-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdownstream.py
172 lines (149 loc) · 6.36 KB
/
downstream.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
from libauc.losses import AUCMLoss, CrossEntropyLoss
from libauc.optimizers import PESG, Adam
from libauc.models import DenseNet121, DenseNet169
from torchvision.models import densenet
from libauc.datasets import CheXpert
import torch
from PIL import Image
import numpy as np
import torchvision.transforms as transforms
from torch.utils.data import Dataset
from sklearn.metrics import roc_auc_score
import torch.nn as nn
import wandb
def set_all_seeds(SEED):
# REPRODUCIBILITY
torch.manual_seed(SEED)
np.random.seed(SEED)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
wandb.init(project="evaluating-mia-densenet", entity="vasudev13")
wandb.config = {
"learning_rate": 0.05,
"epochs": 2,
"batch_size": 32
}
SEED = 13
set_all_seeds(SEED)
PATH = '/scratch/va2134/densenet121_encoder.ckpt'
state_dict = torch.load(PATH)
model = densenet.densenet121(pretrained=False)
model.classifier = nn.Identity()
model.load_state_dict(state_dict)
model.classifier = nn.Linear(in_features=1024, out_features=5, bias=True)
model.cuda()
root = '/scratch/va2134/datasets/CheXpert-v1.0-small/'
Index: -1 denotes multi-label mode including 5 diseases
traindSet = CheXpert(csv_path=root+'train.csv', image_root_path=root, use_upsampling=False, use_frontal=True, image_size=224, mode='train', class_index=-1)
testSet = CheXpert(csv_path=root+'valid.csv', image_root_path=root, use_upsampling=False, use_frontal=True, image_size=224, mode='valid', class_index=-1)
trainloader = torch.utils.data.DataLoader(traindSet, batch_size=32, num_workers=2, shuffle=True)
testloader = torch.utils.data.DataLoader(testSet, batch_size=32, num_workers=2, shuffle=False)
# paramaters
BATCH_SIZE = 32
lr = 1e-4
weight_decay = 1e-5
# define loss & optimizer
CELoss = CrossEntropyLoss()
optimizer = Adam(model.parameters(), lr=lr, weight_decay=weight_decay)
# training
best_val_auc = 0
for epoch in range(1):
for idx, data in enumerate(trainloader):
train_data, train_labels = data
train_data, train_labels = train_data.cuda(), train_labels.cuda()
y_pred = model(train_data)
loss = CELoss(y_pred, train_labels)
optimizer.zero_grad()
loss.backward()
optimizer.step()
# validation
if idx % 400 == 0:
model.eval()
with torch.no_grad():
test_pred = []
test_true = []
for jdx, data in enumerate(testloader):
test_data, test_labels = data
test_data = test_data.cuda()
y_pred = torch.sigmoid(model(test_data))
test_pred.append(y_pred.cpu().detach().numpy())
test_true.append(test_labels.numpy())
test_true = np.concatenate(test_true)
test_pred = np.concatenate(test_pred)
val_auc_mean = roc_auc_score(test_true, test_pred)
model.train()
if best_val_auc < val_auc_mean:
best_val_auc = val_auc_mean
torch.save(model.state_dict(), '/scratch/va2134/ce_pretrained_model.pth')
print ('Epoch=%s, BatchID=%s, Val_AUC=%.4f, Best_Val_AUC=%.4f'%(epoch, idx, val_auc_mean, best_val_auc ))
## Edema Pre-training
class_id = 3 # 0:Cardiomegaly, 1:Edema, 2:Consolidation, 3:Atelectasis, 4:Pleural Effusion
root = '/scratch/va2134/datasets/CheXpert-v1.0-small/'
# You can set use_upsampling=True and pass the class name by upsampling_cols=['Cardiomegaly'] to do upsampling. This may improve the performance
traindSet = CheXpert(csv_path=root+'train.csv', image_root_path=root, use_upsampling=True, use_frontal=True, image_size=224, mode='train', class_index=class_id)
testSet = CheXpert(csv_path=root+'valid.csv', image_root_path=root, use_upsampling=False, use_frontal=True, image_size=224, mode='valid', class_index=class_id)
trainloader = torch.utils.data.DataLoader(traindSet, batch_size=32, num_workers=2, shuffle=True)
testloader = torch.utils.data.DataLoader(testSet, batch_size=32, num_workers=2, shuffle=False)
# paramaters
BATCH_SIZE = 32
imratio = traindSet.imratio
lr = 0.05 # using smaller learning rate is better
gamma = 500
weight_decay = 1e-5
margin = 1.0
# load pretrained model
if True:
PATH = '/scratch/va2134/ce_pretrained_model.pth'
state_dict = torch.load(PATH)
state_dict.pop('classifier.weight', None)
state_dict.pop('classifier.bias', None)
model = densenet.densenet121(pretrained=False)
model.classifier = nn.Identity()
model.load_state_dict(state_dict, strict=False)
model.classifier = nn.Linear(in_features=1024, out_features=1, bias=True)
model.cuda()
# define loss & optimizer
Loss = AUCMLoss(imratio=imratio)
optimizer = PESG(model,
a=Loss.a,
b=Loss.b,
alpha=Loss.alpha,
imratio=imratio,
lr=lr,
gamma=gamma,
margin=margin,
weight_decay=weight_decay)
best_val_auc = 0
for epoch in range(2):
if epoch > 0:
optimizer.update_regularizer(decay_factor=10)
for idx, data in enumerate(trainloader):
train_data, train_labels = data
train_data, train_labels = train_data.cuda(), train_labels.cuda()
y_pred = torch.sigmoid(model(train_data))
loss = Loss(y_pred, train_labels)
optimizer.zero_grad()
loss.backward()
optimizer.step()
# validation
if idx % 400 == 0:
model.eval()
with torch.no_grad():
test_pred = []
test_true = []
for jdx, data in enumerate(testloader):
test_data, test_label = data
test_data = test_data.cuda()
y_pred = model(test_data)
test_pred.append(y_pred.cpu().detach().numpy())
test_true.append(test_label.numpy())
test_true = np.concatenate(test_true)
test_pred = np.concatenate(test_pred)
val_auc = roc_auc_score(test_true, test_pred)
model.train()
if best_val_auc < val_auc:
best_val_auc = val_auc
torch.save(model.state_dict(), '/scratch/va2134/atelectasis_model.pth')
wandb.log({"atelectasis_val_auc": val_auc})
print ('Epoch=%s, BatchID=%s, Val_AUC=%.4f, lr=%.4f'%(epoch, idx, val_auc, optimizer.lr))
print ('Best Val_AUC is %.4f'%best_val_auc)