{"cells":[{"metadata":{},"cell_type":"markdown","source":"# Step 1: Imports #\n\n"},{"metadata":{"trusted":true},"cell_type":"code","source":"!pip install -q efficientnet\nimport efficientnet.tfkeras as efn\n\nimport math, re, os, random\nimport tensorflow as tf\nimport numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nfrom kaggle_datasets import KaggleDatasets\nimport tensorflow.keras.backend as K\nfrom tensorflow import keras\nfrom functools import partial\nfrom sklearn.model_selection import train_test_split\nfrom tensorflow.keras import models, layers\n\nprint(\"Tensorflow version \" + tf.__version__)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Step 2: Distribution Strategy #\n\nA TPU has eight different *cores* and each of these cores acts as its own accelerator. (A TPU is sort of like having eight GPUs in one machine.) We tell TensorFlow how to make use of all these cores at once through a **distribution strategy**. "},{"metadata":{"trusted":true},"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.experimental.TPUStrategy(tpu)\nelse:\n    strategy = tf.distribute.get_strategy() \n\nprint(\"REPLICAS: \", strategy.num_replicas_in_sync)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"We'll use the distribution strategy when we create our neural network model. Then, TensorFlow will distribute the training among the eight TPU cores by creating eight different *replicas* of the model, one for each core.\n\n# Step 3: Loading the Competition Data #\n\n## Get GCS Path ##\n\n- When used with TPUs, datasets need to be stored in a [Google Cloud Storage bucket](https://cloud.google.com/storage/). \n- You can use data from any public GCS bucket by giving its path just like you would data from `'/kaggle/input'`. \n- The following will retrieve the GCS path for this competition's dataset."},{"metadata":{"trusted":true},"cell_type":"code","source":"from kaggle_datasets import KaggleDatasets\n\nGCS_DS_PATH = KaggleDatasets().get_gcs_path('tpu-getting-started')\nEXT_GCS_DS_PATH = KaggleDatasets().get_gcs_path('tf-flower-photo-tfrec')\nGCS_PATH = GCS_DS_PATH + '/tfrecords-jpeg-512x512'\n\nprint(GCS_DS_PATH) # what do gcs paths look like?","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"## Configure the dataset for performance and set up variables"},{"metadata":{"trusted":true},"cell_type":"code","source":"!ls /kaggle/input","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"CLASSES = ['pink primrose',    'hard-leaved pocket orchid', 'canterbury bells', 'sweet pea',     'wild geranium',     'tiger lily',           'moon orchid',              'bird of paradise', 'monkshood',        'globe thistle',         # 00 - 09\n           'snapdragon',       \"colt's foot\",               'king protea',      'spear thistle', 'yellow iris',       'globe-flower',         'purple coneflower',        'peruvian lily',    'balloon flower',   'giant white arum lily', # 10 - 19\n           'fire lily',        'pincushion flower',         'fritillary',       'red ginger',    'grape hyacinth',    'corn poppy',           'prince of wales feathers', 'stemless gentian', 'artichoke',        'sweet william',         # 20 - 29\n           'carnation',        'garden phlox',              'love in the mist', 'cosmos',        'alpine sea holly',  'ruby-lipped cattleya', 'cape flower',              'great masterwort', 'siam tulip',       'lenten rose',           # 30 - 39\n           'barberton daisy',  'daffodil',                  'sword lily',       'poinsettia',    'bolero deep blue',  'wallflower',           'marigold',                 'buttercup',        'daisy',            'common dandelion',      # 40 - 49\n           'petunia',          'wild pansy',                'primula',          'sunflower',     'lilac hibiscus',    'bishop of llandaff',   'gaura',                    'geranium',         'orange dahlia',    'pink-yellow dahlia',    # 50 - 59\n           'cautleya spicata', 'japanese anemone',          'black-eyed susan', 'silverbush',    'californian poppy', 'osteospermum',         'spring crocus',            'iris',             'windflower',       'tree poppy',            # 60 - 69\n           'gazania',          'azalea',                    'water lily',       'rose',          'thorn apple',       'morning glory',        'passion flower',           