I am trying to create a model for this Kaggle competition. A lot of the code that I am using here is from the getting started notebook in that competition. For extra context, I am using Tensorflow 2.15.0 and I am using the Kaggle notebook to write and run this code.
To summarize, the dataset I am using involves a series of images of 104 different species of flowers loaded in from TFRecord files split into training, validation, and test sets.
When pre-processing the training data, I duplicate the dataset a few times and apply data augmentation to all the images in that dataset.
For my model, I am using the Xception model pre-trained on imagenet modified to have a classifier with 104 categories at the end of it.
The model trains fine although with some weird errors:
model_pruner failed: INVALID_ARGUMENT: Graph does not contain terminal node StatefulPartitionedCall.
model_pruner failed: INVALID_ARGUMENT: Graph does not contain terminal node AssignVariableOp.
import math, re, os
import numpy as np
import tensorflow as tf
import matplotlib as plt
print("Tensorflow version " + tf.__version__)
# Detect TPU, return appropriate distribution strategy
try:
tpu = tf.distribute.cluster_resolver.TPUClusterResolver()
print('Running on TPU ', tpu.master())
except ValueError:
tpu = None
if tpu:
tf.config.experimental_connect_to_cluster(tpu)
tf.tpu.experimental.initialize_tpu_system(tpu)
strategy = tf.distribute.TPUStrategy(tpu)
else:
strategy = tf.distribute.get_strategy()
print("REPLICAS: ", strategy.num_replicas_in_sync)
IMAGE_SIZE = [512, 512]
GCS_PATH = "/kaggle/input/tpu-getting-started" + '/tfrecords-jpeg-512x512'
AUTO = tf.data.experimental.AUTOTUNE
TRAINING_FILENAMES = tf.io.gfile.glob(GCS_PATH + '/train/*.tfrec')
VALIDATION_FILENAMES = tf.io.gfile.glob(GCS_PATH + '/val/*.tfrec')
TEST_FILENAMES = tf.io.gfile.glob(GCS_PATH + '/test/*.tfrec')
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
'snapdragon', "colt's foot", 'king protea', 'spear thistle', 'yellow iris', 'globe-flower', 'purple coneflower', 'peruvian lily', 'balloon flower', 'giant white arum lily', # 10 - 19
'fire lily', 'pincushion flower', 'fritillary', 'red ginger', 'grape hyacinth', 'corn poppy', 'prince of wales feathers', 'stemless gentian', 'artichoke', 'sweet william', # 20 - 29
'carnation', 'garden phlox', 'love in the mist', 'cosmos', 'alpine sea holly', 'ruby-lipped cattleya', 'cape flower', 'great masterwort', 'siam tulip', 'lenten rose', # 30 - 39
'barberton daisy', 'daffodil', 'sword lily', 'poinsettia', 'bolero deep blue', 'wallflower', 'marigold', 'buttercup', 'daisy', 'common dandelion', # 40 - 49
'petunia', 'wild pansy', 'primula', 'sunflower', 'lilac hibiscus', 'bishop of llandaff', 'gaura', 'geranium', 'orange dahlia', 'pink-yellow dahlia', # 50 - 59
'cautleya spicata', 'japanese anemone', 'black-eyed susan', 'silverbush', 'californian poppy', 'osteospermum', 'spring crocus', 'iris', 'windflower', 'tree poppy', # 60 - 69
'gazania', 'azalea', 'water lily', 'rose', 'thorn apple', 'morning glory', 'passion flower', 'lotus', 'toad lily', 'anthurium', # 70 - 79
'frangipani', 'clematis', 'hibiscus', 'columbine', 'desert-rose', 'tree mallow', 'magnolia', 'cyclamen ', 'watercress', 'canna lily', # 80 - 89
'hippeastrum ', 'bee balm', 'pink quill', 'foxglove', 'bougainvillea', 'camellia', 'mallow', 'mexican petunia', 'bromelia', 'blanket flower', # 90 - 99
'trumpet creeper', 'blackberry lily', 'common tulip', 'wild rose'] # 100 - 102
def decode_image(image_data):
image = tf.image.decode_jpeg(image_data, channels=3)
image = tf.cast(image, tf.float32) / 255.0 # convert image to floats in [0, 1] range
image = tf.reshape(image, [*IMAGE_SIZE, 3]) # explicit size needed for TPU
return image
def read_labeled_tfrecord(example):
LABELED_TFREC_FORMAT = {
"image": tf.io.FixedLenFeature([], tf.string), # tf.string means bytestring
"class": tf.io.FixedLenFeature([], tf.int64), # shape [] means single element
}
example = tf.io.parse_single_example(example, LABELED_TFREC_FORMAT)
image = decode_image(example['image'])
label = tf.cast(example['class'], tf.int32)
return image, label # returns a dataset of (image, label) pairs
def read_unlabeled_tfrecord(example):
UNLABELED_TFREC_FORMAT = {
"image": tf.io.FixedLenFeature([], tf.string), # tf.string means bytestring
"id": tf.io.FixedLenFeature([], tf.string), # shape [] means single element
# class is missing, this competitions's challenge is to predict flower classes for the test dataset
}
example = tf.io.parse_single_example(example, UNLABELED_TFREC_FORMAT)
image = decode_image(example['image'])
idnum = example['id']
return image, idnum # returns a dataset of image(s)
def load_dataset(filenames, labeled=True, ordered=False):
