-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathmain.py
186 lines (157 loc) · 8.2 KB
/
main.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
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
from __future__ import print_function
import argparse
import os
from utils.file_utils import save_pkl
from utils.utils import *
from utils.core_utils import train
from datasets.dataset_generic import Generic_MIL_Dataset
import torch
import pandas as pd
import numpy as np
device=torch.device("cuda" if torch.cuda.is_available() else "cpu")
# Generic training settings
parser = argparse.ArgumentParser(description='Configurations for WSI Training')
parser.add_argument('--data_root_dir', type=str, default=None, help='data directory')
parser.add_argument('--data_folder_s', type=str, default=None, help='dir under data directory' )
parser.add_argument('--data_folder_l', type=str, default=None, help='dir under data directory' )
parser.add_argument('--max_epochs', type=int, default=200, help='maximum number of epochs to train (default: 200)')
parser.add_argument('--lr', type=float, default=1e-3, help='learning rate (default: 0.0001)')
parser.add_argument('--label_frac', type=float, default=1.0, help='fraction of training labels (default: 1.0)')
parser.add_argument('--seed', type=int, default=1, help='random seed for reproducible experiment (default: 1)')
parser.add_argument('--k', type=int, default=10, help='number of folds (default: 10)')
parser.add_argument('--k_start', type=int, default=-1, help='start fold (default: -1, last fold)')
parser.add_argument('--k_end', type=int, default=-1, help='end fold (default: -1, first fold)')
parser.add_argument('--results_dir', default='./results', help='results directory (default: ./results)')
parser.add_argument('--split_dir', type=str, default=None)
parser.add_argument('--log_data', action='store_true', default=False, help='log data using tensorboard')
parser.add_argument('--testing', action='store_true', default=False, help='debugging tool')
parser.add_argument('--early_stopping', action='store_true', default=False, help='enable early stopping')
parser.add_argument('--opt', type=str, choices = ['adam', 'sgd'], default='adam')
parser.add_argument('--drop_out', action='store_true', default=False, help='enabel dropout (p=0.25)')
parser.add_argument('--model_type', type=str, choices=['ViLa_MIL'], default='ViLa_MIL', help='type of model')
parser.add_argument('--mode', type=str, choices=['transformer'], default='transformer')
parser.add_argument('--exp_code', type=str, help='experiment code for saving results')
parser.add_argument('--weighted_sample', action='store_true', default=False, help='enable weighted sampling')
parser.add_argument('--reg', type=float, default=1e-5, help='weight decay (default: 1e-5)')
parser.add_argument('--bag_loss', type=str, choices=['svm', 'ce', 'focal'], default='ce')
parser.add_argument('--task', type=str)
parser.add_argument("--text_prompt", type=str, default=None)
parser.add_argument("--text_prompt_path", type=str, default=None)
parser.add_argument("--prototype_number", type=int, default=None)
args = parser.parse_args()
args.text_prompt = np.array(pd.read_csv(args.text_prompt_path, header=None)).squeeze()
def seed_torch(seed=7):
import random
random.seed(seed)
os.environ['PYTHONHASHSEED'] = str(seed)
np.random.seed(seed)
torch.manual_seed(seed)
if device.type == 'cuda':
torch.cuda.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
torch.backends.cudnn.benchmark = False
torch.backends.cudnn.deterministic = True
seed_torch(args.seed)
settings = {'num_splits': args.k,
