{"cells":[{"metadata":{},"cell_type":"markdown","source":"# Update:\n## `v12`:\n* ckpt was saving models scoring minimum\n* Stratified tfrecords have been added"},{"metadata":{},"cell_type":"markdown","source":"# Credits:\n1. Notebook was mostly taken from [here](https://www.kaggle.com/itsuki9180/efficientnet-and-cutmixup-with-tpu-train-phase)\n2. Thanks to [@spidermandance](https://www.kaggle.com/spidermandance) for Custom TFrecord Dataset as official one had some issues\n3. Thanks to [@cdeotte](https://www.kaggle.com/cdeotte) for his awesome augmentations"},{"metadata":{},"cell_type":"markdown","source":"# Cassava Leaf Disease Train Phase\n \nReferences, see also them:\n\n[Getting Started: TPUs + Cassava Leaf Disease](https://www.kaggle.com/jessemostipak/getting-started-tpus-cassava-leaf-disease)\n\n[CutMix and MixUp on GPU/TPU](https://www.kaggle.com/cdeotte/cutmix-and-mixup-on-gpu-tpu)\n\n[TFRecord Experiments - Upsample and Coarse Dropout](https://www.kaggle.com/cdeotte/tfrecord-experiments-upsample-and-coarse-dropout)\n\n[Getting started with 100+ flowers on TPU](https://www.kaggle.com/mgornergoogle/getting-started-with-100-flowers-on-tpu)\n\n[Triple Stratified KFold with TFRecords](https://www.kaggle.com/cdeotte/triple-stratified-kfold-with-tfrecords)"},{"metadata":{},"cell_type":"markdown","source":"# Set up environment"},{"metadata":{},"cell_type":"markdown","source":"## For Offline EfficientNet (No Internet Connection is required)\nShows some error messages as input dataset doesn't have `write access` but doesn't make any difference. You can the move dataset it to working directory and avoid the error.\n\n**Weights path have been fixed for kaggle**"},{"metadata":{"trusted":true},"cell_type":"code","source":"%%capture\nimport sys\nsys.path.append('/kaggle/input/efficientnet-keras-dataset/efficientnet_kaggle')\n! pip install -e /kaggle/input/efficientnet-keras-dataset/efficientnet_kaggle","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Importing "},{"metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true},"cell_type":"code","source":"import math, re, os, gc\nimport tensorflow as tf\nimport tensorflow.keras.backend as K\nimport numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nfrom kaggle_datasets import KaggleDatasets\nfrom tensorflow import keras\nfrom functools import partial\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.model_selection import KFold\nprint(\"Tensorflow version \" + tf.__version__)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Detect TPU"},{"metadata":{"_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","trusted":true},"cell_type":"code","source":"try:\n tpu = tf.distribute.cluster_resolver.TPUClusterResolver()\n print('Device:', tpu.master())\n tf.config.experimental_connect_to_cluster(tpu)\n tf.tpu.experimental.initialize_tpu_system(tpu)\n strategy = tf.distribute.experimental.TPUStrategy(tpu)\nexcept:\n strategy = tf.distribute.get_strategy()\nprint('Number of replicas:', strategy.num_replicas_in_sync)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Set up variables"},{"metadata":{"trusted":true},"cell_type":"code","source":"DIM = 512 \nIMAGE_SIZE = [DIM, DIM]\nEFFNET = 6 # 0-7\nPRETRAINED_WEIGHTS = 'imagenet' # noisy-student\n\nGCS_PATH1 = KaggleDatasets().get_gcs_path(f'cassava-tfrecords-{DIM}x{DIM}')\nGCS_PATH2 = KaggleDatasets().get_gcs_path('cassava-leaf-disease-classification')\nAUTOTUNE = tf.data.experimental.AUTOTUNE\nBATCH_SIZE = 8 * strategy.num_replicas_in_sync\nAUG_BATCH = BATCH_SIZE\n\nCLASSES = ['0', '1', '2', '3', '4']\nEPOCHS = 12\nFOLDS = 5\nPHASE = 'train'#'inference'","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Define Augmentation"},{"metadata":{"trusted":true},"cell_type":"code","source":"ROT_ = 180.0\nSHR_ = 2.0\nHZOOM_ = 8.0\nWZOOM_ = 8.0\nHSHIFT_ = 8.0\nWSHIFT_ = 8.0","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def get_mat(rotation, shear, height_zoom, width_zoom, height_shift, width_shift):\n # returns 3x3 transformmatrix which transforms indicies\n \n # CONVERT DEGREES TO RADIANS\n rotation = math.pi * rotation / 180.