-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
133 lines (108 loc) · 4.92 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
import random
import collections
import numpy as np
import skimage
import torch
from torchvision.datasets import CIFAR10, MNIST
from PIL import Image
from model import *
from train import *
from downstream import *
random.seed(42)
torch.manual_seed(42)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
cifar10 = CIFAR10(root='.', train=True, download=True, transform=transforms.ToTensor())
cifar10 = Subset(cifar10, list(range(0, len(cifar10), 5))) # Select only 10000 units
cifar10_train, cifar10_test, cifar10_val = random_split(cifar10, [7144, 1428, 1428])
mnist = MNIST(root='.', train=True, download=True, transform=transforms.ToTensor())
mnist_in = Subset(mnist, list(range(0, len(mnist), 2)))
mnist_out = Subset(mnist, list(range(1, len(mnist), 2))) # Test the generalization capacity
mnist_train, mnist_test, mnist_val = random_split(mnist_in, [21428, 4286, 4286])
"""
Configuration 1 MNIST
"""
torch.manual_seed(42)
md = LatentModulatedSiren(hidden_width=256, hidden_depth=5, out_dim=1, latent_dim=256).to(device)
for siren_layer in md.siren_layers:
siren_layer.to(device)
meta_learning(md, mnist_in, 1000, 128, 3, 100, outer_lr=3e-5, image_size=(28, 28), image_channel=1, save_path='mnist_functa_1.pt') # 230K updates
"""
Configuration 2 MNIST
"""
torch.manual_seed(42)
md = LatentModulatedSiren(hidden_width=512, hidden_depth=15, out_dim=1, latent_dim=512).to(device)
for siren_layer in md.siren_layers:
siren_layer.to(device)
meta_sgd_learning(md, mnist_in, 250, 32, 3, 100, outer_lr=3e-6, image_size=(28, 28), image_channel=1, save_path='mnist_functa_2.pt')
"""
Configuration 1 CIFAR10
"""
torch.manual_seed(42)
md = LatentModulatedSiren(hidden_width=256, hidden_depth=5, out_dim=3, latent_dim=256).to(device)
for siren_layer in md.siren_layers:
siren_layer.to(device)
meta_learning(md, cifar10, 4000, 128, 3, 100, outer_lr=3e-5, save_path='cifar_functa_1.pt') # 312K updates
"""
Configuration 3 CIFAR10
"""
torch.manual_seed(42)
md = SpatialFuncta(hidden_width=256, hidden_depth=5, out_dim=3, latent_size=8, latent_channel=16).to(device)
for siren_layer in md.siren_layers:
siren_layer.to(device)
meta_sgd_learning(md, cifar10, 2000, 64, 3, 100, outer_lr=3e-5, save_path='cifar_functa_2.pt') # 312K updates
"""
Downstream 1 MNIST
"""
model = torch.load("mnist_functa_1.pt").to(device)
for siren_layer in model.siren_layers:
siren_layer.to(device)
mnist_testset = build_functaset(mnist_test, model)
test_loader = DataLoader(mnist_testset, batch_size=256, num_workers=12)
mnist_trainset = build_functaset(mnist_train, model)
train_loader = DataLoader(mnist_trainset, batch_size=256, shuffle=True, num_workers=12)
mnist_valset = build_functaset(mnist_val, model)
val_loader = DataLoader(mnist_valset, batch_size=256, num_workers=12)
classifier = Classifier(256, 10).to(device)
train_classification(train_loader, test_loader, val_loader, classifier)
"""
Downstream 2 MNIST
"""
model = torch.load("mnist_functa_2.pt").to(device)
for siren_layer in model.siren_layers:
siren_layer.to(device)
mnist_testset = build_functaset(mnist_test, model)
test_loader = DataLoader(mnist_testset, batch_size=256, num_workers=12)
mnist_trainset = build_functaset(mnist_train, model)
train_loader = DataLoader(mnist_trainset, batch_size=256, shuffle=True, num_workers=12)
mnist_valset = build_functaset(mnist_val, model)
val_loader = DataLoader(mnist_valset, batch_size=256, num_workers=12)
classifier = Classifier(512, 10).to(device)
train_classification(train_loader, test_loader, val_loader, classifier)
"""
Downstream 1 CIFAR10
"""
model = torch.load("cifar_functa_1.pt").to(device)
for siren_layer in model.siren_layers:
siren_layer.to(device)
cifar10_testset = build_functaset(cifar10_test, model)
test_loader = DataLoader(cifar10_testset, batch_size=256, num_workers=12)
cifar10_trainset = build_functaset(cifar10_train, model)
train_loader = DataLoader(cifar10_trainset, batch_size=256, shuffle=True, num_workers=12)
cifar10_valset = build_functaset(cifar10_val, model)
val_loader = DataLoader(cifar10_valset, batch_size=256, num_workers=12)
classifier = Classifier(256, 10).to(device)
train_classification(train_loader, test_loader, val_loader, classifier)
"""
Downstream 2 CIFAR10
"""
model = torch.load("cifar_functa_2.pt").to(device)
for siren_layer in model.siren_layers:
siren_layer.to(device)
cifar10_testset = build_functaset(cifar10_test, model)
test_loader = DataLoader(cifar10_testset, batch_size=256, num_workers=12)
cifar10_trainset = build_functaset(cifar10_train, model)
train_loader = DataLoader(cifar10_trainset, batch_size=256, shuffle=True, num_workers=12)
cifar10_valset = build_functaset(cifar10_val, model)
val_loader = DataLoader(cifar10_valset, batch_size=256, num_workers=12)
classifier = ConvClassifier(8, 16, 10).to(device)
train_classification(train_loader, test_loader, val_loader, classifier)