Conformal-Prediction-Unlearning/main_unlearn_cpu.py at main · TIML-Group/Conformal-Prediction-Unlearning · GitHub

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
import utils
import torch
import time
import os
import argparse
import unlearn_cpu
import numpy as np
import xlwt
from metrics import MIACR, CR
import wandb
from torch.utils.data import ConcatDataset
from models import resnet, vit
from torch.utils.data import DataLoader, random_split
import torch.nn.functional as F

def get_save_dir():
    if args.unlearn_type == 'random':
        save_dir = os.path.join(args.model_name +'_' + args.data_name,
                            args.unlearn_name +'_nips_forget' + str(int(100-args.retain_ratio*100)) +'_epoch' + str(args.num_epochs))
    elif args.unlearn_type == 'class':
        save_dir = os.path.join(args.model_name + '_' + args.data_name,
                                args.unlearn_name + '_nips_forget_class_epoch' + str(args.num_epochs))
    return save_dir


if __name__ == '__main__':
    flag = time.time()
    parser = argparse.ArgumentParser()
    parser.add_argument(
        '--unlearn_name',
        type=str,
        default='teacher')
    parser.add_argument(
        '--unlearn_type',
        type=str,
        default='random',
        choices=[
            "random",
            "class",
        ])
    parser.add_argument('--model_name', type=str, default='resnet18')
    parser.add_argument('--data_name', type=str, default='cifar10')
    parser.add_argument('--data_dir', type=str, default='../data')
    parser.add_argument('--model_dir', type=str, default='./resnet18-cifar10/fine_model_baseline/final_model.pth')
    parser.add_argument('--batch_size', type=int, default=128)
    parser.add_argument('--num_epochs', type=int, default=20)
    parser.add_argument('--num_classes', type=int, default=10)
    parser.add_argument('--retain_ratio', type=float, default=0.9)
    parser.add_argument('--learning_rate', type=float, default=0.1)
    parser.add_argument('--milestones', type=str, default=None)  # [82,122,163] for retrain 200 epochs None
    parser.add_argument('--seed', type=int, default=1)
    # evaluation
    parser.add_argument('--cal_sizes', type=str, default='1000')
    parser.add_argument('--alphas', type=str, default='0.05')

    parser.add_argument('--delta', type=float, default=0.01)
    parser.add_argument('--alpha', type=float, default=0.05)
    parser.add_argument('--lamda', type=float, default=0.8)

    parser.add_argument('--worst_case', type=bool, default=None, help="worst_case")

    parser.add_argument('--mask_path', type=str, default=None, help="salun")

    args = parser.parse_args()

    utils.setup_seed(args.seed)
    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
    save_dir = get_save_dir()

    if not os.path.exists(save_dir):
        os.makedirs(save_dir)
    train_retain_dl, train_forget_dl, test_retain_dl, test_forget_dl, retain_cal_loader, forget_cal_loader, forget_class = utils.generate_dataset(args.unlearn_type,
                                                                                              args.num_classes,
                                                                                              args.batch_size,
                                                                                              args.data_dir,
                                                                                              args.data_name,
                                                                                              args.model_name,
                                                                                              args.retain_ratio,
                                                                                              worst_case=args.worst_case)
    full_train_dl = DataLoader(
        ConcatDataset((train_retain_dl.dataset, train_forget_dl.dataset)),
        batch_size=args.batch_size,
    )
    cal_train_size = 1000 if args.unlearn_type == 'random' else 200
    cal_ds, test_ds = random_split(test_retain_dl.dataset, [cal_train_size, len(test_retain_dl.dataset)-cal_train_size])


    cal_dl = DataLoader(dataset=cal_ds, batch_size=args.batch_size, shuffle=False, pin_memory=True)
    test_retain_dl_ori = test_retain_dl
    test_retain_dl = DataLoader(dataset=test_ds, batch_size=args.batch_size, shuffle=False, pin_memory=True)

    if args.model_name == 'resnet18':
        net = resnet.ResNet18(num_classes=args.num_classes)
        unlearning_teacher = resnet.ResNet18(num_classes=args.num_classes).to(device) if args.unlearn_name == 'teacher' else None
    elif args.model_name == 'vit':
        if args.unlearn_name == 'retrain':
            net = vit.ViT(num_classes=args.num_classes, pretrained=True)
        else:
            net = vit.ViT(num_classes=args.num_classes, pretrained=False)
        unlearning_teacher = vit.ViT(num_classes=args.num_classes, pretrained=True).to(device) if args.unlearn_name == 'teacher' else None

    if args.unlearn_name != 'retrain':
        net.load_state_dict(torch.load(args.model_dir))

    if torch.cuda.is_available() and torch.cuda.device_count()>1:
        print("Using {} GPUs.".format(torch.cuda.device_count()))
        net = torch.nn.DataParallel(net)

    net = net.to(device)

    # wandb
    proj_name = '{}_{}_{}_epoch{}_loss_new'.format(args.model_name, args.data_name, args.unlearn_name, args.num_epochs)
    watermark = "{}_lr{}_delta{}_alpha{}_lamda{}".format(args.model_name, args.learning_rate, args.delta, args.alpha, args.lamda)
    wandb.init(project=proj_name, name=watermark)
    wandb.config.update(args)
    wandb.watch(net)

    mask = None
    if args.mask_path is not None and args.unlearn_name=='salun':
        mask = torch.load(args.mask_path)

    kwargs = {
        "model": net,
        "train_retain_dl": train_retain_dl,
        "train_forget_dl": train_forget_dl,
        "test_retain_dl": test_retain_dl,
        "test_forget_dl": test_forget_dl,
        "cal_dl": cal_dl,
        "dampening_constant": 1,   # Lambda for ssd
        "selection_weighting": 5 if args.model_name == 'vit' and args.data_name=='cifar100' else 10,   # Alpha for ssd
        "num_classes": args.num_classes,
        "dataset_name": 'cifar10',
        "device": device,
        "model_name": args.model_name,
        "num_epochs": args.num_epochs,
        "learning_rate": args.learning_rate,
        "milestones": np.array(args.milestones.split(',')).astype(int) if args.milestones is not None else None,
        "batch_size": args.batch_size,
        "full_train_dl": full_train_dl,
        "unlearning_teacher": unlearning_teacher,
        "delta": args.delta,
        "alpha": args.alpha,
        "lamda": args.lamda,
        "unlearn_type": args.unlearn_type,
        "unlearn_name": args.unlearn_name,
        "mask": mask,
        "args": args,
    }

    start = time.time()
    acc_test = getattr(unlearn_cpu, args.unlearn_name)(
        **kwargs
    )
    wandb.save("{}_{}_{}_{}_{}_loss.h5".format(args.model_name, args.data_name, args.unlearn_name, args.num_epochs, args.learning_rate))
    torch.save(net.state_dict(), os.path.join(save_dir, f'final_model_delta{args.delta}_alpha{args.alpha}_lamda{args.lamda}_seed{args.seed}.pth'))