\n shear = math.pi * shear / 180.\n\n def get_3x3_mat(lst):\n return tf.reshape(tf.concat([lst],axis=0), [3,3])\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 \n rotation_matrix = get_3x3_mat([c1, s1, zero, \n -s1, c1, zero, \n zero, zero, one]) \n # SHEAR MATRIX\n c2 = tf.math.cos(shear)\n s2 = tf.math.sin(shear) \n \n shear_matrix = get_3x3_mat([one, s2, zero, \n zero, c2, zero, \n zero, zero, one]) \n # ZOOM MATRIX\n zoom_matrix = get_3x3_mat([one/height_zoom, zero, zero, \n zero, one/width_zoom, zero, \n zero, zero, one]) \n # SHIFT MATRIX\n shift_matrix = get_3x3_mat([one, zero, height_shift, \n zero, one, width_shift, \n zero, zero, one])\n \n return K.dot(K.dot(rotation_matrix, shear_matrix), \n K.dot(zoom_matrix, shift_matrix))\n\n\ndef transform_mat(image, DIM=IMAGE_SIZE[0]): \n # input image - is one image of size [dim,dim,3] not a batch of [b,dim,dim,3]\n # output - image randomly rotated, sheared, zoomed, and shifted\n XDIM = DIM%2 \n \n rot = ROT_ * tf.random.normal([1], dtype='float32')\n shr = SHR_ * tf.random.normal([1], dtype='float32') \n h_zoom = 1.0 + tf.random.normal([1], dtype='float32') / HZOOM_\n w_zoom = 1.0 + tf.random.normal([1], dtype='float32') / WZOOM_\n h_shift = HSHIFT_ * tf.random.normal([1], dtype='float32') \n w_shift = WSHIFT_ * tf.random.normal([1], dtype='float32') \n\n # GET TRANSFORMATION MATRIX\n m = get_mat(rot,shr,h_zoom,w_zoom,h_shift,w_shift) \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(m, 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":{},"cell_type":"markdown","source":"## Decode the data"},{"metadata":{"trusted":true},"cell_type":"code","source":"# this code will convert our test image data to a float32 \ndef to_float32(image, label):\n return tf.cast(image, tf.float32), label","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def decode_image(image):\n image = tf.image.decode_jpeg(image, channels=3)\n image = tf.cast(image, tf.float32) / 255.0\n image = tf.reshape(image, [*IMAGE_SIZE, 3])\n return image","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def read_tfrecord(example, labeled):\n tfrecord_format = {\n \"image\": tf.io.FixedLenFeature([], tf.string),\n \"target\": tf.io.FixedLenFeature([], tf.int64)\n } if labeled else {\n \"image\": tf.io.FixedLenFeature([], tf.string),\n \"image_name\": tf.io.FixedLenFeature([], tf.string)\n }\n example = tf.io.parse_single_example(example, tfrecord_format)\n image = decode_image(example['image'])\n if labeled:\n label = tf.cast(example['target'], tf.int32)\n return image, label\n idnum = example['image_name']\n return image, idnum","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def load_dataset(filenames, labeled=True, ordered=False):\n ignore_order = tf.data.Options()\n if not ordered:\n ignore_order.experimental_deterministic = False # disable order, increase speed\n dataset = tf.data.TFRecordDataset(filenames, num_parallel_reads=AUTOTUNE) # 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(partial(read_tfrecord, labeled=labeled), num_parallel_calls=AUTOTUNE)\n return dataset","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"TRAINING_FILENAMES = tf.io.gfile.glob(GCS_PATH1 + '/ld_train*.tfrec')\nTEST_FILENAMES = tf.io.gfile.glob(GCS_PATH2 + '/test_tfrecords/ld_test*.tfrec')\nprint('Train Files:',len(TRAINING_FILENAMES))\nprint('Test Files:',len(TEST_FILENAMES))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def count_data_items(filenames):\n # the number of data items is written in the name of the .tfrec files, i.e. flowers00-230.tfrec = 230 data items\n n = [int(re.compile(r\"-([0-9]*)\\.