{"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":"# Flower Classification Model","metadata":{}},{"cell_type":"code","source":"!pip install -q efficientnet\n!pip install -q tensorflow_addons \n!pip install -q scikit-learn","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:00:06.993167Z","iopub.execute_input":"2023-05-02T08:00:06.993580Z","iopub.status.idle":"2023-05-02T08:00:36.848590Z","shell.execute_reply.started":"2023-05-02T08:00:06.993552Z","shell.execute_reply":"2023-05-02T08:00:36.847363Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import math, re, os, random\nimport numpy as np\nimport pandas as pd\nfrom matplotlib import pyplot as plt\nfrom sklearn.metrics import f1_score, precision_score, \\\n                            recall_score, confusion_matrix\n\nimport tensorflow as tf\nfrom tensorflow_addons.metrics import F1Score\nfrom tensorflow.keras import layers as L\nfrom tensorflow.keras import backend as K\nfrom tensorflow.keras import callbacks\nfrom tensorflow.keras import applications as tf_applications\n\nimport efficientnet.tfkeras as efficientnet\nimport tensorflow_addons as tfa\n\nfrom kaggle_datasets import KaggleDatasets\n\nprint(\"TF version \" + tf.__version__)","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2023-05-02T08:00:36.850903Z","iopub.execute_input":"2023-05-02T08:00:36.851238Z","iopub.status.idle":"2023-05-02T08:01:19.676916Z","shell.execute_reply.started":"2023-05-02T08:00:36.851205Z","shell.execute_reply":"2023-05-02T08:01:19.675580Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## TPU detection","metadata":{}},{"cell_type":"code","source":"# Detect TPU, return appropriate distribution strategy\ntry:\n    tpu = tf.distribute.cluster_resolver.TPUClusterResolver() \n    print('Running on TPU ', tpu.master())\nexcept ValueError:\n    tpu = None\n\nif tpu:\n    tf.config.experimental_connect_to_cluster(tpu)\n    tf.tpu.experimental.initialize_tpu_system(tpu)\n    strategy = tf.distribute.TPUStrategy(tpu)\nelse:\n    strategy = tf.distribute.get_strategy() \n\nprint(\"REPLICAS: \", strategy.num_replicas_in_sync)","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:19.682448Z","iopub.execute_input":"2023-05-02T08:01:19.682796Z","iopub.status.idle":"2023-05-02T08:01:28.953491Z","shell.execute_reply.started":"2023-05-02T08:01:19.682770Z","shell.execute_reply":"2023-05-02T08:01:28.952412Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"AUTO = tf.data.experimental.AUTOTUNE","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:28.954742Z","iopub.execute_input":"2023-05-02T08:01:28.955057Z","iopub.status.idle":"2023-05-02T08:01:28.959315Z","shell.execute_reply.started":"2023-05-02T08:01:28.955028Z","shell.execute_reply":"2023-05-02T08:01:28.958536Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Configuration","metadata":{}},{"cell_type":"code","source":"IMAGE_SIZE = [224, 224] # 192, 224, 331, 512\nEPOCHS = 20\nBATCH_SIZE = 64 * strategy.num_replicas_in_sync","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:28.960407Z","iopub.execute_input":"2023-05-02T08:01:28.960791Z","iopub.status.idle":"2023-05-02T08:01:28.978778Z","shell.execute_reply.started":"2023-05-02T08:01:28.960763Z","shell.execute_reply":"2023-05-02T08:01:28.977862Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Data access and classes","metadata":{}},{"cell_type":"markdown","source":"TPUs read data directly from Google Cloud Storage (GCS), so we need to copy the dataset to a GCS bucket co-located with the TPU. To do that, pass the name of a specific dataset to the get_gcs_path function. The name of the dataset is the name of the directory it is mounted in. ","metadata":{}},{"cell_type":"code","source":"!pip install -q tensorflow-gcs-config","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:28.979900Z","iopub.execute_input":"2023-05-02T08:01:28.980267Z","iopub.status.idle":"2023-05-02T08:01:33.588188Z","shell.execute_reply.started":"2023-05-02T08:01:28.980240Z","shell.execute_reply":"2023-05-02T08:01:33.587100Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from kaggle_secrets import UserSecretsClient\nuser_secrets = UserSecretsClient()\nuser_credential = user_secrets.get_gcloud_credential()\nuser_secrets.set_tensorflow_credential(user_credential)","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:33.589567Z","iopub.execute_input":"2023-05-02T08:01:33.589861Z","iopub.status.idle":"2023-05-02T08:01:33.753812Z","shell.execute_reply.started":"2023-05-02T08:01:33.589833Z","shell.execute_reply":"2023-05-02T08:01:33.752572Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from kaggle_datasets import KaggleDatasets\nGCS_DS_PATH = KaggleDatasets().get_gcs_path('tpu-getting-started')\nGCS_DS_PATH_EXT = KaggleDatasets().get_gcs_path('tf-flower-photo-tfrec')\n\nprint(GCS_DS_PATH)\nprint(GCS_DS_PATH_EXT)","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:33.755218Z","iopub.execute_input":"2023-05-02T08:01:33.755575Z","iopub.status.idle":"2023-05-02T08:01:33.761709Z","shell.execute_reply.started":"2023-05-02T08:01:33.755544Z","shell.execute_reply":"2023-05-02T08:01:33.760729Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"GCS_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\n# External