/
eval.py
130 lines (112 loc) · 3.96 KB
/
eval.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
import os
import logging
import argparse
import numpy as np
import torch
import torch.nn as nn
from utils import (
get_model,
get_metadata,
fix_legacy_dict,
evalfxn,
)
from data import get_real_dataloaders
def main():
parser = argparse.ArgumentParser("Evalution script")
# common args
parser.add_argument("--exp-name", type=str, default="temp")
parser.add_argument("--checkpoint-path", type=str)
parser.add_argument(
"--results-dir", type=str, default="./trained_models/eval_logs/"
)
parser.add_argument(
"--val-method",
type=str,
default="pgd",
choices=("baseline", "pgd", "auto"),
)
# data
parser.add_argument(
"--dataset",
type=str,
choices=("cifar10", "cifar100", "imagnet64", "celebA", "afhq"),
)
parser.add_argument("--data-dir", type=str, help="dir where data is stored")
parser.add_argument("--batch-size", type=int, default=128)
parser.add_argument("--workers", type=int, default=8)
# model
parser.add_argument("--arch", type=str)
# adversarial attack
parser.add_argument("--attack", type=str, choices=("linf", "l2"), default="linf")
parser.add_argument("--epsilon", type=float, default=8.0 / 255)
parser.add_argument("--step-size", type=float, default=2.0 / 255)
parser.add_argument("--num-steps", type=int, default=10)
parser.add_argument("--clip-min", type=float, default=0.0)
parser.add_argument("--clip-max", type=float, default=1.0)
parser.add_argument(
"--autoattack-attack-subset",
nargs="+",
help="subset of attacks from autoattack default set of attacks (default uses all attacks)",
)
# misc
parser.add_argument("--trial", type=int, default=0)
parser.add_argument("--print-freq", type=int, default=50)
parser.add_argument("--seed", type=int, default=12345)
######################### Basic setup #########################
args = parser.parse_args()
args.metadata = get_metadata(args.dataset)
args.gpu, args.rank = "cuda:0", 0 # now its compatible with training utils
print(args)
# set up cuda + seeds
torch.backends.cudnn.benchmark = True # a few percentage speedup
torch.manual_seed(args.seed)
np.random.seed(args.seed)
assert (
torch.cuda.is_available()
), "we assume cuda is available, too slow to eval on cpus."
assert args.checkpoint_path, "Need a checkpoint to evaluate"
# recursively create all directories needed to log results
args.log_dir = os.path.join(
os.path.join(args.results_dir, args.exp_name), f"trial_{args.trial}"
)
os.makedirs(args.log_dir, exist_ok=True)
# logger
logging.basicConfig(level=logging.INFO, format="%(message)s")
logger = logging.getLogger()
logger.addHandler(
logging.FileHandler(os.path.join(args.log_dir, "eval_log.txt"), "a")
)
logger.info(args)
model = get_model(args.arch, args.metadata.num_classes).cuda()
# load checkpoint
checkpoint = torch.load(args.checkpoint_path, map_location="cpu")
d = fix_legacy_dict(checkpoint["state_dict"])
logger.info(f"Loaded state dict from {args.checkpoint_path}")
model.load_state_dict(d, strict=True)
logger.info(f"Mismatched keys {set(d.keys()) ^ set(model.state_dict().keys())}")
logger.info(f"Checkpoint loaded from {args.checkpoint_path}")
model = torch.nn.parallel.DataParallel(model).eval()
criterion = nn.CrossEntropyLoss().cuda()
_, val_loader, _, _ = get_real_dataloaders(
args.dataset,
args.data_dir,
args.batch_size,
args.workers,
args.metadata,
distributed=False,
)
results_val = evalfxn(
args.val_method,
model,
val_loader,
criterion,
args,
args.metadata.num_classes,
)
logger.info(
", ".join(
["{}: {:.3f}".format(k + "_val", v) for (k, v) in results_val.items()]
)
)
if __name__ == "__main__":
main()