'lotus',            'toad lily',        'anthurium',             # 70 - 79\n           'frangipani',       'clematis',                  'hibiscus',         'columbine',     'desert-rose',       'tree mallow',          'magnolia',                 'cyclamen ',        'watercress',       'canna lily',            # 80 - 89\n           'hippeastrum ',     'bee balm',                  'pink quill',       'foxglove',      'bougainvillea',     'camellia',             'mallow',                   'mexican petunia',  'bromelia',         'blanket flower',        # 90 - 99\n           'trumpet creeper',  'blackberry lily',           'common tulip',     'wild rose']                                                                                                                                               # 100 - 102","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"AUTO = tf.data.experimental.AUTOTUNE\nBATCH_SIZE = 16 * strategy.num_replicas_in_sync\n\nEPOCHS = 30\nIMAGE_SIZE = [512, 512]\nHEIGHT = 512\nWIDTH = 512\nCHANNELS = 3","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"## Load Data ##\n\nWhen used with TPUs, datasets are often serialized into [TFRecords](https://www.kaggle.com/ryanholbrook/tfrecords-basics). This is a format convenient for distributing data to each of the TPUs cores. "},{"metadata":{"trusted":true},"cell_type":"code","source":"TRAINING_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\n# Extending data with additional dataset \nIMAGENET = tf.io.gfile.glob(EXT_GCS_DS_PATH + '/imagenet' + '/tfrecords-jpeg-512x512' + '/*.tfrec')\nINATURALIST = tf.io.gfile.glob(EXT_GCS_DS_PATH + '/inaturalist' + '/tfrecords-jpeg-512x512' + '/*.tfrec')\nOPENIMAGE = tf.io.gfile.glob(EXT_GCS_DS_PATH + '/openimage' + '/tfrecords-jpeg-512x512' + '/*.tfrec')\nOXFORD = tf.io.gfile.glob(EXT_GCS_DS_PATH + '/oxford_102' + '/tfrecords-jpeg-512x512' + '/*.tfrec')\nTENSORFLOW = tf.io.gfile.glob(EXT_GCS_DS_PATH + '/tf_flowers' + '/tfrecords-jpeg-512x512' + '/*.tfrec')\n\nTRAINING_FILENAMES = TRAINING_FILENAMES + IMAGENET + INATURALIST + OPENIMAGE + OXFORD + TENSORFLOW\nSUBMISSION_FILENAMES = TRAINING_FILENAMES + VALIDATION_FILENAMES","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Functions "},{"metadata":{"trusted":true},"cell_type":"code","source":"def decode_image(image_data):\n    image = tf.image.decode_jpeg(image_data, channels=3)\n    image = tf.cast(image, tf.float32) / 255.0  # convert image to floats in [0, 1] range\n    image = tf.reshape(image, [*IMAGE_SIZE, 3]) # explicit size needed for TPU\n    return image\n\ndef read_labeled_tfrecord(example):\n    LABELED_TFREC_FORMAT = {\n        \"image\": tf.io.FixedLenFeature([], tf.string), # tf.string means bytestring\n        \"class\": tf.io.FixedLenFeature([], tf.int64),  # shape [] means single element\n    }\n    example = tf.io.parse_single_example(example, LABELED_TFREC_FORMAT)\n    image = decode_image(example['image'])\n    label = tf.cast(example['class'], tf.int32)\n    return image, label # returns a dataset of (image, label) pairs\n\ndef read_unlabeled_tfrecord(example):\n    UNLABELED_TFREC_FORMAT = {\n        \"image\": tf.io.FixedLenFeature([], tf.string), # tf.string means bytestring\n        \"id\": tf.io.FixedLenFeature([], tf.string),  # shape [] means single element\n        # class is missing, this competitions's challenge is to predict flower classes for the test dataset\n    }\n    example = tf.io.parse_single_example(example, UNLABELED_TFREC_FORMAT)\n    image = decode_image(example['image'])\n    idnum = example['id']\n    return image, idnum # returns a dataset of image(s)\n\ndef load_dataset(filenames, labeled=True, ordered=False):\n    # Read from TFRecords. For optimal performance, reading from multiple files at once and\n    # 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","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"## Data Augmentation"},{"metadata":{"trusted":true},"cell_type":"code","source":"def transform_rotation(image, height, rotation):\n    # input image - is one image of size [dim,dim,3] not a batch of [b,dim,dim,3]\n    # output - image randomly rotated\n    DIM = height\n    XDIM = DIM%2 #fix for size 331\n    \n    rotation = rotation * tf.random.uniform([1],dtype='float32')\n    # CONVERT DEGREES TO RADIANS\n    rotation = math.pi * rotation / 180.