def LoadBatch(filename):
"""
Loads batch based on the given filename and produces the X, Y, and y arrays
:param filename: Path of the file
:return: X, Y and y arrays
"""
# borrowed from https://www.cs.toronto.edu/~kriz/cifar.html
def unpickle(file):
with open(file, 'rb') as fo:
dict = pickle.load(fo, encoding='bytes')
return dict
dictionary = unpickle(filename)
# borrowed from https://stackoverflow.com/questions/16977385/extract-the-nth-key-in-a-python-dictionary?utm_medium=organic&utm_source=google_rich_qa&utm_campaign=google_rich_qa
def ix(dic, n): # don't use dict as a variable name
try:
return list(dic)[
n] # or sorted(dic)[n] if you want the keys to be sorted
except IndexError:
print('not enough keys')
garbage = ix(dictionary, 1)
y = dictionary[garbage]
Y = np.transpose(make_class_categorical(y, 10))
garbage = ix(dictionary, 2)
X = np.transpose(dictionary[garbage]) / 255
return X, Y, y
def LoadBatch(filename):
"""
Loads a CIFAR-10 batch of data.
:param filename: The path of the file in your local computer.
:return: CIFAR-10 data X, their one-hot representation Y, and their true labels y
"""
# borrowed from https://www.cs.toronto.edu/~kriz/cifar.html
def unpickle(file):
with open(file, 'rb') as fo:
dict = pickle.load(fo, encoding='latin1')
return dict
dictionary = unpickle(filename)
# borrowed from https://stackoverflow.com/questions/16977385/extract-the-nth-key-in-a-python-dictionary?utm_medium=organic&utm_source=google_rich_qa&utm_campaign=google_rich_qa
def ix(dic, n): # don't use dict as a variable name
try:
return list(dic)[
n] # or sorted(dic)[n] if you want the keys to be sorted
except IndexError:
print('not enough keys')
garbage = ix(dictionary, 1)
y = dictionary[garbage]
Y = np.transpose(make_class_categorical(y, 10))
garbage = dictionary['data']
X = np.transpose(garbage) / 255.0
return X, Y, y
def create_augmented_dataset(X, y):
"""
Creates an augmented dataset, by appying random transformations in each datum.
The transformed images are then concatenated to the originals, thus extending the original dataset size.
One-hot representations and true labels of the generated images are also added in the Y and y matrices.
:param X: Training data.
:param y: Data true labels
:return: Extended training data, one-hot representations and true labels.
"""
X_augmented = np.copy(X)
y_augmented = y.copy()
from keras.preprocessing.image import ImageDataGenerator
datagen = ImageDataGenerator(rotation_range=90,
width_shift_range=0.1,
height_shift_range=0.1,
horizontal_flip=True)
data = np.ndarray(shape=(X.shape[1], 32, 32, 3))
for datum in range(X.shape[1]):
data[datum, :] = X[:, datum].reshape(3, 32, 32).transpose(1, 2, 0)
cnt = 0
augmented = np.ndarray(shape=(3072, X.shape[1]))
labels = []
for X_batch, y_batch in datagen.flow(data, y, batch_size=1):
augmented[:, cnt] = X_batch[0].transpose(2, 1, 0).reshape(3072)
labels.append(y_batch[0])
cnt += 1
if cnt == X.shape[1]:
break
X_augmented = np.copy(np.concatenate((X_augmented, augmented), axis=1))
y_augmented.extend(labels)
Y_augmented = make_class_categorical(y_augmented, 10).T
return X_augmented, Y_augmented, y_augmented