# Read from TFRecords. For optimal performance, reading from multiple files at once and
# disregarding data order. Order does not matter since we will be shuffling the data anyway.
ignore_order = tf.data.Options()
if not ordered:
ignore_order.experimental_deterministic = False # disable order, increase speed
dataset = tf.data.TFRecordDataset(filenames, num_parallel_reads=AUTO) # automatically interleaves reads from multiple files
dataset = dataset.with_options(ignore_order) # uses data as soon as it streams in, rather than in its original order
dataset = dataset.map(read_labeled_tfrecord if labeled else read_unlabeled_tfrecord, num_parallel_calls=AUTO)
# returns a dataset of (image, label) pairs if labeled=True or (image, id) pairs if labeled=False
return dataset
# data augmentation model
data_augmentation = tf.keras.Sequential([
tf.keras.layers.Resizing(IMAGE_SIZE[0], IMAGE_SIZE[1]),
tf.keras.layers.RandomFlip("horizontal_and_vertical"), # Randomly flip the image horizontally and/or vertically
tf.keras.layers.RandomRotation(0.2), # Randomly rotate the image by 20%
tf.keras.layers.RandomTranslation(height_factor=0.2, width_factor=0.2), # Randomly translate the image by up to 20%
tf.keras.layers.RandomZoom(0.2), # Randomly zoom the image in or out by up to 20%
tf.keras.layers.RandomContrast(0.2), # Randomly change the contrast of the image by up to 20%
])
# image resizing
resize_image = tf.keras.Sequential([
tf.keras.layers.Resizing(IMAGE_SIZE[0], IMAGE_SIZE[1])
])
def get_training_dataset(num_augmentations):
loaded_dataset = load_dataset(TRAINING_FILENAMES, labeled=True)
dataset = loaded_dataset
# add additional images to the dataset
for i in range(num_augmentations):
dataset = dataset.concatenate(loaded_dataset)
# perform data augmentation on the dataset
dataset = dataset.map(lambda image, label: (data_augmentation(image), label), num_parallel_calls=AUTO)
dataset = dataset.repeat() # the training dataset must repeat for several epochs
dataset = dataset.shuffle(2048)
dataset = dataset.batch(BATCH_SIZE)
dataset = dataset.prefetch(AUTO) # prefetch next batch while training (autotune prefetch buffer size)
return dataset
def get_validation_dataset(ordered=False):
dataset = load_dataset(VALIDATION_FILENAMES, labeled=True, ordered=ordered)
dataset = dataset.map(lambda image, label: (resize_image(image), label), num_parallel_calls=AUTO)
dataset = dataset.batch(BATCH_SIZE)
dataset = dataset.cache()
dataset = dataset.prefetch(AUTO)
return dataset
def get_test_dataset(ordered=False):
dataset = load_dataset(TEST_FILENAMES, labeled=False, ordered=ordered)
dataset = dataset.map(lambda image, ids: (resize_image(image), ids), num_parallel_calls=AUTO)
dataset = dataset.batch(BATCH_SIZE)
dataset = dataset.prefetch(AUTO)
return dataset
def count_data_items(filenames):
# the number of data items is written in the name of the .tfrec
# files, i.e. flowers00-230.tfrec = 230 data items
n = [int(re.compile(r"-([0-9]*)\.").search(filename).group(1)) for filename in filenames]
return np.sum(n)
NUM_TRAINING_IMAGES = count_data_items(TRAINING_FILENAMES)
NUM_VALIDATION_IMAGES = count_data_items(VALIDATION_FILENAMES)
NUM_TEST_IMAGES = count_data_items(TEST_FILENAMES)
print('Dataset: {} training images, {} validation images, {} unlabeled test images'.format(NUM_TRAINING_IMAGES, NUM_VALIDATION_IMAGES, NUM_TEST_IMAGES))
# Define the batch size. This will be 16 with TPU off and 128 (=16*8) with TPU on
BATCH_SIZE = 16 * strategy.num_replicas_in_sync
AUGMENTATIONS = 4
ds_train = get_training_dataset(AUGMENTATIONS)
ds_valid = get_validation_dataset()
ds_test = get_test_dataset()
print("Training:", ds_train)
print ("Validation:", ds_valid)
print("Test:", ds_test)