\n    \n    # ROTATION MATRIX\n    c1 = tf.math.cos(rotation)\n    s1 = tf.math.sin(rotation)\n    one = tf.constant([1],dtype='float32')\n    zero = tf.constant([0],dtype='float32')\n    rotation_matrix = tf.reshape(tf.concat([c1,s1,zero, -s1,c1,zero, zero,zero,one],axis=0),[3,3])\n\n    # LIST DESTINATION PIXEL INDICES\n    x = tf.repeat( tf.range(DIM//2,-DIM//2,-1), DIM )\n    y = tf.tile( tf.range(-DIM//2,DIM//2),[DIM] )\n    z = tf.ones([DIM*DIM],dtype='int32')\n    idx = tf.stack( [x,y,z] )\n    \n    # ROTATE DESTINATION PIXELS ONTO ORIGIN PIXELS\n    idx2 = K.dot(rotation_matrix,tf.cast(idx,dtype='float32'))\n    idx2 = K.cast(idx2,dtype='int32')\n    idx2 = K.clip(idx2,-DIM//2+XDIM+1,DIM//2)\n    \n    # FIND ORIGIN PIXEL VALUES \n    idx3 = tf.stack( [DIM//2-idx2[0,], DIM//2-1+idx2[1,]] )\n    d = tf.gather_nd(image, tf.transpose(idx3))\n        \n    return tf.reshape(d,[DIM,DIM,3])\n\ndef transform_shear(image, height, shear):\n    # input image - is one image of size [dim,dim,3] not a batch of [b,dim,dim,3]\n    # output - image randomly sheared\n    DIM = height\n    XDIM = DIM%2 #fix for size 331\n    \n    shear = shear * tf.random.uniform([1],dtype='float32')\n    shear = math.pi * shear / 180.\n        \n    # SHEAR MATRIX\n    one = tf.constant([1],dtype='float32')\n    zero = tf.constant([0],dtype='float32')\n    c2 = tf.math.cos(shear)\n    s2 = tf.math.sin(shear)\n    shear_matrix = tf.reshape(tf.concat([one,s2,zero, zero,c2,zero, zero,zero,one],axis=0),[3,3])    \n\n    # LIST DESTINATION PIXEL INDICES\n    x = tf.repeat( tf.range(DIM//2,-DIM//2,-1), DIM )\n    y = tf.tile( tf.range(-DIM//2,DIM//2),[DIM] )\n    z = tf.ones([DIM*DIM],dtype='int32')\n    idx = tf.stack( [x,y,z] )\n    \n    # ROTATE DESTINATION PIXELS ONTO ORIGIN PIXELS\n    idx2 = K.dot(shear_matrix,tf.cast(idx,dtype='float32'))\n    idx2 = K.cast(idx2,dtype='int32')\n    idx2 = K.clip(idx2,-DIM//2+XDIM+1,DIM//2)\n    \n    # FIND ORIGIN PIXEL VALUES \n    idx3 = tf.stack( [DIM//2-idx2[0,], DIM//2-1+idx2[1,]] )\n    d = tf.gather_nd(image, tf.transpose(idx3))\n        \n    return tf.reshape(d,[DIM,DIM,3])","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# def random_blockout(img, sl=0.1, sh=0.2, rl=0.4):\n#     p=random.random()\n#     if p>=0.25:\n#         w, h, c = IMAGE_SIZE[0], IMAGE_SIZE[1], 3\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.cast(erased_img, img.dtype)\n#     else:\n#         return tf.cast(img, img.dtype)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def data_augment(image, label):\n    p_rotation = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n    p_spatial = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n    p_rotate = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n    p_pixel_1 = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n    p_pixel_2 = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n    p_pixel_3 = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n    p_shear = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n    p_crop = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n    \n    # Shear\n    if p_shear > .2:\n        if p_shear > .6:\n            image = transform_shear(image, HEIGHT, shear=20.)\n        else:\n            image = transform_shear(image, HEIGHT, shear=-20.)\n            \n    # Rotation\n    if p_rotation > .2:\n        if p_rotation > .6:\n            image = transform_rotation(image, HEIGHT, rotation=45.)\n        else:\n            image = transform_rotation(image, HEIGHT, rotation=-45.)\n            \n    # Flips\n    image = tf.image.random_flip_left_right(image)\n    image = tf.image.random_flip_up_down(image)\n    if p_spatial > .75:\n        image = tf.image.transpose(image)\n        \n    # Rotates\n    if p_rotate > .75:\n        image = tf.image.rot90(image, k=3) # rotate 270º\n    elif p_rotate > .5:\n        image = tf.image.rot90(image, k=2) # rotate 180º\n    elif p_rotate > .25:\n        image = tf.image.rot90(image, k=1) # rotate 90º\n        \n    # Pixel-level transforms\n    if p_pixel_1 >= .4:\n        image = tf.image.random_saturation(image, lower=.7, upper=1.3)\n    if p_pixel_2 >= .4:\n        image = tf.image.random_contrast(image, lower=.8, upper=1.2)\n    if p_pixel_3 >= .4:\n        image = tf.image.random_brightness(image, max_delta=.1)\n        \n    # Crops\n    if p_crop > .7:\n        if p_crop > .9:\n            image = tf.image.central_crop(image, central_fraction=.6)\n        elif p_crop > .8:\n            image = tf.image.central_crop(image, central_fraction=.7)\n        else:\n            image = tf.image.central_crop(image, central_fraction=.8)\n    elif p_crop > .4:\n        crop_size = tf.random.uniform([], int(HEIGHT*.6), HEIGHT, dtype=tf.int32)\n        image = tf.image.random_crop(image, size=[crop_size, crop_size, CHANNELS])\n        \n    #image = random_blockout(image)       \n    image = tf.image.resize(image, size=[HEIGHT, WIDTH])\n\n    return image, label","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def data_augment_1(image, label):\n    # Thanks to the dataset.prefetch(AUTO)\n    # statement in the next function (below), this happens essentially\n    # for free on TPU. Data pipeline code is executed on the \"CPU\"\n    # part of the TPU while the TPU itself is computing gradients.\n    image = tf.image.random_flip_left_right(image)\n    image = tf.image.random_flip_up_down(image)\n    image = tf.image.random_saturation(image, 0, 2)\n    return image, label   ","execution_count":null,"outputs":[]},{"metadata":{"_kg_hide-input":true,"trusted":true},"cell_type":"code","source":"def get_training_dataset():\n    dataset = load_dataset(TRAINING_FILENAMES, labeled=True)\n    dataset = dataset.map(data_augment_1, 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)\n    return dataset\n\ndef get_submission_dataset():\n    dataset = load_dataset(SUBMISSION_FILENAMES, labeled=True)\n    dataset = dataset.map(data_augment_1, 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_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)\n    return dataset","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\n    # 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)\nNUM_SUB_IMAGES = NUM_TRAINING_IMAGES + NUM_VALIDATION_IMAGES\nprint('Dataset: {} training images, {} validation images, {} Submission images,  {} unlabeled test images'.format(NUM_TRAINING_IMAGES, NUM_VALIDATION_IMAGES,NUM_SUB_IMAGES, NUM_TEST_IMAGES))","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"This next cell will create the datasets that we'll use with Keras during training and inference. Notice how we scale the size of the batches to the number of TPU cores."},{"metadata":{"trusted":true},"cell_type":"code","source":"# Define the batch size. This will be 16 with TPU off and 128 (=16*8) with TPU on\n\nds_train = get_training_dataset()\nds_valid = get_validation_dataset()\nds_test = get_test_dataset()\nds_sub = get_submission_dataset()\nprint(\"Training:\", ds_train)\nprint (\"Validation:\", ds_valid)\nprint(\"Test:\", ds_test)\nprint(\"Submission:\", ds_sub)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"These datasets are `tf.data.Dataset` objects. You can think about a dataset in TensorFlow as a *stream* of data records. The training and validation sets are streams of `(image, label)` pairs."},{"metadata":{"trusted":true},"cell_type":"code","source":"np.set_printoptions(threshold=15, linewidth=80)\n\nprint(\"Training data shapes:\")\nfor image, label in ds_train.take(3):\n    print(image.numpy().shape, label.numpy().shape)\nprint(\"Training data label examples:\", label.numpy())","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"The test set is a stream of `(image, idnum)` pairs; `idnum` here is the unique identifier given to the image that we'll use later when we make our submission as a `csv` file."},{"metadata":{"trusted":true},"cell_type":"code","source":"print(\"Test data shapes:\")\nfor image, idnum in ds_test.take(3):\n    print(image.numpy().shape, idnum.numpy().shape)\nprint(\"Test data IDs:\", idnum.numpy().astype('U')) # U=unicode string","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Step 4: Explore Data #\n\nLet's take a moment to look at some of the images in the dataset."},{"metadata":{"_kg_hide-input":true,"trusted":true},"cell_type":"code","source":"\nfrom matplotlib import pyplot as plt\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,\n                                     # 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\n    # 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\n    # 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\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":{},"cell_type":"markdown","source":"You can display a single batch of images from a dataset with another of our helper functions. The next cell will turn the dataset into an iterator of batches of 20 images."