-
Notifications
You must be signed in to change notification settings - Fork 2
/
s3_CaseSplitting.py
146 lines (111 loc) · 8.18 KB
/
s3_CaseSplitting.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
import os, argparse
import numpy as np
from datasets.dataset_generic import Generic_WSI_Classification_Dataset, split_slideinfo
"""
class: Case Splitting
"""
class CaseSplitting(object):
def __init__(self, task_name = None, csv_info_path=None, split_to_dir=None, label_column_name="label", label_list=['0', '1'], filter_dict={},
seed=2020, shuffle=False, tvtmode=False, slide_featspath=None):
f"""
task_name: task name; such as "recurrence" or "survival"
csv_info_path: clinical info csv file with slide_id column and label info
split_to_dir: dir for saving cases splitting files
label_column_name: (default: `label`) name of label column to be used.
label_list: (default: ['0', '1']) label lists for mapping class names in label column name to 0, 1, 2 ...
seed: (default: 2020) 'random seed
shuffle: (default: False) shuffle slide data or not
tvtmode (default: False) default is cross validation mode, if True, train-val-test mode
"""
assert task_name is not None, "WHY not assign a TASK NAME??? tell me the task name please!"
assert split_to_dir is not None, f"directory to save cases splits files {split_to_dir}"
if tvtmode:
split_to_dir = os.path.join(split_to_dir, task_name + "_TrainValTest")
else:
split_to_dir = os.path.join(split_to_dir, task_name + "_KFoldsCV")
os.makedirs(split_to_dir, exist_ok=True)
self.split_dir = split_to_dir
assert csv_info_path is not None and os.path.isfile(csv_info_path), f"csv info path {csv_info_path} must be given or file do not EXIST :)"
dataset = Generic_WSI_Classification_Dataset(csv_path = csv_info_path,
label_col = label_column_name, label_dict = dict(zip(label_list, range(len(label_list)))),
filter_dict=filter_dict,
seed = seed, shuffle = shuffle, print_info = True,
patient_strat= True, # TRUE 表示按照patient进行划分split，保证一个patient的所有slide被split在同一区间内，如train或val；防止数据泄露 任何时候都应该这样
patient_voting='maj', # maj合理；max是对于一个patient多个slide标签选标签值最大，不合理
ignore=[],
slide_featspath=slide_featspath
)
self.dataset = dataset
def run(self, times=1, kfold=0, val_frac=0.2, test_frac=0.2):
num_slides_cls = np.array([len(cls_ids) for cls_ids in self.dataset.patient_cls_ids])
test_num = np.round(num_slides_cls * test_frac).astype(int)
# train_num = num_slides_cls - test_num
if kfold: # 在kfold不为0时，表示采用kfolds Cross validation mode mode，此时args.val_frac无效
assert kfold > 1, "K folds num parameter should great than 1"
for t in range(times):
print(f"[*(//@_@)*]@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@[*(//@_@)*] Time: {t} Cross Validation Folds {kfold}")
self.mode_cv_kfolds(t, kfold, test_num)
elif kfold == 0: # 在kfold为0时，表示采用train-val-test mode，此时args.val_frac有效
val_num = np.round(num_slides_cls * val_frac).astype(int)
# train_num = train_num - val_num # 确定train val test num
for t in range(times):
print(f"[*(//@_@)*]@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@[*(//@_@)*] Time: {t} train-val-test")
self.mode_train_val_test(t, val_num, test_num)
def mode_cv_kfolds(self, times=1, kfold=10, test_num=0):
self.dataset.create_splits_CrossValidation(k = kfold, test_num = test_num)
for kval in range(kfold):
print(f">>>>>>>>>>>>>>>>>>>>>>>> Splitting, Now Fold {kval}")
self.dataset.set_splits()
descriptor_df = self.dataset.split_trainval_cls_num_gen(return_descriptor=True)
splits = self.dataset.return_splits(from_id=True)
df_split_list = split_slideinfo(splits, ['train', 'val', 'test'])
df_splist_bool = split_slideinfo(splits, ['train', 'val', 'test'], boolean_style=True)
descriptor_df.to_csv(os.path.join(self.split_dir, f'splits_descriptor_time{times}_fold{kval}.csv'))
df_split_list.to_csv(os.path.join(self.split_dir, f'splits_time{times}_fold{kval}.csv'))
df_splist_bool.to_csv(os.path.join(self.split_dir, f'splits_bool_time{times}_fold{kval}.csv'))
def mode_train_val_test(self, times=1, val_num=0, test_num=0):
self.dataset.create_splits(k = 1, val_num = val_num, test_num = test_num, label_frac=1)
self.dataset.set_splits()
descriptor_df = self.dataset.split_trainval_cls_num_gen(return_descriptor=True)
splits = self.dataset.return_splits(from_id=True)
df_split_list = split_slideinfo(splits, ['train', 'val', 'test'])
df_splist_bool = split_slideinfo(splits, ['train', 'val', 'test'], boolean_style=True)
descriptor_df.to_csv(os.path.join(self.split_dir, f'splits_descriptor_time{times}.csv'))
df_split_list.to_csv(os.path.join(self.split_dir, f'splits_time{times}.csv'))
df_splist_bool.to_csv(os.path.join(self.split_dir, f'splits_bool_time{times}.csv'))
def set_args():
parser = argparse.ArgumentParser(description='Creating cases splits for whole slide classification')
parser.add_argument('--task_name', type=str, default=None)
parser.add_argument('--csv_info_path', type=str, default=None, help='csv file with slide_id and label information')
parser.add_argument('--split_to_dir', type=str, default=None)
parser.add_argument('--times', type=int, default=1,
help='(default: 1) number of times, valid both in Cross validation mode and train-val-test mode')
parser.add_argument('--kfold', type=int, default=10,
help='(default: 10) number of folds, common: Cross validation mode; if 0, means NOT CrossValidation, use train-val-test mode')
parser.add_argument('--val_frac', type=float, default= 0.2,
help='(default: 0.2) fraction of samples for validation, only use in train-val-test mode, not used for cross validation')
parser.add_argument('--test_frac', type=float, default= 0.2,
help='(default: 0.2) fraction of samples for test, use in both cross validation and train-val-test modes')
parser.add_argument('--label_column_name', type=str, default="label",
help='(default: label) name of label column to be used.')
parser.add_argument('--label_list', nargs='+', default=['0', '1'],
help=f'(default: [0, 1]) label lists for mapping class names in label column name to 0, 1, 2 ... IN Order.')
parser.add_argument('--slide_featspath', type=str, default=None, help='slide feats path for useful slide selection')
parser.add_argument('--seed', type=int, default=2020,
help='random seed (default: 2020)')
parser.add_argument('--shuffle', default=False, action='store_true')
args = parser.parse_args()
return args
if __name__ == "__main__":
args = set_args()
# args.task_name = "tmptest"
# args.csv_info_path = "/home/cyyan/Projects/HER2proj/data/HEcasesFullInfo再整理0317.csv"
# args.split_to_dir = "/home/cyyan/Projects/HER2proj/results/TJMUCH_3CaseSplits"
# args.label_column_name = "复发"
# args.test_frac = 0
# args.times = 5
# args.kfold = 5
args.filter_dict = {args.label_column_name: args.label_list}
casesplit = CaseSplitting(args.task_name, args.csv_info_path, args.split_to_dir, args.label_column_name, args.label_list, args.filter_dict,
args.seed, args.shuffle, tvtmode=args.kfold==0, slide_featspath=args.slide_featspath)
casesplit.run(args.times, args.kfold, args.val_frac, args.test_frac)