf.random.uniform(shape=[], minval=e_size_l, maxval=e_height_h, dtype=tf.int32)\n    erase_width = tf.random.uniform(shape=[], minval=e_size_l, maxval=e_width_h, dtype=tf.int32)\n\n    erase_area = tf.zeros(shape=[erase_height, erase_width, c])\n    erase_area = tf.cast(erase_area, tf.uint8)\n\n    pad_h = h - erase_height\n    pad_top = tf.random.uniform(shape=[], minval=0, maxval=pad_h, dtype=tf.int32)\n    pad_bottom = pad_h - pad_top\n\n    pad_w = w - erase_width\n    pad_left = tf.random.uniform(shape=[], minval=0, maxval=pad_w, dtype=tf.int32)\n    pad_right = pad_w - pad_left\n\n    erase_mask = tf.pad([erase_area], [[0,0],[pad_top, pad_bottom], [pad_left, pad_right], [0,0]], constant_values=1)\n    erase_mask = tf.squeeze(erase_mask, axis=0)\n    erased_img = tf.multiply(tf.cast(img,tf.float32), tf.cast(erase_mask, tf.float32))\n\n    return tf.cond(tf.random.uniform([], 0, 1) > p, lambda: tf.cast(img, img.dtype), lambda:  tf.cast(erased_img, img.dtype))","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.483207Z","iopub.execute_input":"2023-11-17T07:32:58.483424Z","iopub.status.idle":"2023-11-17T07:32:58.493310Z","shell.execute_reply.started":"2023-11-17T07:32:58.483402Z","shell.execute_reply":"2023-11-17T07:32:58.492614Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Dataset Functions","metadata":{}},{"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\n    image = tf.reshape(image, [*IMAGE_SIZE,3])\n    return image","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.494167Z","iopub.execute_input":"2023-11-17T07:32:58.494378Z","iopub.status.idle":"2023-11-17T07:32:58.506813Z","shell.execute_reply.started":"2023-11-17T07:32:58.494356Z","shell.execute_reply":"2023-11-17T07:32:58.505882Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def onehot(image,label):\n    return image,tf.one_hot(label, len(CLASSES))","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.507780Z","iopub.execute_input":"2023-11-17T07:32:58.508030Z","iopub.status.idle":"2023-11-17T07:32:58.516791Z","shell.execute_reply.started":"2023-11-17T07:32:58.508004Z","shell.execute_reply":"2023-11-17T07:32:58.515972Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def read_labeled_tfrecord(example):\n    LABELED_TFREC_FORMAT = {\n        \"image\": tf.io.FixedLenFeature([], tf.string), \n        \"class\": tf.io.FixedLenFeature([], tf.int64),  \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","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.517726Z","iopub.execute_input":"2023-11-17T07:32:58.517978Z","iopub.status.idle":"2023-11-17T07:32:58.526451Z","shell.execute_reply.started":"2023-11-17T07:32:58.517952Z","shell.execute_reply":"2023-11-17T07:32:58.525617Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def read_unlabeled_tfrecord(example):\n    UNLABELED_TFREC_FORMAT = {\n        \"image\": tf.io.FixedLenFeature([], tf.string),\n        \"id\": tf.io.FixedLenFeature([], tf.string),  \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)","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.527346Z","iopub.execute_input":"2023-11-17T07:32:58.527593Z","iopub.status.idle":"2023-11-17T07:32:58.540269Z","shell.execute_reply.started":"2023-11-17T07:32:58.527568Z","shell.execute_reply":"2023-11-17T07:32:58.539587Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def 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","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.541405Z","iopub.execute_input":"2023-11-17T07:32:58.541618Z","iopub.status.idle":"2023-11-17T07:32:58.552608Z","shell.execute_reply.started":"2023-11-17T07:32:58.541595Z","shell.execute_reply":"2023-11-17T07:32:58.551898Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def 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\n    image = tf.image.random_flip_left_right(image)\n    image = random_erasing(image)\n    return image, label","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.553410Z","iopub.execute_input":"2023-11-17T07:32:58.553617Z","iopub.status.idle":"2023-11-17T07:32:58.562982Z","shell.execute_reply.started":"2023-11-17T07:32:58.553596Z","shell.execute_reply":"2023-11-17T07:32:58.562309Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def data_hflip(image, idnum):\n    image = tf.image.flip_left_right(image)\n    return image, idnum","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.563779Z","iopub.execute_input":"2023-11-17T07:32:58.563996Z","iopub.status.idle":"2023-11-17T07:32:58.576231Z","shell.execute_reply.started":"2023-11-17T07:32:58.563974Z","shell.execute_reply":"2023-11-17T07:32:58.575472Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def