-
Notifications
You must be signed in to change notification settings - Fork 1
/
baseline.py
150 lines (100 loc) · 5.28 KB
/
baseline.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
import torch
import random
import argparse
import math
import torchvision
import utils
import pathlib
import numpy as np
import utils.cross_validate as CV
from model import trainer, pipeline
# from pytorch_pretrained_vit import ViT
# import pathlib
# import os
# from torch.utils.data.sampler import SubsetRandomSampler
# import model.metric as metric
# main program
def run_spatial(args=None):
### Seed ###
random.seed(args.seed)
np.random.seed(args.seed)
torch.manual_seed(args.seed)
### Select device for computation ###
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
### Split patients into train/test ###
patients = sorted(utils.util.get_spatial_patients().keys())
test_patients = ["BC23450", "BC23903"]
train_patients = [p for p in patients if p not in test_patients]
print("Train patients: ", train_patients)
print("Test patients: ", test_patients)
print("Parameters: ", args)
print()
# ### cross-validation to get the best epoch
best_loss_epoch, best_aMAE_epoch, best_aRMSE_epoch, best_aCC_epoch = CV.get_cv_resluts(train_patients, args.cv_fold, args, device)
# print("Best loss epoch is: ", best_loss_epoch)
# print("Best aMAE epoch is: ", best_aMAE_epoch)
# print("Best aRMSE epoch is: ", best_aRMSE_epoch)
# print("Best aCC epoch is: ", best_aCC_epoch)
# print()
best_epoch = math.ceil(np.mean(np.array(([best_loss_epoch, best_aMAE_epoch, best_aRMSE_epoch, best_aCC_epoch]))))
print("Best CV epoch: ", best_epoch)
print()
# best_epoch = 10
### cross-validation to get the best epoch
print("### START MAIN PROGRAM:")
print()
print("Train patients: ", train_patients)
print("Test patients: ", test_patients)
print("Parameters: ", args)
### main network
model, train_loader, test_loader, optim, lr_scheduler, criterion = pipeline.setup(train_patients, test_patients, args, device)
if best_epoch <= 3: # for debug
best_epoch = 3
for epoch in range(best_epoch):
if args.debug and epoch==3:
break
print("Epoch #" + str(epoch + 1) + ":")
train_loss = trainer.fit(model, train_loader, optim, criterion, args, device)
lr_scheduler.step()
# here test will save should change the file name
test_loss = trainer.test(model, test_loader, criterion, device, args, best_epoch)
### TODO: best_epoch = 0, skip the for loop and direct save model?
# if args.debug:
# pass
# else:
torch.save(model, args.pred_root + '/model.pkl')
print()
torch.cuda.empty_cache()
#
# # model.save (the best model should be the current epoch - patience)
# # relative
# lr_scheduler(val_loss)
# early_stopping(val_loss)
# if early_stopping.early_stop: # when break, the model is at the final epoch
# break
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Process the paths.')
parser.add_argument('--seed', type=int, default=0, help='seed for reproduction')
parser.add_argument("--cv_fold", type=int, default=5, help="cv fold for cross-validation")
parser.add_argument("--batch", type=int, default=32, help="training batch size")
parser.add_argument("--learning_rate", type=float, default=1e-3, help="learning rate")
parser.add_argument("--cv_epochs", type=int, default=50, help="number of cross-validation epochs")
parser.add_argument("--workers", type=int, default=8, help="number of workers for dataloader")
parser.add_argument("--window", type=int, default=224, help="window size") # try 128 150 224 299 512 (smaller, normal, and bigger)
parser.add_argument("--resolution", type=int, default=224, help="resolution") # try
parser.add_argument("--model", type=str, default= 'densenet121', help="choose different model") # alexnet, vgg16, resnet101, densenet121, inception_v3, efficientnet_b7
parser.add_argument("--pretrained", action="store_true", help="use ImageNet pretrained model?")
parser.add_argument("--finetuning", type=str, default= None, help="use ImageNet pretrained model with fine tuning fcs?")
parser.add_argument("--gene_filter", default=250, type =int,
help="specific predicted main genes (defalt use all the rest for aux tasks)")
parser.add_argument("--aux_ratio", default=1, type =float,
help="specific the number of aux genes")
parser.add_argument("--aux_weight", default=1, type =float,
help="specific the loss weight of aux genes")
parser.add_argument("--epochs", type=int, default=50, help="number of epochs")
parser.add_argument("--pred_root", type=str, default="output/", help="root for prediction outputs")
parser.add_argument("--debug", action="store_true", help="debug")
args = parser.parse_args()
# set different log name
# pathlib.Path(os.path.dirname(args.pred_root)).mkdir(parents=True, exist_ok=True)
run_spatial(args)