forked from duanyiyao/E-TS-in-VFL
-
Notifications
You must be signed in to change notification settings - Fork 0
/
model.py
136 lines (119 loc) · 5.32 KB
/
model.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
import torch
from torchvision import datasets, transforms
from torch import nn, optim
import numpy as np
class SplitNN(torch.nn.Module):
def __init__(self, models, data_owner, label_owner, server,device,model_locations):
super().__init__()
self.label_owner = label_owner
self.data_owners = data_owner
self.models = models
self.server = server
self.device = device
self.optimizers = {location: optim.Adam(self.models[location].parameters(), lr=0.001) for location in model_locations}
def activate_model(self, device):
self.to(device)
for owner, model in self.models.items():
self.models[owner] = model.to(device)
self.device = device
def forward(self, data_pointer):
embedding = {}
remote_outputs = []
#iterate over each client and pass thier inputs to respective model segment and move outputs to server
for owner in self.data_owners:
embedding[owner] = self.models[owner](data_pointer[owner].reshape([-1, 28*4]))
embedding[owner] = embedding[owner].view(-1,64,7,1)
remote_outputs.append(embedding[owner].requires_grad_())
remote_outputs = torch.cat(remote_outputs, -1)
server_output = self.models["server"](remote_outputs.view(-1, 64, 7, 7))
server_output = server_output.view(-1, 512)
# move to label_owner
server_output = server_output.requires_grad_()
pred = self.models["label_owner"](server_output)
return pred
def forward_client(self, data_pointer):
embedding = {}
for owner in self.data_owners:
embedding[owner] = self.models[owner](data_pointer[owner].reshape([-1, 28*4]))
embedding[owner] = embedding[owner].view(-1,64,7,1)
return embedding
def forward_server(self, input):
server_input = []
for client in self.data_owners:
server_input.append(input[client])
server_inputs = torch.cat(server_input, -1)
server_output = self.models["server"](server_inputs.view(-1, 64, 7, 7))
server_output = server_output.view(-1, 512)
server_output = server_output.requires_grad_()
pred = self.models["label_owner"](server_output)
return pred
def zero_grads(self):
for opt in self.optimizers.values():
opt.zero_grad()
def steps(self):
for opt in self.optimizers.values():
opt.step()
def train(self):
for model in self.models.values():
model.train()
def eval(self):
for model in self.models.values():
model.eval()
def models_generate(data_owners, input_size, hidden_size):
models = {}
for i in range(len(data_owners)):
models[data_owners[i]] = nn.Sequential(
nn.Linear(input_size, hidden_size),
nn.BatchNorm1d(hidden_size),
nn.ReLU(),
)
models["server"] = nn.Sequential(
# 3
nn.Conv2d(64, 128, kernel_size=2, padding=1),
nn.BatchNorm2d(128),
nn.ReLU(True),
# 4
nn.Conv2d(128, 128, kernel_size=3, padding=1),
nn.BatchNorm2d(128),
nn.ReLU(True),
nn.MaxPool2d(kernel_size=2, stride=2),
# 5
nn.Conv2d(128, 256, kernel_size=3, padding=1),
nn.BatchNorm2d(256),
nn.ReLU(True),
# 6
nn.Conv2d(256, 256, kernel_size=3, padding=1),
nn.BatchNorm2d(256),
nn.ReLU(True),
# 7
nn.Conv2d(256, 256, kernel_size=3, padding=1),
nn.BatchNorm2d(256),
nn.ReLU(True),
nn.MaxPool2d(kernel_size=2, stride=2),
# 8
nn.Conv2d(256, 512, kernel_size=3, padding=1),
nn.BatchNorm2d(512),
nn.ReLU(True),
# 9
nn.Conv2d(512, 512, kernel_size=3, padding=1),
nn.BatchNorm2d(512),
nn.ReLU(True),
# 10
nn.Conv2d(512, 512, kernel_size=3, padding=1),
nn.BatchNorm2d(512),
nn.ReLU(True),
nn.MaxPool2d(kernel_size=2, stride=2),
)
models["label_owner"] = nn.Sequential(
# 11
nn.Linear(512, 512),
nn.ReLU(True),
nn.Dropout(0.5),
# 12
nn.Linear(512, 512),
nn.ReLU(True),
nn.Dropout(0.5),
# 13
nn.Linear(512, 10),
)
return models