\").search(filename).group(1)) for filename in filenames]\n return np.sum(n)\n\nNUM_TRAINING_IMAGES = int( count_data_items(TRAINING_FILENAMES) * (FOLDS-1.)/FOLDS )\nNUM_VALIDATION_IMAGES = int( count_data_items(TRAINING_FILENAMES) * (1./FOLDS) )\nNUM_TEST_IMAGES = count_data_items(TEST_FILENAMES)\nSTEPS_PER_EPOCH = NUM_TRAINING_IMAGES // BATCH_SIZE\n\nprint('Dataset: {} training images, {} validation images, {} unlabeled test images'.format(NUM_TRAINING_IMAGES, NUM_VALIDATION_IMAGES, NUM_TEST_IMAGES))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"TRAINING_FILENAMES, VALID_FILENAMES = train_test_split(\n tf.io.gfile.glob(GCS_PATH1 + '/ld_train*.tfrec'),\n test_size=0.35, random_state=5\n)\n\nTEST_FILENAMES = tf.io.gfile.glob(GCS_PATH2 + '/test_tfrecords/ld_test*.tfrec')","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def dropout(image, DIM=DIM, PROBABILITY = 0.75, CT = 8, SZ = 0.2):\n # input image - is one image of size [dim,dim,3] not a batch of [b,dim,dim,3]\n # output - image with CT squares of side size SZ*DIM removed\n \n # DO DROPOUT WITH PROBABILITY DEFINED ABOVE\n P = tf.cast( tf.random.uniform([],0,1) 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 plt.xticks([])\n plt.yticks([])\n else:\n plt.figure(figsize=(FIGSIZE/rows*cols,FIGSIZE))\n plt.xticks([])\n plt.yticks([]) \n # display\n for i, (image, label) in enumerate(zip(images[:rows*cols], labels[:rows*cols])):\n title = '' if label is None else label2name[CLASSES[label]] + f'({str(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_plant(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.tight_layout()\n plt.show()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# load our training dataset for EDA\ntraining_dataset = get_training_dataset()\ntraining_dataset = training_dataset.unbatch().batch(20)\ntrain_batch = iter(training_dataset)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# run this cell again for another randomized set of training images\ndisplay_batch_of_images(next(train_batch))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# load our validation dataset for EDA\nvalidation_dataset = get_validation_dataset()\nvalidation_dataset = validation_dataset.unbatch().batch(20)\nvalid_batch = iter(validation_dataset)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# run this cell again for another randomized set of training images\ndisplay_batch_of_images(next(valid_batch))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# load our test dataset for EDA\ntesting_dataset = get_test_dataset()\ntesting_dataset = testing_dataset.unbatch().batch(20)\ntest_batch = iter(testing_dataset)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# we only have one test image\ndisplay_batch_of_images(next(test_batch))","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# CutMix and MixUp Augmentation"},{"metadata":{"trusted":true},"cell_type":"code","source":"def onehot(image,label):\n CLASSES = 5\n return image,tf.one_hot(label,CLASSES)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def cutmix(image, label, PROBABILITY = 1.0):\n # input image - is a batch of images of size [n,dim,dim,3] not a single image of [dim,dim,3]\n # output - a batch of images with cutmix applied\n DIM = IMAGE_SIZE[0]\n CLASSES = 5\n \n imgs = []; labs = []\n for j in range(AUG_BATCH):\n # DO CUTMIX WITH PROBABILITY DEFINED ABOVE\n P = tf.cast( tf.random.uniform([],0,1)<=PROBABILITY, tf.int32)\n # CHOOSE