np.set_printoptions(threshold=15, linewidth=80)
print("Training data shapes:")
for image, label in ds_train.take(3):
print(image.numpy().shape, label.numpy().shape)
print("Training data label examples:", label.numpy())
print("Test data shapes:")
for image, idnum in ds_test.take(3):
print(image.numpy().shape, idnum.numpy().shape)
print("Test data IDs:", idnum.numpy().astype('U')) # U=unicode string
EPOCHS = 12
with strategy.scope():
# Xception model pre-trained on the imagenet dataset
pretrained_model = tf.keras.applications.Xception(
weights='imagenet',
include_top=False,
input_shape=[*IMAGE_SIZE, 3]
)
pretrained_model.trainable = True
model = tf.keras.Sequential([
pretrained_model,
# ... attach a new head to act as a classifier
tf.keras.layers.GlobalAveragePooling2D(),
tf.keras.layers.Dense(units=len(CLASSES), activation='softmax'),
])
# compile model
model.compile(
optimizer='adam',
loss = 'sparse_categorical_crossentropy',
metrics=['sparse_categorical_accuracy'],
)
model.summary()
# Learning Rate Schedule for Fine Tuning #
def exponential_lr(epoch,
start_lr = 0.00001, min_lr = 0.00001, max_lr = 0.00005,
rampup_epochs = 5, sustain_epochs = 0,
exp_decay = 0.8):
def lr(epoch, start_lr, min_lr, max_lr, rampup_epochs, sustain_epochs, exp_decay):
# linear increase from start to rampup_epochs
if epoch < rampup_epochs:
lr = ((max_lr - start_lr) /
rampup_epochs * epoch + start_lr)
# constant max_lr during sustain_epochs
elif epoch < rampup_epochs + sustain_epochs:
lr = max_lr
# exponential decay towards min_lr
else:
lr = ((max_lr - min_lr) *
exp_decay**(epoch - rampup_epochs - sustain_epochs) +
min_lr)
return lr
return lr(epoch,
start_lr,
min_lr,
max_lr,
rampup_epochs,
sustain_epochs,
exp_decay)
lr_callback = tf.keras.callbacks.LearningRateScheduler(exponential_lr, verbose=True)
# Define training epochs
STEPS_PER_EPOCH = NUM_TRAINING_IMAGES * AUGMENTATIONS // BATCH_SIZE
with strategy.scope():
history = model.fit(
ds_train,
validation_data=ds_valid,
epochs=EPOCHS,
steps_per_epoch=STEPS_PER_EPOCH,
callbacks=[lr_callback],
)
cmdataset = get_validation_dataset(ordered=True)
images_ds = cmdataset.map(lambda image, label: image)
labels_ds = cmdataset.map(lambda image, label: label).unbatch()
cm_correct_labels = next(iter(labels_ds.batch(NUM_VALIDATION_IMAGES))).numpy()
cm_probabilities = model.predict(cmdataset, steps=NUM_VALIDATION_IMAGES//BATCH_SIZE)
print(cm_probabilities)
print(len(cm_probabilities))
print(tf.shape(cm_probabilities))
cm_predictions = np.argmax(cm_probabilities, axis=-1)
print(cm_predictions)
I then ran model.predict() on the validation dataset. What I wanted to do is to see how well the model performs when classifying each of the classes and create a confusion matrix using this data since model.evaluate() only determines how well the model performs overall. I will also be planning to use model.predict() to run the model on the test set and submit the predictions.
When running the model.predict() function, if I recall correctly, I should get a list of arrays for each image in the validation set which contain the weights for each class. However, I run into an issue where model.predict() is instead returning a single array with a length equal to the step size which there were 29 in this case (3712 validation images / 128 batch size).
[127.99999 127.999985 128. ... 127.999985 127.999985 127.99999 ]
29
tf.Tensor([29], shape=(1,), dtype=int32)
2
I've tried different models (I did ResNet50V2, VGG16, and a basic CNN model) and they all yielded the same problem where I only get a single array of size 29. My only guess as to what is going on here is ether an issue with my classification layer or with the dataset, perhaps shape related.