-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy patheval_CIFAR10C.py
147 lines (114 loc) · 4.62 KB
/
eval_CIFAR10C.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
import warnings
warnings.filterwarnings('ignore')
import torch
import numpy as np
from models import wrn
from laplace import kfla
import laplace.util as lutil
import util.evaluation as evalutil
import util.dataloaders as dl
import util.misc
from math import *
from tqdm import tqdm, trange
import argparse
import os, sys
from tqdm import tqdm, trange
from collections import defaultdict
import reluq
parser = argparse.ArgumentParser()
parser.add_argument('--ood_dset', default='imagenet', choices=['imagenet', 'uniform', 'smooth'])
args = parser.parse_args()
torch.manual_seed(9999)
np.random.seed(9999)
path = f'./pretrained_models'
train_loader = dl.CIFAR10(train=True, augm_flag=False)
val_loader, test_loader = dl.CIFAR10(train=False, val_size=2000)
print(len(train_loader.dataset), len(val_loader.dataset), len(test_loader.dataset))
num_classes = 10
data_shape = [3, 32, 32]
method_types = ['MAP', 'DE', 'LA', 'LULA']
method_strs = ['MAP', 'DE', 'LA', 'LA-LULA']
distortion_types = dl.CorruptedCIFAR10Dataset.distortions
severity_levels = range(1, 6) # 1 ... 5
tab_acc = defaultdict(lambda: defaultdict(lambda: defaultdict(list)))
tab_mmc = defaultdict(lambda: defaultdict(lambda: defaultdict(list)))
tab_ece = defaultdict(lambda: defaultdict(lambda: defaultdict(list)))
tab_brier = defaultdict(lambda: defaultdict(lambda: defaultdict(list)))
tab_loglik = defaultdict(lambda: defaultdict(lambda: defaultdict(list)))
def load_model(type='MAP'):
def create_model():
return wrn.WideResNet(16, 4, num_classes).cuda()
if type == 'DE':
K = 5
model = [create_model() for _ in range(K)]
state_dicts = torch.load(f'./pretrained_models/CIFAR10_wrn_de.pt')
for k in range(K):
model[k].load_state_dict(state_dicts[k])
model[k].eval()
else:
model = create_model()
model.load_state_dict(torch.load(f'./pretrained_models/CIFAR10_wrn_plain.pt'))
model.eval()
# Additionally, load these for LULA
if type == 'LULA':
lula_params = torch.load(f'./pretrained_models/kfla/CIFAR10_wrn_lula_{args.ood_dset}.pt')
if args.ood_dset == 'best':
state_dict, n_units, noise = lula_params
print(f'LULA uses this OOD dataset: {noise}')
else:
state_dict, n_units = lula_params
model = lula.model.LULAModel_LastLayer(model, n_units).cuda()
model.to_gpu()
model.load_state_dict(state_dict)
model.disable_grad_mask()
model.unmask()
model.eval()
if type in ['LA', 'LULA']:
var0 = torch.tensor(1/(5e-4*len(train_loader.dataset))).float().cuda()
model = kfla.KFLA(model)
model.get_hessian(train_loader)
model.estimate_variance(var0)
return model
def predict_(test_loader, model, model_name, params=None):
assert model_name in method_types
if model_name in ['LA', 'LULA']:
py = lutil.predict(test_loader, model, n_samples=20)
elif model_name == 'DE':
py = evalutil.predict_ensemble(test_loader, model)
else: # MAP
py = evalutil.predict(test_loader, model)
return py.cpu().numpy()
def evaluate(model_name):
assert model_name in method_types
model = load_model(model_name)
params = None
if model_name == 'LULA':
model_str = 'LA-LULA'
else:
model_str = model_name
print(f'Processing for {model_str}')
# For all distortions, for all severity
for d in tqdm(distortion_types, leave=False):
for s in tqdm(severity_levels, leave=False):
shift_loader = dl.CorruptedCIFAR10(d, s)
py_shift = predict_(shift_loader, model, model_name, params=params)
targets = torch.cat([y for x, y in shift_loader], dim=0).numpy()
tab_acc[model_str][d][str(s)].append(evalutil.get_acc(py_shift, targets))
tab_mmc[model_str][d][str(s)].append(evalutil.get_mmc(py_shift))
tab_ece[model_str][d][str(s)].append(evalutil.get_calib(py_shift, targets)[0])
tab_brier[model_str][d][str(s)].append(evalutil.get_brier(py_shift, targets))
tab_loglik[model_str][d][str(s)].append(evalutil.get_loglik(py_shift, targets))
evaluate('MAP')
evaluate('DE')
evaluate('LA')
evaluate('LULA')
# Save results
dir_name = f'results/CIFAR10C/'
dir_name += f'{args.ood_dset}'
if not os.path.exists(dir_name):
os.makedirs(dir_name)
np.save(f'{dir_name}/mmcs', util.misc.ddict2dict(tab_mmc))
np.save(f'{dir_name}/accs', util.misc.ddict2dict(tab_acc))
np.save(f'{dir_name}/eces', util.misc.ddict2dict(tab_ece))
np.save(f'{dir_name}/briers', util.misc.ddict2dict(tab_brier))
np.save(f'{dir_name}/logliks', util.misc.ddict2dict(tab_loglik))