data\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\nIMAGENET_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')\n\nADDITIONAL_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') # predictions on this dataset should be submitted for the competition \n\nTRAINING_FILENAMES = TRAINING_FILENAMES + ADDITIONAL_TRAINING_FILENAMES","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:33.765497Z","iopub.execute_input":"2023-05-02T08:01:33.765811Z","iopub.status.idle":"2023-05-02T08:01:33.918497Z","shell.execute_reply.started":"2023-05-02T08:01:33.765783Z","shell.execute_reply":"2023-05-02T08:01:33.917479Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Visualization functions","metadata":{}},{"cell_type":"markdown","source":"A set of functions to visualize data.","metadata":{}},{"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[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.'])","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:33.919816Z","iopub.execute_input":"2023-05-02T08:01:33.920107Z","iopub.status.idle":"2023-05-02T08:01:33.950426Z","shell.execute_reply.started":"2023-05-02T08:01:33.920080Z","shell.execute_reply":"2023-05-02T08:01:33.949520Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Random erasing (blockout) augmentation","metadata":{}},{"cell_type":"code","source":"# https://www.kaggle.com/tusharkendre/tpu-flowers\ndef 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":"2023-05-02T08:01:33.951487Z","iopub.execute_input":"2023-05-02T08:01:33.951746Z","iopub.status.idle":"2023-05-02T08:01:33.968344Z","shell.execute_reply.started":"2023-05-02T08:01:33.951723Z","shell.execute_reply":"2023-05-02T08:01:33.967426Z"},"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  # 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 onehot(image,label):\n    return image,tf.one_hot(label, len(CLASSES))\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    # 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, 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    #image = random_transform(image)\n    image = random_erasing(image)\n    return image, label\n\ndef data_hflip(image, idnum):\n    image = tf.image.flip_left_right(image)\n    return image, idnum\n\ndef 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\n\ndef 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\n\ndef 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\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)\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(f'Dataset: {NUM_TRAINING_IMAGES} training images, {NUM_VALIDATION_IMAGES} validation images, {NUM_TEST_IMAGES} unlabeled test images')","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:33.969710Z","iopub.execute_input":"2023-05-02T08:01:33.969980Z","iopub.status.idle":"2023-05-02T08:01:33.997052Z","shell.execute_reply.started":"2023-05-02T08:01:33.969957Z","shell.execute_reply":"2023-05-02T08:01:33.996024Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Dataset visualizations","metadata":{}},{"cell_type":"code","source":"# data dump\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')) # U=unicode string","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:33.998105Z","iopub.execute_input":"2023-05-02T08:01:33.998365Z","iopub.status.idle":"2023-05-02T08:01:37.570528Z","shell.execute_reply.started":"2023-05-02T08:01:33.998343Z","shell.execute_reply":"2023-05-02T08:01:37.569174Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Peek at training data\ntraining_dataset = get_training_dataset()\ntraining_dataset = training_dataset.unbatch().batch(20)\ntrain_batch = iter(training_dataset)","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:37.571904Z","iopub.execute_input":"2023-05-02T08:01:37.572225Z","iopub.status.idle":"2023-05-02T08:01:37.724577Z","shell.execute_reply.started":"2023-05-02T08:01:37.572196Z","shell.execute_reply":"2023-05-02T08:01:37.723536Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# run this cell again for next set of images\ndisplay_batch_of_images(next(train_batch))","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:37.725905Z","iopub.execute_input":"2023-05-02T08:01:37.726217Z","iopub.status.idle":"2023-05-02T08:01:40.878071Z","shell.execute_reply.started":"2023-05-02T08:01:37.726188Z","shell.execute_reply":"2023-05-02T08:01:40.876841Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# peer at test data\ntest_dataset = get_test_dataset()\ntest_dataset = test_dataset.unbatch().batch(20)\ntest_batch = iter(test_dataset)","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:40.879445Z","iopub.execute_input":"2023-05-02T08:01:40.879746Z","iopub.status.idle":"2023-05-02T08:01:40.970085Z","shell.execute_reply.started":"2023-05-02T08:01:40.879718Z","shell.execute_reply":"2023-05-02T08:01:40.968853Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# run