-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathCL_ViT_MLP_vs_KAN.py
159 lines (106 loc) · 5.66 KB
/
CL_ViT_MLP_vs_KAN.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
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torchvision import datasets, transforms
from torch.utils.data import DataLoader
# from torchsummary import summary
import torch
from efficientkan import KAN as efficientKAN
from models.vision_transformer import VisionTransformer, PatchEmbed
from continual_learning_trainer import ContinualLearningTrainer
from utils import DivideDataset
if __name__ == '__main__':
## device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
## batch size
batch_size = 16
## code_version_flag
isMNIST = True
## dataset transform
transform = transforms.Compose(
[transforms.Resize(224),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
num_of_task = 10
## dataset and dataloader
train_dataset = datasets.CIFAR100(root='./data', train=True, download=True, transform=transform)
test_dataset = datasets.CIFAR100(root='./data', train=False, transform=transform)
train_CIFAR_task_divider = DivideDataset(train_dataset, num_of_task, list(range(100)))
train_CIFAR_task_datasets, tasks_classes = train_CIFAR_task_divider.get_the_datasets()
train_task_dataset_loader = {}
for task in range(num_of_task):
train_task_dataset_loader[task] = DataLoader(train_CIFAR_task_datasets[task], batch_size=batch_size, shuffle=True)
test_Cifar_task_divider = DivideDataset(test_dataset, num_of_task, list(range(100)))
test_CIFAR_task_datasets, tasks_classes = test_Cifar_task_divider.get_the_datasets()
test_task_dataset_loader = {}
for task in range(num_of_task):
test_task_dataset_loader[task] = DataLoader(test_CIFAR_task_datasets[task], batch_size=batch_size, shuffle=False)
## replay dataset and loader
replay_dataset = {}
no_of_replay_samples = 2000
for current_task in range(1, num_of_task):
no_of_sample_per_task = no_of_replay_samples//current_task
for task in range(current_task):
## get random no_of_sample_per_task samples from task dataset
indices = torch.randperm(len(train_CIFAR_task_datasets[task])).tolist()[:no_of_sample_per_task]
replay_samples = torch.utils.data.Subset(train_CIFAR_task_datasets[task], indices)
if current_task not in replay_dataset:
replay_dataset[current_task] = [replay_samples]
else:
replay_dataset[current_task].append(replay_samples)
## concatenate all replay samples and create a dataset loader
replay_dataset_loader = {}
for task in range(1, num_of_task):
replay_dataset[task] = torch.utils.data.ConcatDataset(replay_dataset[task])
replay_dataset_loader[task] = DataLoader(replay_dataset[task], batch_size=batch_size, shuffle=True)
## define models
Vit_MLP = VisionTransformer(img_size=224, patch_size=16, in_c=3, num_classes=100, embed_dim=256, depth=8, num_heads=8, mlp_ratio=4.0,
drop_path_ratio=0., embed_layer=PatchEmbed, isKAN=False).to(device)
Vit_KAN = VisionTransformer(img_size=224, patch_size=16, in_c=3, num_classes=100, embed_dim=256, depth=8, num_heads=8, mlp_ratio=4.0,
embed_layer=PatchEmbed, isKAN=True).to(device)
## define optimizer
Vit_MLP_optimizer = optim.AdamW(Vit_MLP.parameters(), lr=1e-3)
Vit_KAN_optimizer = optim.AdamW(Vit_KAN.parameters(), lr=1e-3)
## define loss function
criterion = nn.CrossEntropyLoss()
## training schedular
Vit_MLP_schedular = optim.lr_scheduler.StepLR(Vit_MLP_optimizer, step_size=5, gamma=0.9)
Vit_KAN_schedular = optim.lr_scheduler.StepLR(Vit_KAN_optimizer, step_size=5, gamma=0.9)
models = [Vit_MLP, Vit_KAN]
model_names = ['Vit_MLP', 'Vit_KAN']
dataset_name = ['CIFAR100']
train_dataset_loader = [train_task_dataset_loader]
test_dataset_loader = [test_task_dataset_loader]
replay_dataset_loaders = [replay_dataset_loader]
optimizers = [Vit_MLP_optimizer, Vit_KAN_optimizer]
schedulars = [Vit_MLP_schedular, Vit_KAN_schedular]
epoch_ditribution = {}
for task in range(num_of_task):
if task == 0:
epoch_ditribution[task] = 35
else:
epoch_ditribution[task] = 15
file_path = 'saved_models\\CL_ViT_KAN_vs_MLP_with_replay.txt'
file_path_cl_tasks = 'saved_models\\CL_ViT_KAN_vs_MLP_tasks_with_replay.txt'
args_dict = {}
args_dict['num_models'] = len(models)
args_dict['num_datasets'] = 1
for index in range(args_dict['num_datasets']):
args_dict[('dataset_name', index)] = dataset_name[index]
args_dict[('trainloader', index)] = train_dataset_loader[index]
args_dict[('testloader', index)] = test_dataset_loader[index]
args_dict[('replayloader', index)] = replay_dataset_loaders[index]
args_dict['record_save_path'] = file_path
args_dict['record_save_path_cl_tasks'] = file_path_cl_tasks
args_dict['epoch_distribution'] = epoch_ditribution
args_dict['device'] = device
args_dict['loss_function'] = criterion
args_dict['weights_save_path'] = 'saved_models'
args_dict['num_tasks'] = num_of_task
for m in range(args_dict['num_models']):
args_dict[('model', m)] = models[m]
args_dict[('model_name', m)] = model_names[m]
args_dict[('optimizers', m)] = optimizers[m]
args_dict[('schedulers', m)] = schedulars[m]
cl_trainer = ContinualLearningTrainer(args_dict)
cl_trainer.train_models()