def split_generator(X, Y, n_cv=5, random_seed=1234):
""""""
if Y[2].isna().unique() == True:
label = Y[3].values
# returning generator
split_gen = StratifiedKFold(n_splits=n_cv,
shuffle=True,
random_state=random_seed).split(X, label)
else:
train_ix = Y[Y[2].isin({'train', 'valid'})][0].values
test_ix = Y[Y[2].isin({'test'})][0].values
split_gen = []
for _ in range(n_cv):
np.random.shuffle(train_ix)
split_gen.append((train_ix, test_ix))
for train_ix, test_ix in split_gen:
yield train_ix, test_ix
def data_partition(sample_size_list):
num_sum = 0
seed_ = 42
folds_num = 5
label_all = []
for i in range(len(sample_size_list)):
tmp_labels = [float(i)] * sample_size_list[i]
label_all += tmp_labels
num_sum += sample_size_list[i]
label_all = np.array(label_all)
pse_data = np.random.normal(loc=0.0, scale=1.0, size=[num_sum, num_sum])
folds = StratifiedKFold(folds_num,
shuffle=True,
random_state=np.random.RandomState(seed_))
folds_temp = list(folds.split(pse_data, label_all))
folds = []
for i in range(folds_num):
train_index = folds_temp[i][0]
test_index = folds_temp[i][1]
folds.append((train_index, test_index))
return folds