get_training_dataset(do_onehot=False):\n    dataset = load_dataset(TRAINING_FILENAMES, labeled=True)\n    dataset = dataset.map(data_augment, num_parallel_calls=AUTO)\n    if do_onehot:\n        dataset = dataset.map(onehot, 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","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.577130Z","iopub.execute_input":"2023-11-17T07:32:58.577387Z","iopub.status.idle":"2023-11-17T07:32:58.586583Z","shell.execute_reply.started":"2023-11-17T07:32:58.577361Z","shell.execute_reply":"2023-11-17T07:32:58.585791Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def get_validation_dataset(ordered=False, do_onehot=False):\n    dataset = load_dataset(VALIDATION_FILENAMES, labeled=True, ordered=ordered)\n    if do_onehot:\n        dataset = dataset.map(onehot, num_parallel_calls=AUTO)\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","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.587494Z","iopub.execute_input":"2023-11-17T07:32:58.587737Z","iopub.status.idle":"2023-11-17T07:32:58.595861Z","shell.execute_reply.started":"2023-11-17T07:32:58.587711Z","shell.execute_reply":"2023-11-17T07:32:58.595190Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def get_test_dataset(ordered=False, augmented=False):\n    dataset = load_dataset(TEST_FILENAMES, labeled=False, ordered=ordered)\n    dataset = dataset.map(data_hflip, num_parallel_calls=AUTO)\n    dataset = dataset.batch(BATCH_SIZE)\n    dataset = dataset.prefetch(AUTO) # prefetch next batch while training (autotune prefetch buffer size)\n    return dataset","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.596773Z","iopub.execute_input":"2023-11-17T07:32:58.597031Z","iopub.status.idle":"2023-11-17T07:32:58.606053Z","shell.execute_reply.started":"2023-11-17T07:32:58.597005Z","shell.execute_reply":"2023-11-17T07:32:58.605318Z"},"trusted":true},"execution_count":null,"outputs":[]},{"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)","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.607017Z","iopub.execute_input":"2023-11-17T07:32:58.607273Z","iopub.status.idle":"2023-11-17T07:32:58.615167Z","shell.execute_reply.started":"2023-11-17T07:32:58.607247Z","shell.execute_reply":"2023-11-17T07:32:58.614477Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"NUM_TRAINING_IMAGES = count_data_items(TRAINING_FILENAMES)\nNUM_VALIDATION_IMAGES = count_data_items(VALIDATION_FILENAMES)\nNUM_TEST_IMAGES = count_data_items(TEST_FILENAMES)\nSTEPS_PER_EPOCH = NUM_TRAINING_IMAGES // BATCH_SIZE\nVALIDATION_STEPS = -(-NUM_VALIDATION_IMAGES // BATCH_SIZE) # The \"-(-//)\" trick rounds up instead of down :-)\nTEST_STEPS = -(-NUM_TEST_IMAGES // BATCH_SIZE)             # The \"-(-//)\" trick rounds up instead of down :-)\nprint('Dataset: {} training images, {} validation images, {} unlabeled test images'.format(NUM_TRAINING_IMAGES, NUM_VALIDATION_IMAGES, NUM_TEST_IMAGES))","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.616118Z","iopub.execute_input":"2023-11-17T07:32:58.616384Z","iopub.status.idle":"2023-11-17T07:32:58.630978Z","shell.execute_reply.started":"2023-11-17T07:32:58.616358Z","shell.execute_reply":"2023-11-17T07:32:58.630315Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"TRAINING_FILENAMES = TRAINING_FILENAMES + VALIDATION_FILENAMES\nprint(\"Training data shapes:\")\nfor image, label in get_training_dataset().take(3):\n    print(image.numpy().shape, label.numpy().shape)\nprint(\"Training data label examples:\", label.numpy())\nprint(\"Validation data shapes:\")\nfor image, label in get_validation_dataset().take(3):\n    print(image.numpy().shape, label.numpy().shape)\nprint(\"Validation data label examples:\", label.numpy())\nprint(\"Test data shapes:\")\nfor image, idnum in get_test_dataset().take(3):\n    print(image.numpy().shape, idnum.numpy().shape)\nprint(\"Test data IDs:\", idnum.numpy().astype('U'))","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:32:58.631819Z","iopub.execute_input":"2023-11-17T07:32:58.632034Z","iopub.status.idle":"2023-11-17T07:33:01.883349Z","shell.execute_reply.started":"2023-11-17T07:32:58.632012Z","shell.execute_reply":"2023-11-17T07:33:01.882125Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"training_dataset = get_training_dataset()\ntraining_dataset = training_dataset.unbatch().batch(20)\ntrain_batch = iter(training_dataset)","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:33:01.887862Z","iopub.execute_input":"2023-11-17T07:33:01.888230Z","iopub.status.idle":"2023-11-17T07:33:02.019010Z","shell.execute_reply.started":"2023-11-17T07:33:01.888189Z","shell.execute_reply":"2023-11-17T07:33:02.017973Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"display_batch_of_images(next(train_batch))","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:33:02.020209Z","iopub.execute_input":"2023-11-17T07:33:02.020490Z","iopub.status.idle":"2023-11-17T07:33:04.723651Z","shell.execute_reply.started":"2023-11-17T07:33:02.020462Z","shell.execute_reply":"2023-11-17T07:33:04.722694Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_dataset = get_test_dataset()\ntest_dataset = test_dataset.unbatch().batch(20)\ntest_batch = iter(test_dataset)\ndisplay_batch_of_images(next(test_batch))\n","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:33:04.724728Z","iopub.execute_input":"2023-11-17T07:33:04.724986Z","iopub.status.idle":"2023-11-17T07:33:06.457188Z","shell.execute_reply.started":"2023-11-17T07:33:04.724959Z","shell.execute_reply":"2023-11-17T07:33:06.455910Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# custom LR scheduler\n# 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","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:33:06.458419Z","iopub.execute_input":"2023-11-17T07:33:06.458696Z","iopub.status.idle":"2023-11-17T07:33:06.464064Z","shell.execute_reply.started":"2023-11-17T07:33:06.458668Z","shell.execute_reply":"2023-11-17T07:33:06.463324Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"lr_callback = tf.keras.callbacks.LearningRateScheduler(lrfn, verbose =1)\n\nrng = [i for i in range(25 if EPOCHS<25 else 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]))","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:33:06.464918Z","iopub.execute_input":"2023-11-17T07:33:06.465131Z","iopub.status.idle":"2023-11-17T07:33:06.932498Z","shell.execute_reply.started":"2023-11-17T07:33:06.465109Z","shell.execute_reply":"2023-11-17T07:33:06.931553Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def get_model(base_model):\n    base_model.trainable = True\n    \n    model = tf.keras.Sequential([\n        base_model,\n        tf.keras.layers.Dense(len(CLASSES), activation = 'softmax')\n    ])\n    \n    model.compile(\n        optimizer = 'adam',\n        loss = 'categorical_crossentropy',\n        metrics = [tfa.metrics.F1Score(len(CLASSES), average= 'macro')],\n        # NEW on TPU in TensorFlow 24: sending multiple batches to the TPU at once saves communications\n        # overheads and allows the XLA compiler to unroll the loop on TPU and optimize hardware utilization.\n        steps_per_execution = 16\n    )\n    \n    return model","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:33:06.933621Z","iopub.execute_input":"2023-11-17T07:33:06.933923Z","iopub.status.idle":"2023-11-17T07:33:06.939634Z","shell.execute_reply.started":"2023-11-17T07:33:06.933884Z","shell.execute_reply":"2023-11-17T07:33:06.938742Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"with strategy.scope():\n    enet = efn.EfficientNetB7(\n        weights = 'noisy-student',\n        include_top = False,\n        pooling = 'avg',\n        input_shape = (*IMAGE_SIZE,3)\n    )\n    \n    model1 = get_model(enet)\nmodel1.summary()","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:33:06.940596Z","iopub.execute_input":"2023-11-17T07:33:06.940851Z","iopub.status.idle":"2023-11-17T07:34:10.068920Z","shell.execute_reply.started":"2023-11-17T07:33:06.940826Z","shell.execute_reply":"2023-11-17T07:34:10.067934Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"chk_callback1 = tf.keras.callbacks.ModelCheckpoint('effnetb7_best.h5',\n                                                   save_weights_only=True,\n                                                   monitor='val_f1_score',\n                                                   mode='max',\n                                                   save_best_only=True,\n                                                   verbose=1)\n         \nhistory1 = model1.fit(get_training_dataset(do_onehot=True), \n                steps_per_epoch=STEPS_PER_EPOCH, \n                epochs=EPOCHS, \n                validation_data=get_validation_dataset(do_onehot=True),\n                validation_steps=VALIDATION_STEPS,\n                callbacks=[lr_callback, chk_callback1],\n                verbose=2)","metadata":{"execution":{"iopub.status.busy":"2023-11-17T07:34:10.070034Z","iopub.execute_input":"2023-11-17T07:34:10.070315Z","iopub.status.idle":"2023-11-17T08:02:35.414059Z","shell.execute_reply.started":"2023-11-17T07:34:10.070289Z","shell.execute_reply":"2023-11-17T08:02:35.412922Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"with strategy.scope():\n    densenet = tf.keras.applications.DenseNet201(weights='imagenet', \n                                                include_top=False,\n                                                pooling='avg',\n                                                input_shape=(*IMAGE_SIZE, 3))\n    model2 = get_model(densenet)\n    \nmodel2.summary()","metadata":{"execution":{"iopub.status.busy":"2023-11-17T08:02:35.416814Z","iopub.execute_input":"2023-11-17T08:02:35.417100Z","iopub.status.idle":"2023-11-17T08:03:37.879470Z","shell.execute_reply.started":"2023-11-17T08:02:35.417057Z","shell.execute_reply":"2023-11-17T08:03:37.878475Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"chk_callback2 = tf.keras.callbacks.ModelCheckpoint('densenet201_best.h5',\n                                                   save_weights_only=True,\n                                                   monitor='val_f1_score',\n                                                   mode='max',\n                                                   save_best_only=True,\n                                                   verbose=1)\n\nhistory2 = model2.fit(get_training_dataset(do_onehot=True), \n                steps_per_epoch=STEPS_PER_EPOCH, \n                epochs=EPOCHS, \n                validation_data=get_validation_dataset(do_onehot=True),\n                validation_steps=VALIDATION_STEPS,\n                callbacks=[lr_callback, chk_callback2],\n                verbose=2)","metadata":{"execution":{"iopub.status.busy":"2023-11-17T08:03:37.880560Z","iopub.execute_input":"2023-11-17T08:03:37.880819Z","iopub.status.idle":"2023-11-17T08:22:48.007590Z","shell.execute_reply.started":"2023-11-17T08:03:37.880792Z","shell.execute_reply":"2023-11-17T08:22:48.006475Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"cmdataset = get_validation_dataset(ordered=True) # since we are splitting the dataset and iterating separately on images and labels, order matters.\nimages_ds = cmdataset.map(lambda image, label: image)\nlabels_ds = cmdataset.map(lambda image, label: label).unbatch()\ncm_correct_labels = next(iter(labels_ds.batch(NUM_VALIDATION_IMAGES))).numpy() # get everything as one batch\nm1 = model1.predict(images_ds, steps=VALIDATION_STEPS)\nm2 = model2.predict(images_ds, steps=VALIDATION_STEPS)\n\ncm_probabilities = (m1 + m2)/2\ncm_predictions = np.argmax(cm_probabilities, axis=-1)\ncmat = confusion_matrix(cm_correct_labels, cm_predictions, labels=range(len(CLASSES)))\nscore = f1_score(cm_correct_labels, cm_predictions, labels=range(len(CLASSES)), average='macro')\nprecision = precision_score(cm_correct_labels, cm_predictions, labels=range(len(CLASSES)), average='macro')\nrecall = recall_score(cm_correct_labels, cm_predictions, labels=range(len(CLASSES)), average='macro')\n#cmat = (cmat.T / cmat.sum(axis=1)).T # normalized\ndisplay_confusion_matrix(cmat, score, precision, recall)\nprint('f1 score: {:.3f}, precision: {:.3f}, recall: {:.3f}'.format(score, precision, recall))","metadata":{"execution":{"iopub.status.busy":"2023-11-17T08:22:48.009621Z","iopub.execute_input":"2023-11-17T08:22:48.009937Z","iopub.status.idle":"2023-11-17T08:24:52.714104Z","shell.execute_reply.started":"2023-11-17T08:22:48.009904Z","shell.execute_reply":"2023-11-17T08:24:52.713003Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def run_inference(model):\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    preds = model.predict(test_images_ds,verbose=0, steps=TEST_STEPS)\n    return preds","metadata":{"execution":{"iopub.status.busy":"2023-11-17T08:24:52.715180Z","iopub.execute_input":"2023-11-17T08:24:52.715457Z","iopub.status.idle":"2023-11-17T08:24:52.719743Z","shell.execute_reply.started":"2023-11-17T08:24:52.715428Z","shell.execute_reply":"2023-11-17T08:24:52.719044Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_ds = get_test_dataset(ordered=True) # since we are splitting the dataset and iterating separately on images and ids, order matters.\n\nprint('Calculating predictions...')\nprobs1 = run_inference(model1)\nprobs2 = run_inference(model2)\nprobabilities = (probs1 + probs2)/2\npredictions = np.argmax(probabilities, axis=-1)\n\nprint('Generating submission file...')\ntest_ids_ds = test_ds.map(lambda image, idnum: idnum).unbatch()\ntest_ids = next(iter(test_ids_ds.batch(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='')","metadata":{"execution":{"iopub.status.busy":"2023-11-17T08:24:52.720567Z","iopub.execute_input":"2023-11-17T08:24:52.720832Z","iopub.status.idle":"2023-11-17T08:25:53.705484Z","shell.execute_reply.started":"2023-11-17T08:24:52.720804Z","shell.execute_reply":"2023-11-17T08:25:53.704455Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}