this cell again for next set of images\ndisplay_batch_of_images(next(test_batch))","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:40.971461Z","iopub.execute_input":"2023-05-02T08:01:40.972139Z","iopub.status.idle":"2023-05-02T08:01:43.209212Z","shell.execute_reply.started":"2023-05-02T08:01:40.972108Z","shell.execute_reply":"2023-05-02T08:01:43.207868Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Models and training","metadata":{}},{"cell_type":"code","source":"len_classes = len(CLASSES)\nprint('class length: ', len_classes)","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:43.210646Z","iopub.execute_input":"2023-05-02T08:01:43.210943Z","iopub.status.idle":"2023-05-02T08:01:43.216208Z","shell.execute_reply.started":"2023-05-02T08:01:43.210916Z","shell.execute_reply":"2023-05-02T08:01:43.215158Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"STEPS_PER_EPOCH =  68094 // BATCH_SIZE\nprint('steps per epoch: ', STEPS_PER_EPOCH)","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:43.217456Z","iopub.execute_input":"2023-05-02T08:01:43.217754Z","iopub.status.idle":"2023-05-02T08:01:43.230029Z","shell.execute_reply.started":"2023-05-02T08:01:43.217730Z","shell.execute_reply":"2023-05-02T08:01:43.229124Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Custom LR scheduler","metadata":{}},{"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.\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\n\nlr_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":"2023-05-02T08:01:43.231061Z","iopub.execute_input":"2023-05-02T08:01:43.231741Z","iopub.status.idle":"2023-05-02T08:01:43.449396Z","shell.execute_reply.started":"2023-05-02T08:01:43.231714Z","shell.execute_reply":"2023-05-02T08:01:43.448515Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"early_stopping = tf.keras.callbacks.EarlyStopping(\n    monitor = 'val_f1_score', # metric to monitor\n    patience = 10, # number of epochs with no improvement after which training will be stopped\n    restore_best_weights = True # restore the weights from the epoch with the best monitored metric\n)","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:01:43.450548Z","iopub.execute_input":"2023-05-02T08:01:43.450817Z","iopub.status.idle":"2023-05-02T08:01:43.455558Z","shell.execute_reply.started":"2023-05-02T08:01:43.450793Z","shell.execute_reply":"2023-05-02T08:01:43.454554Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Define output shape and data type\noutput_shape = tf.TensorShape((None, len_classes))\noutput_dtype = tf.float32\n\n# Define output variable\noutputs = tf.TensorSpec(shape = output_shape, dtype = output_dtype)\n\n# Define model\ndef EfficientNet_model():\n    pretrained_model = tf.keras.applications.efficientnet_v2.EfficientNetV2S(\n        weights = 'imagenet', \n        include_top = False, \n        input_shape = [224, 224, 3],\n        pooling = 'avg')\n    pretrained_model.trainable = True\n\n    model = tf.keras.Sequential([\n        pretrained_model,\n        #tf.keras.layers.GlobalAveragePooling2D(),\n        tf.keras.layers.Dense(len_classes, \n                              activation = 'softmax',\n                              dtype = output_dtype, \n                              name = 'output')\n    ])\n\n    return model","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:15:31.490100Z","iopub.execute_input":"2023-05-02T08:15:31.490865Z","iopub.status.idle":"2023-05-02T08:15:31.498950Z","shell.execute_reply.started":"2023-05-02T08:15:31.490827Z","shell.execute_reply":"2023-05-02T08:15:31.497577Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"tf.keras.backend.clear_session()\nos.environ['TF_ENABLE_EAGER_CLIENT_STREAMING_ENQUEUE'] = 'False'\n\n\nwith strategy.scope():\n    eff = EfficientNet_model()\n    eff.compile(optimizer = 'adam',\n                loss = 'categorical_crossentropy',\n                metrics = [tfa.metrics.F1Score(len_classes, average = 'macro')])\n    callback = tf.keras.callbacks.ModelCheckpoint('effnetv2l_best.h5',\n                                              save_weights_only = True,\n                                              monitor = 'val_f1_score',\n                                              mode = 'max',\n                                              save_best_only = True,\n                                              verbose = 1)\n    eff.summary()","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:15:48.734906Z","iopub.execute_input":"2023-05-02T08:15:48.735661Z","iopub.status.idle":"2023-05-02T08:16:32.567835Z","shell.execute_reply.started":"2023-05-02T08:15:48.735629Z","shell.execute_reply":"2023-05-02T08:16:32.566711Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"history = eff.fit(get_training_dataset(do_onehot=True),\n                  steps_per_epoch = STEPS_PER_EPOCH, \n                  epochs = EPOCHS, \n                  callbacks = [lr_callback, early_stopping],\n                  validation_data = get_validation_dataset(do_onehot=True),\n                  validation_steps = VALIDATION_STEPS,\n                  verbose = 2,\n                  workers = 3)","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:04:26.943522Z","iopub.execute_input":"2023-05-02T08:04:26.944440Z","iopub.status.idle":"2023-05-02T08:07:23.041456Z","shell.execute_reply.started":"2023-05-02T08:04:26.944393Z","shell.execute_reply":"2023-05-02T08:07:23.039854Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#history = model.