def get_data_paths(self, path, folds=5):
'''
Get splitted data paths.
'''
total_folds_itens = np.zeros(folds)
# Get data files.
brains_data = Files(path)
brains_data = brains_data.get_file_names()
total_brains = len(brains_data)
fold_size = np.int(total_brains * 1.0 / folds + 0.5)
# Get total of brains by category.
brains_by_patient = group_brains_by_patient_id(brains_data)
patient_ids = brains_by_patient.keys()
np.random.shuffle(patient_ids)
fold_itens = []
for _ in range(folds):
fold_itens.append([])
# Split data, avoiding brains of the same patient being moved to
# different sets.
for patient in patient_ids:
delta_folds_itens = fold_size - total_folds_itens
selected_fold = np.argmax(delta_folds_itens)
for brain in brains_by_patient[patient]:
fold_itens[selected_fold].append(brain)
total_folds_itens[selected_fold] += 1
self.fold_itens = np.array(fold_itens)
def split_train_data(self, paths, ratio=None):
'''
Get splitted data paths.
'''
if ratio == None:
ratio = 0.05 / (1 - 1.0 / len(self.fold_itens))
# Get data files.
brains_data = Files("")
brains_data.paths = paths
brains_data = brains_data.get_file_names()
total_brains = len(brains_data)
validation_size = np.int(total_brains * ratio)
# Get total of brains by category.
brains_by_patient = group_brains_by_patient_id(brains_data)
validation_paths = []
train_paths = []
brains_by_category = group_brains_by_category(brains_data)
statistic = {}
# Initialize statistic data.
for label in brains_by_category:
statistic[label] = np.round(
len(brains_by_category[label]) * validation_size * 1.0 /
total_brains)
# Create train and validation set.
for label in brains_by_category:
index_brains = 0
while statistic[label] > 0:
patient_id = get_patient_id(
brains_by_category[label][index_brains])
brains_patient = brains_by_patient[patient_id]
for brain_patient in brains_patient:
validation_paths.append(brain_patient)
statistic[get_category(brain_patient)] -= 1
index_brains += 1
for brain in brains_data:
if brain not in validation_paths:
train_paths.append(brain)
# Shuffle data.
np.random.shuffle(np.array(validation_paths))
np.random.shuffle(np.array(train_paths))
return train_paths, validation_paths