RANDOM IMAGE TO CUTMIX WITH\n k = tf.cast( tf.random.uniform([],0,AUG_BATCH),tf.int32)\n # CHOOSE RANDOM LOCATION\n x = tf.cast( tf.random.uniform([],0,DIM),tf.int32)\n y = tf.cast( tf.random.uniform([],0,DIM),tf.int32)\n b = tf.random.uniform([],0,1) # this is beta dist with alpha=1.0\n WIDTH = tf.cast( DIM * tf.math.sqrt(1-b),tf.int32) * P\n ya = tf.math.maximum(0,y-WIDTH//2)\n yb = tf.math.minimum(DIM,y+WIDTH//2)\n xa = tf.math.maximum(0,x-WIDTH//2)\n xb = tf.math.minimum(DIM,x+WIDTH//2)\n # MAKE CUTMIX IMAGE\n one = image[j,ya:yb,0:xa,:]\n two = image[k,ya:yb,xa:xb,:]\n three = image[j,ya:yb,xb:DIM,:]\n middle = tf.concat([one,two,three],axis=1)\n img = tf.concat([image[j,0:ya,:,:],middle,image[j,yb:DIM,:,:]],axis=0)\n imgs.append(img)\n # MAKE CUTMIX LABEL\n a = tf.cast(WIDTH*WIDTH/DIM/DIM,tf.float32)\n if len(label.shape)==1:\n lab1 = tf.one_hot(label[j],CLASSES)\n lab2 = tf.one_hot(label[k],CLASSES)\n else:\n lab1 = label[j,]\n lab2 = label[k,]\n labs.append((1-a)*lab1 + a*lab2)\n \n # RESHAPE HACK SO TPU COMPILER KNOWS SHAPE OF OUTPUT TENSOR (maybe use Python typing instead?)\n image2 = tf.reshape(tf.stack(imgs),(AUG_BATCH,DIM,DIM,3))\n label2 = tf.reshape(tf.stack(labs),(AUG_BATCH,CLASSES))\n return image2,label2\n\n\ndef mixup(image, label, PROBABILITY = 1.0):\n # input image - is a batch of images of size [n,dim,dim,3] not a single image of [dim,dim,3]\n # output - a batch of images with mixup applied\n DIM = IMAGE_SIZE[0]\n CLASSES = 5\n \n imgs = []; labs = []\n for j in range(AUG_BATCH):\n # DO MIXUP WITH PROBABILITY DEFINED ABOVE\n P = tf.cast( tf.random.uniform([],0,1)<=PROBABILITY, tf.float32)\n # CHOOSE RANDOM\n k = tf.cast( tf.random.uniform([],0,AUG_BATCH),tf.int32)\n a = tf.random.uniform([],0,1)*P # this is beta dist with alpha=1.0\n # MAKE MIXUP IMAGE\n img1 = image[j,]\n img2 = image[k,]\n imgs.append((1-a)*img1 + a*img2)\n # MAKE CUTMIX LABEL\n if len(label.shape)==1:\n lab1 = tf.one_hot(label[j],CLASSES)\n lab2 = tf.one_hot(label[k],CLASSES)\n else:\n lab1 = label[j,]\n lab2 = label[k,]\n labs.append((1-a)*lab1 + a*lab2)\n \n # RESHAPE HACK SO TPU COMPILER KNOWS SHAPE OF OUTPUT TENSOR (maybe use Python typing instead?)\n image2 = tf.reshape(tf.stack(imgs),(AUG_BATCH,DIM,DIM,3))\n label2 = tf.reshape(tf.stack(labs),(AUG_BATCH,CLASSES))\n return image2,label2\n\ndef transform(image,label):\n # THIS FUNCTION APPLIES BOTH CUTMIX AND MIXUP\n DIM = IMAGE_SIZE[0]\n CLASSES = 5\n SWITCH = 0.5\n CUTMIX_PROB = 0.666\n MIXUP_PROB = 0.666\n # FOR SWITCH PERCENT OF TIME WE DO CUTMIX AND (1-SWITCH) WE DO MIXUP\n image1 = []\n for j in range(AUG_BATCH):\n img = transform_mat(image[j,])\n img = tf.image.random_flip_left_right(img)\n img = tf.image.random_saturation(img, 0.7, 1.3)\n img = tf.image.random_contrast(img, 0.8, 1.2)\n img = tf.image.random_brightness(img, 0.1)\n image1.append(img)\n \n image1 = tf.reshape(tf.stack(image1),(AUG_BATCH,DIM,DIM,3))\n image2, label2 = cutmix(image1, label, CUTMIX_PROB)\n image3, label3 = mixup(image1, label, MIXUP_PROB)\n imgs = []; labs = []\n for j in range(AUG_BATCH):\n P = tf.cast( tf.random.uniform([],0,1)<=SWITCH, tf.float32)\n imgs.append(P*image2[j,]+(1-P)*image3[j,])\n labs.append(P*label2[j,]+(1-P)*label3[j,])\n # RESHAPE HACK SO TPU COMPILER KNOWS SHAPE OF OUTPUT TENSOR (maybe use Python typing instead?)