},{"metadata":{"trusted":true},"cell_type":"code","source":"ds_iter = iter(ds_train.unbatch().batch(20))","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"Use the Python `next` function to pop out the next batch in the stream and display it with the helper function."},{"metadata":{"trusted":true},"cell_type":"code","source":"one_batch = next(ds_iter)\ndisplay_batch_of_images(one_batch)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"By defining `ds_iter` and `one_batch` in separate cells, you only need to rerun the cell above to see a new batch of images."},{"metadata":{},"cell_type":"markdown","source":"# Callbacks"},{"metadata":{"trusted":true},"cell_type":"code","source":"\n# Learning Rate Schedule for Fine Tuning #\ndef exponential_lr(epoch,\n                   start_lr = 0.00001, min_lr = 0.00001, max_lr = 0.00005,\n                   rampup_epochs = 5, sustain_epochs = 0,\n                   exp_decay = 0.8):\n\n    def lr(epoch, start_lr, min_lr, max_lr, rampup_epochs, sustain_epochs, exp_decay):\n        # linear increase from start to rampup_epochs\n        if epoch < rampup_epochs:\n            lr = ((max_lr - start_lr) /\n                  rampup_epochs * epoch + start_lr)\n        # constant max_lr during sustain_epochs\n        elif epoch < rampup_epochs + sustain_epochs:\n            lr = max_lr\n        # exponential decay towards min_lr\n        else:\n            lr = ((max_lr - min_lr) *\n                  exp_decay**(epoch - rampup_epochs - sustain_epochs) +\n                  min_lr)\n        return lr\n    return lr(epoch,\n              start_lr,\n              min_lr,\n              max_lr,\n              rampup_epochs,\n              sustain_epochs,\n              exp_decay)\n\nlr_callback = tf.keras.callbacks.LearningRateScheduler(exponential_lr, verbose=True)\n\nrng = [i for i in range(EPOCHS)]\ny = [exponential_lr(x) for x in rng]\nplt.plot(rng, y)\nprint(\"Learning rate schedule: {:.3g} to {:.3g} to {:.3g}\".format(y[0], max(y), y[-1]))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def lrfn(epoch):\n    LR_START = 0.00001\n    LR_MAX = 0.00005 * strategy.num_replicas_in_sync\n    LR_MIN = 0.00001\n    LR_RAMPUP_EPOCHS = 5\n    LR_SUSTAIN_EPOCHS = 0\n    LR_EXP_DECAY = .8\n    \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","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"from keras.callbacks import ReduceLROnPlateau,EarlyStopping, ModelCheckpoint\n\n# Stop training when the validation loss metric has stopped decreasing for 5 epochs.\nearly_stopping = EarlyStopping(monitor = 'val_loss',\n                               patience = 3,\n                               mode = 'min',\n                               restore_best_weights = True)\n\n# Save the model with the minimum validation loss\ncheckpoint = ModelCheckpoint('best_model.hdf5', \n                             monitor = 'val_loss',\n                             verbose = 1,\n                             mode = 'min', \n                             save_best_only = True)\n# # reduce learning rate\n# reduce_lr = ReduceLROnPlateau(monitor = 'val_loss',\n#                               factor = 0.2,\n#                               patience = 3,\n#                               min_lr = 0.0001,\n#                               mode = 'min',\n#                               verbose = 1)\n\nlr_schedule = tf.keras.callbacks.LearningRateScheduler(lrfn, verbose=1)\n#optimizer = tf.keras.optimizers.Nadam(learning_rate=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-07)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Step 5: Define Model #\n\nWe'll use what's known as **transfer learning**. With transfer learning, you reuse part of a pretrained model to get a head-start on a new dataset.\n\n\nThe distribution strategy we created earlier contains a [context manager](https://docs.python.org/3/reference/compound_stmts.html#with), `strategy.scope`. This context manager tells TensorFlow how to divide the work of training among the eight TPU cores. When using TensorFlow with a TPU, it's important to define your model in a `strategy.scope()` context."