-
Notifications
You must be signed in to change notification settings - Fork 7
/
eval_maml_permutation.py
181 lines (154 loc) · 8.81 KB
/
eval_maml_permutation.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
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
import argparse
import os.path as osp
import numpy as np
import torch
import torch.nn.functional as F
from torch.utils.data import DataLoader
from model.dataloader.samplers import CategoriesSampler,RandomSampler
from model.utils import pprint, set_gpu, Averager, Timer, count_acc, euclidean_metric, compute_confidence_interval
from tqdm import tqdm
from model.utils import one_hot
from copy import deepcopy
from itertools import permutations
import pickle
np.random.seed(0)
torch.manual_seed(0)
''' Evaluate MAML-Type Methods for ALL 120 permutations, and some statistics are saved'''
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--model_class', type=str, default='MAML',
choices=['U2S', 'MAML', 'ProtoMAML'])
parser.add_argument('--backbone_class', type=str, default='Res12',
choices=['Res12'])
parser.add_argument('--dataset', type=str, default='MiniImageNet', choices=['MiniImageNet', 'CUB', 'TieredImageNet'])
parser.add_argument('--model_path', type=str, default='./MAML-1-shot.pth')
parser.add_argument('--gpu', default='0')
parser.add_argument('--shot_list', type=str, default='1,5')
parser.add_argument('--gd_lr', default=0.001, type=float,
help='The inner learning rate for MAML-Based model')
parser.add_argument('--inner_iters', default=5, type=int,
help='The inner iterations for MAML-Based model')
# parser.add_argument('--fsl', action='store_true', default=False) # test FSL or not
args = parser.parse_args()
args.shot = 1
args.orig_imsize = -1
args.n_view = 1
args.fix_BN = False
args.way = 5
args.temperature = 1
pprint(vars(args))
set_gpu(args.gpu)
# Dataset and Data Loader
if args.dataset == 'MiniImageNet':
# Handle MiniImageNet
from model.dataloader.mini_imagenet import MiniImageNet as Dataset
args.dropblock_size = 5
elif args.dataset == 'CUB':
from model.dataloader.cub import CUB as Dataset
args.dropblock_size = 5
elif args.dataset == 'TieredImageNet':
from model.dataloader.tiered_imagenet_raw import tieredImageNet as Dataset
args.dropblock_size = 5
else:
raise ValueError('Non-supported Dataset.')
trainset = Dataset('train', args)
testset = Dataset('test', args)
args.num_class = trainset.num_class
# shot = [1, 5, 10, 20, 30, 50]
# FSL Test, 1-Shot, 5-Way
num_shots = [int(e) for e in args.shot_list.split(',')]
pemute_list = list(permutations(range(5), 5))
test_acc_record_shot = np.zeros((2000, len(pemute_list), len(num_shots)))
test_loss_record_shot = np.zeros((2000, len(pemute_list), len(num_shots)))
test_acc_record_query = np.zeros((2000, len(pemute_list), len(num_shots)))
test_loss_record_query = np.zeros((2000, len(pemute_list), len(num_shots)))
for shot_ind, shot in enumerate(num_shots):
few_shot_sampler = CategoriesSampler(testset.label, 2000, 5, shot + 15)
few_shot_loader = DataLoader(dataset=testset, batch_sampler=few_shot_sampler, num_workers=4, pin_memory=True)
shot_logit_list = np.zeros((2000, len(pemute_list), args.way * shot, args.way))
query_logit_list = np.zeros((2000, len(pemute_list), args.way * 15, args.way))
update_norm_list = np.zeros((2000, len(pemute_list), 50))
label_shot, label_query = torch.arange(args.way).repeat(shot).long(), torch.arange(args.way).repeat(15).long()
if torch.cuda.is_available():
label_shot = label_shot.cuda()
label_query = label_query.cuda()
# Get Model
if args.model_class == 'MAML':
from model.models.maml import MAML
model = MAML(args)
else:
raise ValueError('No Such Model')
model_dict = model.state_dict()
pretrained_dict = torch.load(args.model_path, map_location='cpu')['params']