\n image4 = tf.reshape(tf.stack(imgs),(AUG_BATCH,DIM,DIM,3))\n label4 = tf.reshape(tf.stack(labs),(AUG_BATCH,CLASSES))\n return image4,label4","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def get_training_dataset(dataset=TRAINING_FILENAMES, do_aug=True):\n #dataset = load_dataset(dataset, labeled=True) \n if do_aug: \n dataset = dataset.map(data_augment, num_parallel_calls=AUTOTUNE)\n dataset = dataset.repeat() # the training dataset must repeat for several epochs\n dataset = dataset.batch(AUG_BATCH)\n if do_aug: \n dataset = dataset.map(transform, num_parallel_calls=AUTOTUNE) # note we put AFTER batching \n dataset = dataset.unbatch()\n dataset = dataset.shuffle(2048)\n dataset = dataset.batch(BATCH_SIZE)\n dataset = dataset.prefetch(AUTOTUNE) # prefetch next batch while training (autotune prefetch buffer size)\n return dataset","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def get_validation_dataset(dataset, do_onehot=True):\n dataset = dataset.batch(BATCH_SIZE)\n if do_onehot: dataset = dataset.map(onehot, num_parallel_calls=AUTOTUNE) # we must use one hot like augmented train data\n dataset = dataset.cache()\n dataset = dataset.prefetch(AUTOTUNE) # prefetch next batch while training (autotune prefetch buffer size)\n return dataset","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"row = 6; col = 4;\nrow = min(row,AUG_BATCH//col)\nall_elements = get_training_dataset(load_dataset(TRAINING_FILENAMES),do_aug=False).unbatch()\naugmented_element = all_elements.repeat().batch(AUG_BATCH).map(transform)\n\nfor (img,label) in augmented_element:\n plt.figure(figsize=(15,int(15*row/col)))\n for j in range(row*col):\n plt.subplot(row,col,j+1)\n plt.axis('off')\n plt.imshow(img[j,])\n plt.tight_layout()\n plt.show()\n break","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Building the model\n## Learning rate schedule"},{"metadata":{"trusted":true},"cell_type":"code","source":"def get_lr_callback(batch_size=8, show = False):\n lr_start = 0.000015000\n lr_max = 0.000000250 * strategy.num_replicas_in_sync * batch_size\n lr_min = 0.000001\n lr_ramp_ep = 5\n lr_sus_ep = 0\n lr_decay = 0.8\n \n def lrfn(epoch):\n if epoch < lr_ramp_ep:\n lr = (lr_max - lr_start) / lr_ramp_ep * epoch + lr_start\n \n elif epoch < lr_ramp_ep + lr_sus_ep:\n lr = lr_max\n \n else:\n lr = (lr_max - lr_min) * lr_decay**(epoch - lr_ramp_ep - lr_sus_ep) + lr_min\n \n return lr\n if show:\n plt.figure(figsize = (8, 5))\n plt.plot(np.arange(1, 12), [lrfn(x) for x in np.arange(1, 12)], marker = 'o')\n plt.xlabel('epoch')\n plt.ylabel('learaning_rate');\n plt.show()\n\n lr_callback = tf.keras.callbacks.LearningRateScheduler(lrfn, verbose=False)\n return lr_callback\n\nget_lr_callback(batch_size=BATCH_SIZE, show = True)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"## Building our model"},{"metadata":{"trusted":true},"cell_type":"code","source":"import efficientnet.tfkeras as efn \nmodel_dict = {0: efn.EfficientNetB0,\n 1: efn.EfficientNetB1,\n 2: efn.EfficientNetB2,\n 3: efn.EfficientNetB3,\n 4: efn.EfficientNetB4,\n 5: efn.EfficientNetB5,\n 6: efn.EfficientNetB6,\n 7: efn.EfficientNetB7,}\ndef get_model():\n with strategy.scope(): \n inp = tf.keras.layers.Input(shape=(DIM,DIM,3))\n base = model_dict[EFFNET](input_shape=(DIM,DIM,3),weights=PRETRAINED_WEIGHTS,include_top=False)\n x = base(inp)\n x = tf.keras.layers.GlobalAveragePooling2D()(x)\n x = tf.keras.layers.Dense(64,activation='relu')(x)\n x = tf.keras.layers.Dense(5,activation='softmax')(x)\n model = tf.keras.Model(inputs=inp, outputs=x)\n opt = tf.keras.optimizers.Adam(learning_rate=0.001)\n loss = tf.keras.losses.CategoricalCrossentropy(\n from_logits=False, label_smoothing=0.0001,\n name='categorical_crossentropy'\n )\n model.compile(optimizer=opt,loss=loss,metrics=['categorical_accuracy'])\n return model","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"TRAINING_FILENAMES = tf.io.gfile.glob(GCS_PATH1 + '/ld_train*.tfrec')\nkfold = KFold(FOLDS, shuffle = True, random_state = 42)\n\nprobabilities = np.zeros((count_data_items(TEST_FILENAMES),1))\nfor f, (trn_ind, val_ind) in