-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmulti_runs_joint.py
190 lines (147 loc) · 6.74 KB
/
multi_runs_joint.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
182
183
184
185
186
187
188
189
190
import datetime
import os
import numpy as np
import torch
from torch.optim import Adam
from torchvision import datasets, transforms
from agent import get_agent
from models import get_model
from models.buffer import Buffer
from utils.util import Logger
def get_cifar_data_joint(dataset_name, batch_size, n_workers):
size = [3, 32, 32]
if dataset_name == "cifar10":
class_num = 10
elif dataset_name == "cifar100":
class_num = 100
dataset_path = './data/'
mean = [x / 255 for x in [125.3, 123.0, 113.9]]
std = [x / 255 for x in [63.0, 62.1, 66.7]]
dataset = {}
if dataset_name == "cifar10":
dataset['train'] = datasets.CIFAR10(dataset_path, train=True, download=True, transform=transforms.Compose(
[transforms.ToTensor(), transforms.Normalize(mean, std)]))
dataset['test'] = datasets.CIFAR10(dataset_path, train=False, download=True, transform=transforms.Compose(
[transforms.ToTensor(), transforms.Normalize(mean, std)]))
elif dataset_name == "cifar100":
dataset['train'] = datasets.CIFAR100(dataset_path, train=True, download=True, transform=transforms.Compose(
[transforms.ToTensor(), transforms.Normalize(mean, std)]))
dataset['test'] = datasets.CIFAR100(dataset_path, train=False, download=True, transform=transforms.Compose(
[transforms.ToTensor(), transforms.Normalize(mean, std)]))
Loader = {}
Loader['train'] = torch.utils.data.DataLoader(
dataset['train'],
batch_size=batch_size,
shuffle=True,
num_workers=n_workers,
)
Loader['test'] = torch.utils.data.DataLoader(
dataset['test'],
batch_size=64,
shuffle=True,
num_workers=n_workers,
)
return class_num, Loader, size
def multiple_run_joint(args):
test_all_acc = torch.zeros(args.run_nums)
last_test_all_acc = torch.zeros(args.run_nums)
for run in range(args.run_nums):
tmp_acc = []
last_tmp_acc = []
buffer_tmp_acc = []
buffer_last_tmp_acc = []
train_tmp_acc = []
train_last_tmp_acc = []
print('=' * 100)
print(f"-----------------------------run {run} start--------------------------")
print(datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S'))
print('=' * 100)
class_num, task_loader, input_size = get_cifar_data_joint(
dataset_name=args.dataset, batch_size=args.batch_size, n_workers=args.n_workers
)
args.n_classes = class_num
setattr(args, 'run_name', f"{args.exp_name} run_{run:02d}")
print(f"\nRun {run}: {args.run_name} {'*' * 50}\n")
logger = Logger(args, base_dir=f"./outputs/{args.method}/{args.dataset}")
buffer = Buffer(args, input_size).cuda()
model = get_model(method_name=args.method, nclasses=class_num).cuda()
optimizer = Adam(model.parameters(), args.lr, weight_decay=args.wd)
agent = get_agent(
method_name=args.method, model=model,
buffer=buffer, optimizer=optimizer, input_size=input_size, args=args
)
print(f"number of classifier parameters:\t {model.n_params/1e6:.2f}M", )
print(f"buffer parameters (image size prod):\t {np.prod(buffer.bx.size())/1e6:.2f}M", )
for i in range(1):
print(f"\n-----------------------------run {run} task id:{i} start training-----------------------------")
train_log_holder = agent.train(i, task_loader['train'])
acc_list, all_acc_list = agent.test(i, task_loader)
# empirical analysis of overfitting-underfitting dilemma
buffer_acc_list, buffer_all_acc_list = agent.test_buffer(i, task_loader)
train_acc_list, train_all_acc_list = agent.test_train(i, task_loader)
tmp_acc.append(acc_list)
last_tmp_acc.append(all_acc_list['3'])
buffer_tmp_acc.append(buffer_acc_list)
buffer_last_tmp_acc.append(buffer_all_acc_list['3'])
train_tmp_acc.append(train_acc_list)
train_last_tmp_acc.append(train_all_acc_list['3'])
logger.log_losses(train_log_holder)
logger.log_accs(all_acc_list)
# record the intermediate final accs
for feat_id, acc_list_id in all_acc_list.items():
test_accuracy_id = acc_list_id[:i+1].mean()
logger.log_scalars({
f"test/{feat_id}_avg_acc": test_accuracy_id,
}, step=agent.total_step)
test_accuracy = acc_list.mean()
test_all_acc[run] = test_accuracy
tmp_acc = np.array(tmp_acc)
logger.log_scalars({
'test/final_avg_acc': test_accuracy,
'metrics/buffer_n_bits': agent.buffer.n_bits / 1e6,
'metrics/model_n_params': agent.model.n_params / 1e6
}, step=agent.total_step+1, verbose=True)
logger.log_accs_table(
name='task_accs_table', accs_list=tmp_acc,
step=agent.total_step+1, verbose=True
)
# record the last scalars
last_acc_list = all_acc_list['3']
last_test_accuracy = last_acc_list.mean()
last_test_all_acc[run] = last_test_accuracy
last_tmp_acc = np.array(last_tmp_acc)
logger.log_scalars({
'test/last_final_avg_acc': last_test_accuracy,
}, step=agent.total_step+1, verbose=True)
logger.log_accs_table(
name='last_task_accs_table', accs_list=last_tmp_acc,
step=agent.total_step+1, verbose=True
)
buffer_tmp_acc = np.array(buffer_tmp_acc)
buffer_last_tmp_acc = np.array(buffer_last_tmp_acc)
train_tmp_acc = np.array(train_tmp_acc)
train_last_tmp_acc = np.array(train_last_tmp_acc)
logger.log_accs_table(
name='buffer_task_accs_table', accs_list=buffer_tmp_acc,
step=agent.total_step+1, verbose=True
)
logger.log_accs_table(
name='buffer_last_task_accs_table', accs_list=buffer_last_tmp_acc,
step=agent.total_step+1, verbose=True
)
logger.log_accs_table(
name='train_task_accs_table', accs_list=train_tmp_acc,
step=agent.total_step+1, verbose=True
)
logger.log_accs_table(
name='train_last_task_accs_table', accs_list=train_last_tmp_acc,
step=agent.total_step+1, verbose=True
)
print('=' * 100)
print("{}th run's Test result: Accuracy: {:.2f}%".format(run, test_accuracy))
print('=' * 100)
logger.close()
print(f"\n{'=' * 100}")
print(f"total {args.run_nums}runs last test acc results: {last_test_all_acc}")
print(f"\n{'=' * 100}")
print(f"total {args.run_nums}runs test acc results: {test_all_acc}")