-
Notifications
You must be signed in to change notification settings - Fork 6
/
siScore.py
183 lines (148 loc) · 7 KB
/
siScore.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
import torch
import torch.nn as nn
import torchvision.transforms as transforms
import torchvision.models as models
import torch.backends.cudnn as cudnn
import pandas as pd
from torch.utils.data import DataLoader
from utils.graph import *
from utils.siScore_utils import *
from utils.parameters import *
def make_data_loader(cluster_list, batch_sz):
cluster_dataset = ClusterDataset(cluster_list, dir_name = args.dir_name, transform = transforms.Compose([
RandomRotate(),
ToTensor(),
Grayscale(prob = 0.1),
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
]))
cluster_loader = torch.utils.data.DataLoader(cluster_dataset, batch_size=batch_sz, shuffle=True, num_workers=4, drop_last=True)
return cluster_loader
def generate_loader_dict(total_list, unified_cluster_list, batch_sz):
loader_dict = {}
for cluster_id in total_list:
cluster_loader = make_data_loader([cluster_id], batch_sz)
loader_dict[cluster_id] = cluster_loader
for cluster_tuple in unified_cluster_list:
cluster_loader = make_data_loader(cluster_tuple, batch_sz)
for cluster_num in cluster_tuple:
loader_dict[cluster_num] = cluster_loader
return loader_dict
def deactivate_batchnorm(model):
for m in model.modules():
if isinstance(m, nn.BatchNorm2d):
m.reset_parameters()
m.eval()
with torch.no_grad():
m.weight.fill_(1.0)
m.bias.zero_()
def train(args, epoch, model, optimizer, loader_list, cluster_path_list, device):
model.train()
# Deactivate the batch normalization before training
deactivate_batchnorm(model.module)
train_loss = AverageMeter()
reg_loss = AverageMeter()
# For each cluster route
path_idx = 0
avg_loss = 0
count = 0
for cluster_path in cluster_path_list:
path_idx += 1
dataloaders = []
for cluster_id in cluster_path:
dataloaders.append(loader_list[cluster_id])
for batch_idx, data in enumerate(zip(*dataloaders)):
cluster_num = len(data)
data_zip = torch.cat(data, 0).to(device)
# Generating Score
scores = model(data_zip).squeeze()
scores = torch.clamp(scores, min=0, max=1)
score_list = torch.split(scores, args.batch_sz, dim = 0)
# Standard deviation as a loss
loss_var = torch.zeros(1).to(device)
for score in score_list:
loss_var += score.var()
loss_var /= len(score_list)
# Differentiable Ranking with sigmoid function
rank_matrix = torch.zeros((args.batch_sz, cluster_num, cluster_num)).to(device)
for itertuple in list(permutations(range(cluster_num), 2)):
score1 = score_list[itertuple[0]]
score2 = score_list[itertuple[1]]
diff = args.lamb * (score2 - score1)
results = torch.sigmoid(diff)
rank_matrix[:, itertuple[0], itertuple[1]] = results
rank_matrix[:, itertuple[1], itertuple[0]] = 1 - results
rank_predicts = rank_matrix.sum(1)
temp = torch.Tensor(range(cluster_num))
target_rank = temp.unsqueeze(0).repeat(args.batch_sz, 1).to(device)
# Equivalent to spearman rank correlation loss
loss_train = ((rank_predicts - target_rank)**2).mean()
loss = loss_train + loss_var * args.alpha
optimizer.zero_grad()
loss.backward()
optimizer.step()
train_loss.update(loss_train.item(), args.batch_sz)
reg_loss.update(loss_var.item(), args.batch_sz)
avg_loss += loss.item()
count += 1
# Print status
if batch_idx % 10 == 0:
print('Epoch: [{epoch}][{path_idx}][{elps_iters}] '
'Train loss: {train_loss.val:.4f} ({train_loss.avg:.4f}) '
'Reg loss: {reg_loss.val:.4f} ({reg_loss.avg:.4f})'.format(
epoch=epoch, path_idx=path_idx, elps_iters=batch_idx, train_loss=train_loss, reg_loss=reg_loss))
return avg_loss / count
def main(args):
torch.manual_seed(args.seed)
torch.cuda.manual_seed_all(args.seed)
np.random.seed(args.seed)
# Input example
cluster_number = args.cluster_num
# Graph generation mode
if args.graph_config:
graph_config = args.graph_config
elif args.mode == "census":
df = pd.read_csv(args.census_path)
hist = pd.read_csv(os.path.join('./data', args.dir_name, args.histogram_path), header = None)
graph_config = graph_inference_census(df, hist, cluster_number, args.graph_name)
elif args.mode == "nightlight":
grid_df = pd.read_csv(os.path.join('./data', args.dir_name, args.grid_path))
nightlight_df = pd.read_csv(args.nightlight_path)
graph_config = graph_inference_nightlight(grid_df, nightlight_df, cluster_number, args.graph_name)
else:
raise ValueError
# Dataloader definition
start, end, partial_order, cluster_unify = graph_process(graph_config)
loader_list = generate_loader_dict(range(cluster_number), cluster_unify, args.batch_sz)
cluster_graph = generate_graph(partial_order, cluster_number)
cluster_path_list = cluster_graph.printPaths(start, end)
print("Cluster_path: ", cluster_path_list)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = models.resnet18(pretrained=False)
model.fc = nn.Sequential()
if torch.cuda.device_count() > 1:
model = nn.DataParallel(model)
cudnn.benchmark = True
model.load_state_dict(torch.load(args.pretrained_path)['state_dict'], strict = False)
model.module.fc = nn.Sequential(nn.Linear(512, 1))
model.to(device)
optimizer = torch.optim.Adam(model.parameters(), lr = args.lr)
print("Pretrained net load finished")
best_loss = float('inf')
if args.load == False:
for epoch in range(args.epochs):
loss = train(args, epoch, model, optimizer, loader_list, cluster_path_list, device)
if epoch % 10 == 0 and epoch != 0:
if best_loss > loss:
print("state saving...")
state = {
'model': model.state_dict(),
'loss': loss
}
if not os.path.isdir('checkpoint'):
os.mkdir('checkpoint')
torch.save(state, './checkpoint/{}'.format(args.name))
best_loss = loss
print("best loss: %.4f\n" % (best_loss))
if __name__ == "__main__":
args = siScore_parser()
main(args)