FedSSL/options.py at master · davidgill97/FedSSL · 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
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
#!/usr/bin/env python
# -*- coding: utf-8 -*-
# Python version: 3.6

import argparse
import os
import time, datetime

def str2bool(v):
    #https://eehoeskrap.tistory.com/521
    if isinstance(v, bool):
        return v
    if v.lower() in ('yes', 'true', 't', 'y', '1'):
        return True
    elif v.lower() in ('no', 'false', 'f', 'n', '0'):
        return False
    else:
        raise argparse.ArgumentTypeError('Boolean value expected.')

def getTimestamp():
    utc_timestamp = int(time.time())
    date = datetime.datetime.fromtimestamp(utc_timestamp).strftime("%Y_%m_%d_%H_%M_%S")
    return date

def args_parser():
    parser = argparse.ArgumentParser()

    # model arguments
    parser.add_argument('--model',            type=str,       default='resnet18',  help='resnet18|resnet50')
    parser.add_argument('--pretrained',       type=str2bool,  default=False,       help='pretrained backbone')
    parser.add_argument('--num_classes',      type=int,       default=10,          help="number of classes")
    parser.add_argument('--group_norm',       type=str2bool,  default=False,       help="group normalization")

    # Experimental setup
    parser.add_argument("--exp",              type=str,       default="FLSL",      choices=["simclr", "simsiam", "centralized", "FLSL"])
    parser.add_argument("--wandb_tag",        type=str,       default="",        help="optional tag for wandb logging")
    parser.add_argument("--alpha",            type=float,     default=0.5,       help="dirichlet param 0<alpha<infty controls iidness alpha=0:non iid")

    # SimCLR
    parser.add_argument("--temperature",      type=float,     default=0.1,       help="softmax temperature")
    parser.add_argument('--n_views',          type=int,       default=2,         help="default simclr n_views=2")

    # SimSiam
    parser.add_argument("--pred_dim",         type=int,       default=256,       help="pred dim for simsiam")

    # SimCLR & SimSiam
    parser.add_argument('--strength',         type=float,     default=0.5,       help="augmentation strength 0<s<1")
    parser.add_argument('--target_size',      type=int,       default=32,        help="augmentation target width (=height)")
    parser.add_argument('--freeze',           type=str2bool,  default=False,      help='freeze feature extractor during linear eval')
    parser.add_argument('--out_dim',          type=int,       default=512,       help="output dimension of the feature for simclr and simsiam")

    # FedProx
    parser.add_argument('--mu',             type=float,     default=0.01)

    # FL
    parser.add_argument("--num_users",      type=int,      default=10,        help="num users")
    parser.add_argument("--num_items",      type=int,      default=32,        help="num data each client holds")
    parser.add_argument("--iid",            type=str2bool, default=True,      help="iid on clients")
    parser.add_argument('--epochs',         type=int,      default=200,        help="number of rounds of training") # FedMatch
    parser.add_argument('--frac',           type=float,    default=1,       help='the fraction of clients: C')
    parser.add_argument('--local_ep',       type=int,      default=10,         help="the number of local epochs: E")
    parser.add_argument('--local_bs',       type=int,      default=32,         help="local batch size")
    parser.add_argument('--lr',             type=float,    default=0.001,      help='learning rate')
    parser.add_argument('--momentum',       type=float,    default=0.9,        help='SGD momentum (default: 0.5)')
    parser.add_argument('--weight_decay',   type=float,    default=1e-4,       help='weight decay (default: 1e-4)')
    parser.add_argument('--dataset',        type=str,       default='cifar',  choices=["mnist", "cifar"],              help="mnist|cifar")
    parser.add_argument('--optimizer',      type=str,       default='adam',               help="type of optimizer")
    parser.add_argument('--server_num_items', type=int,  default=1000,  help="number of items per class used for training at server")
    parser.add_argument('--server_batch_size', type=int, default=32, help="server batch size for iid pre-training at server")
    parser.add_argument('--bn_stat_momentum', type=float, default=0.01, help="bn stat EMA momentum should be set < 0.1")

    # Train setting
    parser.add_argument("--parallel",       type=str2bool,  default=True,                help="parallel training with threads")
    parser.add_argument("--num_workers",    type=int,       default=8,                    help="num workers for dataloader")
    parser.add_argument('--log_path',       type=str,       default='./logs',             help="tensorboard log dir")
    parser.add_argument('--seed',           type=int,       default=2022,                 help='random seed')
    parser.add_argument('--ckpt_path',      type=str,       default="./checkpoints/checkpoint.pth.tar", help="model ckpt save path")
    parser.add_argument('--data_path',      type=str,       default="./data",             help="path to dataset")
    parser.add_argument('--train_device',   type=str,       default="0",                  help="gpu device number for train")
    parser.add_argument('--finetune_epoch', type=int,       default=1,                   help='finetune epochs at server')
    parser.add_argument('--finetune',       type=str2bool,  default=True,                 help="finetune at the server")
    parser.add_argument("--ckpt_criterion",  type=str,     default="loss", help="ckpt criterion loss|top1")
    parser.add_argument("--agg",             type=str,     default="fedavg", choices=["fedavg", "fedprox", "fedmatch"])

    args = parser.parse_args()

    # if no experiment is specified on path
    if not args.log_path.split("logs")[-1]:
        args.log_path = os.path.join(args.log_path, getTimestamp())
    if not args.finetune:
        args.finetune_epoch = 1


    if args.exp == "FLSL":
        get_FLSL_opts(args, args.iid)

    elif args.exp == "centralized":
        get_centralized_opts(args, args.iid)

    elif args.exp == "simclr":
        get_simclr_opts(args, args.iid)

    elif args.exp == "simsiam":
        get_simsiam_opts(args, args.iid)

    return args

def get_centralized_opts(args, iid = False):
    if iid == True:
        args.alpha = 100000 # arbitrary large number for iid

    else:
        args.alpha = 0.5    # Non-i.i.d.

    args.num_users = 100
    args.num_items = 300
    args.epochs  = 100
    args.frac = 0.1
    args.local_ep = 10
    args.local_bs = 16
    args.finetune = False
    args.freeze = False

def get_FLSL_opts(args, iid = False):
    if iid == True:
        args.alpha = 100000

    else:
        args.alpha = 0.1

    args.num_users = 100
    args.num_items = 300
    args.epochs = 100
    args.frac = 0.1
    args.local_ep = 5
    args.local_bs = 16
    # aggregate only
    args.finetune = False


def get_simclr_opts(args, iid = False):
    if iid == True:
        args.alpha = 100000
    else:
        args.alpha = 0.5

    args.num_users = 100
    args.num_items = 300
    args.epochs = 100
    args.frac = 0.1
    args.local_ep = 10
    args.local_bs = 16
    # aggregate and train linear at server (finetune)
    args.freeze
    args.finetune = True
    # freeze backbone
    args.freeze = True

def get_simsiam_opts(args, iid = False):
    if iid == True:
        args.alpha = 100000
    else:
        args.alpha = 0.5

    args.num_users = 100
    args.num_items = 300
    args.epochs = 100
    args.frac = 0.1
    args.local_ep = 10
    args.local_bs = 16
    args.finetune = True
    # freeze backbone
    args.freeze = True