'k_start': args.k_start,
'k_end': args.k_end,
'task': args.task,
'max_epochs': args.max_epochs,
'results_dir': args.results_dir,
'lr': args.lr,
'experiment': args.exp_code,
'label_frac': args.label_frac,
'seed': args.seed,
'model_type': args.model_type,
'mode': args.mode,
"use_drop_out": args.drop_out,
'weighted_sample': args.weighted_sample,
'opt': args.opt}
print('\nLoad Dataset')
if args.task == 'task_tcga_rcc_subtyping':
args.n_classes=3
dataset = Generic_MIL_Dataset(csv_path = 'dataset_csv/TCGA_RCC_subtyping.csv',
mode = args.mode,
data_dir_s = os.path.join(args.data_root_dir, args.data_folder_s),
data_dir_l = os.path.join(args.data_root_dir, args.data_folder_l),
shuffle = False,
print_info = True,
label_dict = {'CCRCC':0, 'PRCC':1, 'CRCC':2},
patient_strat= False,
ignore=[])
elif args.task == 'task_tcga_lung_subtyping':
args.n_classes=2
dataset = Generic_MIL_Dataset(csv_path = 'dataset_csv/TCGA_Lung_subtyping.csv',
mode = args.mode,
data_dir_s = os.path.join(args.data_root_dir, args.data_folder_s),
data_dir_l = os.path.join(args.data_root_dir, args.data_folder_l),
shuffle = False,
print_info = True,
label_dict = {'LUAD':0, 'LUSC':1},
patient_strat= False,
ignore=[])
else:
raise NotImplementedError
if not os.path.exists(args.results_dir):
os.makedirs(args.results_dir)
args.results_dir = os.path.join(args.results_dir, str(args.exp_code) + '_s{}'.format(args.seed))
if not os.path.exists(args.results_dir):
os.makedirs(args.results_dir)
if args.split_dir is None:
args.split_dir = os.path.join('splits', args.task+'_{}'.format(int(args.label_frac*100)))
else:
args.split_dir = os.path.join('splits', args.split_dir)
print('split_dir: ', args.split_dir)
assert os.path.isdir(args.split_dir)
settings.update({'split_dir': args.split_dir})
with open(args.results_dir + '/experiment_{}.txt'.format(args.exp_code), 'w') as f:
print(settings, file=f)
f.close()
print("################# Settings ###################")
for key, val in settings.items():
print("{}: {}".format(key, val))
def main(args):
if args.k_start == -1:
start = 0
else:
start = args.k_start
if args.k_end == -1:
end = args.k
else:
end = args.k_end
all_test_auc = []
all_val_auc = []
all_test_acc = []
all_val_acc = []
all_test_f1 = []
folds = np.arange(start, end)
for i in folds:
seed_torch(args.seed)
train_dataset, val_dataset, test_dataset = dataset.return_splits(from_id=False, csv_path='{}/splits_{}.csv'.format(args.split_dir, i))
datasets = (train_dataset, val_dataset, test_dataset)
results, test_auc, val_auc, test_acc, val_acc, _, test_f1 = train(datasets, i, args)
all_test_auc.append(test_auc)
all_val_auc.append(val_auc)
all_test_f1.append(test_f1)
all_test_acc.append(test_acc)
all_val_acc.append(val_acc)
filename = os.path.join(args.results_dir, 'split_{}_results.pkl'.format(i))
save_pkl(filename, results)
final_df = pd.DataFrame({'folds': folds, 'test_auc': all_test_auc, 'test_acc': all_test_acc, 'test_f1': all_test_f1})
result_df = pd.DataFrame({'metric': ['mean', 'var'],
'test_auc': [np.mean(all_test_auc), np.std(all_test_auc)],
'test_f1': [np.mean(all_test_f1), np.std(all_test_f1)],
'test_acc': [np.mean(all_test_acc), np.std(all_test_acc)],
})
if len(folds) != args.k:
save_name = 'summary_partial_{}_{}.csv'.format(folds[0], folds[-1])
result_name = 'result_partial_{}_{}.csv'.format(folds[0], folds[-1])
else:
save_name = 'summary.csv'
result_name = 'result.csv'
result_df.to_csv(os.path.join(args.results_dir, result_name), index=False)
final_df.to_csv(os.path.join(args.results_dir, save_name))
if __name__ == "__main__":
results = main(args)
print("finished!")
print("end script")