pretrained_dict = {k.replace('module.',''): v for k, v in pretrained_dict.items()}
pretrained_dict = {k: v for k, v in pretrained_dict.items() if k in model_dict}
print(pretrained_dict.keys())
model_dict.update(pretrained_dict)
model.load_state_dict(model_dict)
model.train()
model.args.shot = shot
if torch.cuda.is_available():
torch.backends.cudnn.benchmark = True
model = model.cuda()
support_emb = []
for i, batch in tqdm(enumerate(few_shot_loader), ncols=50, desc='1-Shot 5-Way Test'):
if torch.cuda.is_available():
data = batch[0].cuda()
else:
data = batch[0]
support, query = data[:5 * shot].view(shot, 5, 3, 84, 84), data[5 * shot:].view(15, 5, 3, 84, 84)
for p_index, p_value in enumerate(pemute_list):
support_c, query_c = support[:, p_value, :, :, :].view(-1, 3, 84, 84), query[:, p_value, :, :, :].view(-1, 3, 84, 84)
model.load_state_dict(model_dict)
if p_index == 0:
with torch.no_grad():
support_emb.append(model.encoder(support_c, embedding = True).detach().cpu())
model.load_state_dict(model_dict)
logits_shot, logits_query, updated_norm = model.forward_eval_perm(support_c, query_c)
shot_acc = count_acc(logits_shot, label_shot)
query_acc = count_acc(logits_query, label_query)
test_acc_record_shot[i, p_index, shot_ind] = shot_acc
test_loss_record_shot[i, p_index, shot_ind] = F.cross_entropy(logits_shot, label_shot).item()
test_acc_record_query[i, p_index, shot_ind] = query_acc
test_loss_record_query[i, p_index, shot_ind] = F.cross_entropy(logits_query, label_query).item()
shot_logit_list[i, p_index, :, :] = logits_shot.detach().cpu()
query_logit_list[i, p_index, :, :] = logits_query.detach().cpu()
update_norm_list[i, p_index, :] = updated_norm
del support_c, query_c, logits_shot, logits_query
torch.cuda.empty_cache()
del support, query
torch.cuda.empty_cache()
current_acc_record_shot = test_acc_record_shot[:,:, shot_ind]
current_loss_record_shot = test_loss_record_shot[:,:, shot_ind]
current_acc_record_query = test_acc_record_query[:,:, shot_ind]
current_loss_record_query = test_loss_record_query[:,:, shot_ind]
# save current record
savename = args.model_path.split('.')[-2].split('/')[-1].strip()
with open('{}-Test-{}-Shot.pkl'.format(savename, shot), 'wb') as f:
pickle.dump([current_acc_record_shot, current_loss_record_shot, current_acc_record_query, current_loss_record_query], f)
with open('{}-Test-{}-Shot-EMB.pkl'.format(savename, shot), 'wb') as f:
pickle.dump(support_emb, f)
with open('{}-Test-{}-logit.pkl'.format(savename, shot), 'wb') as f:
pickle.dump([shot_logit_list, query_logit_list], f)
with open('{}-Test-{}-updated_norm.pkl'.format(savename, shot), 'wb') as f:
pickle.dump(update_norm_list, f)
m1, pm1 = compute_confidence_interval(current_acc_record_query.mean(-1))
print('Shot-{}: Test Acc - {:.5f} + {:.5f}'.format(shot, m1, pm1))
# compute variance
max_value = current_acc_record_query.max(-1)
min_value = current_acc_record_query.min(-1)
m2, pm2 = compute_confidence_interval(max_value)
m3, pm3 = compute_confidence_interval(min_value)
print('Shot-{}: Test Max Acc - {:.5f} + {:.5f}'.format(shot, m2, pm2))
print('Shot-{}: Test Min Acc - {:.5f} + {:.5f}'.format(shot, m3, pm3))
for shot_ind, shot in enumerate(num_shots):
current_acc_record = test_acc_record_query[:,:, shot_ind]
m1, pm1 = compute_confidence_interval(current_acc_record.mean(-1))
print('Shot-{}: Test Acc - {:.5f} + {:.5f}'.format(shot, m1, pm1))
# compute variance
max_value = current_acc_record.max(-1)
min_value = current_acc_record.min(-1)
m2, pm2 = compute_confidence_interval(max_value)
m3, pm3 = compute_confidence_interval(min_value)
print('Shot-{}: Test Max Acc - {:.5f} + {:.5f}'.format(shot, m2, pm2))
print('Shot-{}: Test Min Acc - {:.5f} + {:.5f}'.format(shot, m3, pm3))