enumerate(kfold.split(TRAINING_FILENAMES)):\n print(); print('='*50)\n print(f' fold: {f+1} | model: EfficientNetB{EFFNET} | image_size: {DIM}')\n print('='*50)\n if PHASE == 'train':\n train_dataset = load_dataset(list(pd.DataFrame({'TRAINING_FILENAMES': TRAINING_FILENAMES}).loc[trn_ind]['TRAINING_FILENAMES']), labeled = True)\n val_dataset = load_dataset(list(pd.DataFrame({'TRAINING_FILENAMES': TRAINING_FILENAMES}).loc[val_ind]['TRAINING_FILENAMES']), labeled = True, ordered = True)\n sv = tf.keras.callbacks.ModelCheckpoint(\n 'effnet-b%i_img-%i_fold-%02d_best.h5'%(EFFNET, DIM, f+1), monitor='val_categorical_accuracy', verbose=0, save_best_only=True,\n save_weights_only=True, mode='max', save_freq='epoch')\n model = get_model()\n history = model.fit(\n get_training_dataset(train_dataset), \n steps_per_epoch = STEPS_PER_EPOCH,\n epochs = EPOCHS,\n callbacks = [get_lr_callback(BATCH_SIZE), sv],\n validation_data = get_validation_dataset(val_dataset),\n# verbose=1\n )\n plt.figure(figsize=(15,5))\n plt.plot(np.arange(len(model.history.history['categorical_accuracy'])),model.history.history['categorical_accuracy'],'-o',label='Train categorical_accuracy',color='#ff7f0e')\n plt.plot(np.arange(len(model.history.history['categorical_accuracy'])),model.history.history['val_categorical_accuracy'],'-o',label='Val categorical_accuracy',color='#1f77b4')\n x = np.argmax( model.history.history['val_categorical_accuracy'] ); y = np.max( model.history.history['val_categorical_accuracy'] )\n xdist = plt.xlim()[1] - plt.xlim()[0]; ydist = plt.ylim()[1] - plt.ylim()[0]\n plt.scatter(x,y,s=200,color='#1f77b4'); plt.text(x-0.03*xdist,y-0.13*ydist,'max categorical_accuracy\\n%.2f'%y,size=14)\n plt.ylabel('categorical_accuracy',size=14); plt.xlabel('Epoch',size=14)\n plt.legend(loc=2)\n plt2 = plt.gca().twinx()\n plt2.plot(np.arange(len(model.history.history['loss'])),model.history.history['loss'],'-o',label='Train Loss',color='#2ca02c')\n plt2.plot(np.arange(len(model.history.history['loss'])),model.history.history['val_loss'],'-o',label='Val Loss',color='#d62728')\n x = np.argmin( model.history.history['val_loss'] ); y = np.min( model.history.history['val_loss'] )\n ydist = plt.ylim()[1] - plt.ylim()[0]\n plt.scatter(x,y,s=200,color='#d62728'); plt.text(x-0.03*xdist,y+0.05*ydist,'min loss',size=14)\n plt.ylabel('Loss',size=14)\n plt.title('Fold: %i | Image Size: %i | model: EfficientNetB%i | Batch_size: %i'%(f+1, DIM ,EFFNET, BATCH_SIZE))\n plt.legend(loc=3)\n plt.show()\n\n elif PHASE == 'inference':\n model = get_model()\n model.load_weights('../input/leafdiseaseefnet0/fold-%i.h5'%f)\n test_ds = get_test_dataset(ordered=True) \n test_ds = test_ds.map(to_float32)\n\n print('Computing predictions...')\n test_images_ds = testing_dataset\n test_images_ds = test_ds.map(lambda image, idnum: image)\n print(model.predict(test_images_ds))\n probabilities += model.predict(test_images_ds) / FOLDS\n\n \n \n else:\n print('PHASE must be train or inference.')\n \n model.save_weights('effnet-b%i_img-%i_fold-%02d_last.h5'%(EFFNET, DIM, f+1))\n del model; z = gc.collect()\n \npredictions = np.argmax(probabilities, axis=-1)\nprint(predictions)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"if PHASE == 'inference':\n print('Generating submission.csv file...')\n test_ids_ds = test_ds.map(lambda image, idnum: idnum).unbatch()\n test_ids = next(iter(test_ids_ds.batch(NUM_TEST_IMAGES))).numpy().astype('U') # all in one batch\n np.savetxt('submission.csv', np.rec.fromarrays([test_ids, predictions]), fmt=['%s', '%d'], delimiter=',', header='id,label', comments='')\n !head submission.csv","execution_count":null,"outputs":[]}],"metadata":{"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}