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, callback],\n#                verbose=2)","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:02:18.184648Z","iopub.status.idle":"2023-05-02T08:02:18.185005Z","shell.execute_reply.started":"2023-05-02T08:02:18.184834Z","shell.execute_reply":"2023-05-02T08:02:18.184850Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"display_training_curves(\n    history.history['loss'],\n    history.history['val_loss'],\n    'loss',\n    211,\n)","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:02:18.186047Z","iopub.status.idle":"2023-05-02T08:02:18.186369Z","shell.execute_reply.started":"2023-05-02T08:02:18.186206Z","shell.execute_reply":"2023-05-02T08:02:18.186221Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Find best alpha to ensemble","metadata":{}},{"cell_type":"markdown","source":"Let's find best alpha, the coefficient we will use for ensembling predictions from both models","metadata":{}},{"cell_type":"code","source":"def find_best_alpha(valid_dataset, model_lst):\n    images_ds = valid_dataset.map(lambda image, label: image)\n    labels_ds = valid_dataset.map(lambda image, label: label).unbatch()\n    y_true = next(iter(labels_ds.batch(NUM_VALIDATION_IMAGES))).numpy() # get everything as one batch\n    p = []\n    for model in model_lst:\n        p.append(model.predict(images_ds))\n\n    scores = []\n    for alpha in np.linspace(0,1,100):\n        preds = np.argmax(alpha*p[0]+(1-alpha)*p[1], axis=-1)\n        scores.append(f1_score(y_true, preds, labels=range(len(CLASSES)), average='macro'))\n\n    best_alpha = np.argmax(scores)/100\n    return best_alpha","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:02:18.187676Z","iopub.status.idle":"2023-05-02T08:02:18.188009Z","shell.execute_reply.started":"2023-05-02T08:02:18.187845Z","shell.execute_reply":"2023-05-02T08:02:18.187861Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"valid_ds = get_validation_dataset(ordered=True) # since we are splitting the dataset and iterating separately on images and labels, order matters.\nalpha = find_best_alpha(valid_ds, models)\nprint(f'Best alpha is {alpha}')","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:02:18.189246Z","iopub.status.idle":"2023-05-02T08:02:18.189640Z","shell.execute_reply.started":"2023-05-02T08:02:18.189443Z","shell.execute_reply":"2023-05-02T08:02:18.189462Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Plot confusion matrix and predict on test dataset","metadata":{}},{"cell_type":"code","source":"def predict_ensemble(dataset, model_lst, alpha, steps):\n    print('Calculating predictions...')\n    images_ds = dataset.map(lambda image, idnum: image)\n    probs = []\n    for model in model_lst:\n        p = model.predict(images_ds,verbose=0, steps=steps)\n        probs.append(p)\n    preds = np.argmax(alpha*probs[0] + (1-alpha)*probs[1], axis=-1)\n    return preds","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:02:18.190787Z","iopub.status.idle":"2023-05-02T08:02:18.191108Z","shell.execute_reply.started":"2023-05-02T08:02:18.190945Z","shell.execute_reply":"2023-05-02T08:02:18.190960Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Let's plot confusion matrix to evaluate our model ensemble accuracy.","metadata":{}},{"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.\ncm_predictions = predict_ensemble(valid_ds, models, alpha, steps=VALIDATION_STEPS)\n\nlabels_ds = valid_ds.map(lambda image, label: label).unbatch()\ncm_correct_labels = next(iter(labels_ds.batch(NUM_VALIDATION_IMAGES))).numpy() # get everything as one batch\n\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)","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:02:18.192461Z","iopub.status.idle":"2023-05-02T08:02:18.192816Z","shell.execute_reply.started":"2023-05-02T08:02:18.192631Z","shell.execute_reply":"2023-05-02T08:02:18.192647Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Now we predict on the test dataset and write results to the submission file (submission.csv)","metadata":{}},{"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\npredictions = predict_ensemble(test_ds, models, alpha, steps=TEST_STEPS)\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\n                 \nsub_df = pd.DataFrame({'id': test_ids, 'label': predictions})\nsub_df.to_csv('submission.csv', index=False)","metadata":{"execution":{"iopub.status.busy":"2023-05-02T08:02:18.193850Z","iopub.status.idle":"2023-05-02T08:02:18.194170Z","shell.execute_reply.started":"2023-05-02T08:02:18.194008Z","shell.execute_reply":"2023-05-02T08:02:18.194023Z"},"trusted":true},"execution_count":null,"outputs":[]}]}