},{"metadata":{"trusted":true},"cell_type":"code","source":"with strategy.scope():\n    model = tf.keras.Sequential([\n        efn.EfficientNetB3( drop_connect_rate=0.35,\n                            input_shape=(HEIGHT, WIDTH, CHANNELS),\n                            weights='noisy-student',\n                            include_top=False),\n        \n        layers.GlobalAveragePooling2D(),\n        #layers.Dropout(0.2),\n        layers.Dense(len(CLASSES), activation='softmax')\n    ])\n    \n    #model.trainable = False   ","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"The `'sparse_categorical'` versions of the loss and metrics are appropriate for a classification task with more than two labels, like this one."},{"metadata":{"trusted":true},"cell_type":"code","source":"model.compile(\n    optimizer='adam',\n    loss = 'sparse_categorical_crossentropy',\n    metrics=['sparse_categorical_accuracy'],\n)\n\nmodel.summary()","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"## Fit Model ##\n\nAnd now we're ready to train the model. After defining a few parameters, we're good to go!"},{"metadata":{"trusted":true},"cell_type":"code","source":"# Define training epochs\nSTEPS_PER_EPOCH = NUM_TRAINING_IMAGES // BATCH_SIZE\nVALID_STEPS = NUM_VALIDATION_IMAGES // BATCH_SIZE\n\nhistory = model.fit(\n    ds_train,\n    steps_per_epoch=STEPS_PER_EPOCH,\n    epochs=50,\n    validation_data=ds_valid,\n    validation_steps=VALID_STEPS,\n    callbacks=[early_stopping, checkpoint, lr_callback], #  lr_schedule\n)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"This next cell shows how the loss and metrics progressed during training. Thankfully, it converges!"},{"metadata":{"trusted":true},"cell_type":"code","source":"display_training_curves(\n    history.history['loss'],\n    history.history['val_loss'],\n    'loss',\n    211,\n)\ndisplay_training_curves(\n    history.history['sparse_categorical_accuracy'],\n    history.history['val_sparse_categorical_accuracy'],\n    'accuracy',\n    212,\n)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Step 7: Evaluate Predictions #\n\nBefore making your final predictions on the test set, it's a good idea to evaluate your model's predictions on the validation set. This can help you diagnose problems in training or suggest ways your model could be improved. We'll look at two common ways of validation: plotting the **confusion matrix** and **visual validation**."},{"metadata":{"_kg_hide-input":true,"trusted":true},"cell_type":"code","source":"\nimport matplotlib.pyplot as plt\nfrom sklearn.metrics import f1_score, precision_score, recall_score, confusion_matrix\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":{},"cell_type":"markdown","source":"## Confusion Matrix ##\n\nA [confusion matrix](https://en.wikipedia.org/wiki/Confusion_matrix) shows the actual class of an image tabulated against its predicted class. It is one of the best tools you have for evaluating the performance of a classifier.\n\nThe following cell does some processing on the validation data and then creates the matrix with the `confusion_matrix` function included in [`scikit-learn`](https://scikit-learn.org/stable/index.html)."},{"metadata":{"trusted":true},"cell_type":"code","source":"cmdataset = get_validation_dataset(ordered=True)\nimages_ds = cmdataset.map(lambda image, label: image)\nlabels_ds = cmdataset.map(lambda image, label: label).unbatch()\n\ncm_correct_labels = next(iter(labels_ds.batch(NUM_VALIDATION_IMAGES))).numpy()\ncm_probabilities = model.predict(images_ds)\ncm_predictions = np.argmax(cm_probabilities, axis=-1)\n\nlabels = range(len(CLASSES))\ncmat = confusion_matrix(\n    cm_correct_labels,\n    cm_predictions,\n    labels=labels,\n)\ncmat = (cmat.T / cmat.sum(axis=1)).T # normalize","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"You might be familiar with metrics like [F1-score](https://en.wikipedia.org/wiki/F1_score) or [precision and recall](https://en.wikipedia.org/wiki/Precision_and_recall). This cell will compute these metrics and display them with a plot of the confusion matrix. (These metrics are defined in the Scikit-learn module `sklearn.metrics`; we've imported them in the helper script for you.)"