def __init__(self, args):
size = 100
self.h = 10
self.w = 10
self.rs = args.rs
random_state = np.random.RandomState(seed=self.rs)
self.images = []
self.bag_labels = []
self.instance_labels = []
for i in range(size):
data, instance_label, b_label = generate_positive_image(
self.h, self.w, random_state)
self.images.append(data)
self.bag_labels.append(b_label)
self.instance_labels.append(instance_label)
#Uncomment for MIMN definition of positive image
data, instance_label, b_label = generate_negative_image(
self.h, self.w, random_state, 0)
#data, instance_label, b_label = generate_negative_image(self.h, self.w, random_state)
self.images.append(data)
self.bag_labels.append(b_label)
self.instance_labels.append(instance_label)
self.features = ravel_image(self.images)
self.instance_labels = ravel_image(self.instance_labels)
self.features, self.bag_labels, self.instance_labels = shuffle_dataset(
self.features, self.bag_labels, self.rs, self.instance_labels)
self.random_shuffle()
def random_shuffle(self):
self.features, self.bag_labels = shuffle_dataset(self.features, self.bag_labels, self.rs)
x_train, x_test, y_train, y_test = train_test_split(self.features, self.bag_labels)
self.training_data = x_train
self.testing_data = x_test
self.training_labels = y_train
self.testing_labels = y_test
def get_dataset(random_seed):
"""
Fetches dataset from precalculated descriptors.
:return: features, bag_labels
"""
list_of_positive_testing_bags = []
list_of_negative_testing_bags = []
list_of_positive_training_bags = []
list_of_negative_training_bags = []
features = [] # list for all bags (list containing instances)
bag_labels = [] # list for all bag labels
# Appends all pkl files to list of their
for file in os.listdir(
"/home/dub/python/multiple_instance_learning/data/synthetic/testing/normal/"
):
if file.endswith('.pkl'):
list_of_negative_testing_bags.append(
os.path.join(
"/home/dub/python/multiple_instance_learning/data/synthetic/testing/normal/",
file))
for file in os.listdir(
'/home/dub/python/multiple_instance_learning/data/synthetic/testing/tumor/'
):
if file.endswith('.pkl'):
list_of_positive_testing_bags.append(
os.path.join(
"/home/dub/python/multiple_instance_learning/data/synthetic/testing/tumor/",
file))
for file in os.listdir(
'/home/dub/python/multiple_instance_learning/data/synthetic/training/normal/'
):
if file.endswith('.pkl'):
list_of_negative_training_bags.append(
os.path.join(
"/home/dub/python/multiple_instance_learning/data/synthetic/training/normal/",
file))
for file in os.listdir(
'/home/dub/python/multiple_instance_learning/data/synthetic/training/tumor/'
):
if file.endswith('.pkl'):
list_of_positive_training_bags.append(
os.path.join(
"/home/dub/python/multiple_instance_learning/data/synthetic/training/tumor/",
file))
# appends all instances and bag labels into lists
for instance in list_of_positive_training_bags: # positive instances
with open(instance, 'rb') as tfh:
tsd = pickle.load(tfh)
if np.array(len(tsd['instances'][0])) == 16:
features.append(np.array(tsd['instances']))
bag_labels.append(tsd['bag_label'])
for instance in list_of_negative_training_bags: # negative instances
with open(instance, 'rb') as tfh:
tsd = pickle.load(tfh)
if np.array(len(tsd['instances'][0])) == 16:
features.append(np.array(tsd['instances']))
bag_labels.append(-tsd['bag_label'])
for instance in list_of_positive_testing_bags: # positive instances
with open(instance, 'rb') as tfh:
tsd = pickle.load(tfh)
if np.array(len(tsd['instances'][0])) == 16:
features.append(np.array(tsd['instances']))
bag_labels.append(-tsd['bag_label'])
for instance in list_of_negative_testing_bags: # negative instances
with open(instance, 'rb') as tfh:
tsd = pickle.load(tfh)
if np.array(len(tsd['instances'][0])) == 16:
features.append(np.array(tsd['instances']))
bag_labels.append(tsd['bag_label'])
features, bag_label = shuffle_dataset(features, bag_label, random_seed)
return features, bag_labels