},{"metadata":{"trusted":true},"cell_type":"code","source":"score = f1_score(\n    cm_correct_labels,\n    cm_predictions,\n    labels=labels,\n    average='macro',\n)\nprecision = precision_score(\n    cm_correct_labels,\n    cm_predictions,\n    labels=labels,\n    average='macro',\n)\nrecall = recall_score(\n    cm_correct_labels,\n    cm_predictions,\n    labels=labels,\n    average='macro',\n)\ndisplay_confusion_matrix(cmat, score, precision, recall)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"## Visual Validation ##\n\nIt can also be helpful to look at some examples from the validation set and see what class your model predicted. This can help reveal patterns in the kinds of images your model has trouble with.\n\nThis cell will set up the validation set to display 20 images at a time -- you can change this to display more or fewer, if you like."},{"metadata":{"trusted":true},"cell_type":"code","source":"dataset = get_validation_dataset()\ndataset = dataset.unbatch().batch(20)\nbatch = iter(dataset)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"And here is a set of flowers with their predicted species. Run the cell again to see another set."},{"metadata":{"trusted":true},"cell_type":"code","source":"images, labels = next(batch)\nprobabilities = model.predict(images)\npredictions = np.argmax(probabilities, axis=-1)\ndisplay_batch_of_images((images, labels), predictions)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Step 8: Make Test Predictions #\n\nOnce you're satisfied with everything, you're ready to make predictions on the test set."},{"metadata":{"trusted":true},"cell_type":"code","source":"# load model\nfrom keras.models import load_model\n\nbest_model = load_model('/kaggle/working/best_model.hdf5')\n# summarize model.\n#best_model.summary()","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Here, We continue to training the model with all the data that we have (Train and Validation)"},{"metadata":{"trusted":true},"cell_type":"code","source":"STEPS_PER_EPOCH_SUB = NUM_SUB_IMAGES // BATCH_SIZE\n\nhistory_sub = best_model.fit(\n    ds_sub,\n    steps_per_epoch=STEPS_PER_EPOCH_SUB,\n    epochs=10,\n    #callbacks=[early_stopping, checkpoint, lr_callback], #  lr_schedule\n)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"model.save('./model_submission.h5')","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"!ls /kaggle/working","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"test_ds = get_test_dataset(ordered=True)\n\nprint('Computing predictions...')\ntest_images_ds = test_ds.map(lambda image, idnum: image)\nprobabilities = best_model.predict(test_images_ds)\npredictions = np.argmax(probabilities, axis=-1)\nprint(predictions)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"We'll generate a file `submission.csv`. This file is what you'll submit to get your score on the leaderboard."},{"metadata":{"trusted":true},"cell_type":"code","source":"print('Generating submission.csv file...')\n\n# Get image ids from test set and convert to unicode\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')\n\n# Write the submission file\nnp.savetxt(\n    'submission.csv',\n    np.rec.fromarrays([test_ids, predictions]),\n    fmt=['%s', '%d'],\n    delimiter=',',\n    header='id,label',\n    comments='',\n)\n\n# Look at the first few predictions\n!head submission.csv","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Step 9: Make a submission #\n\n\n1. Begin by clicking on the blue **Save Version** button in the top right corner of the window.  This will generate a pop-up window.  \n2. Ensure that the **Save and Run All** option is selected, and then click on the blue **Save** button.\n3. This generates a window in the bottom left corner of the notebook.  After it has finished running, click on the number to the right of the **Save Version** button.  This pulls up a list of versions on the right of the screen.  Click on the ellipsis **(...)** to the right of the most recent version, and select **Open in Viewer**.  This brings you into view mode of the same page. You will need to scroll down to get back to these instructions.\n4. Click on the **Output** tab on the right of the screen.  Then, click on the file you would like to submit, and click on the blue **Submit** button to submit your results to the leaderboard.\n"}],"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}