-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
200 lines (164 loc) · 10.6 KB
/
train.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
191
192
193
194
195
196
197
198
199
200
# -*- coding: utf-8 -*-
import torch
import numpy as np
from tqdm import tqdm
import torch.optim as optim
from torch.utils.data import DataLoader
from args import args, device
from data.helper import get_model, get_dataset_omniglot, get_dataset_miniimagenet
from algorithms.meta_sgd import MetaSGD, meta_sgd_loop
from algorithms.maml import maml_loop
from algorithms.mada_sgd import adaptiveSGD, adaptiveSGD_V2, adaptiveSGD2, adaptiveSGD3, mada_sgd_loop
import copy
import scipy.io
from data.dataset.MiniImagenet import MiniImagenet
if __name__ == '__main__':
if args.adaptiveSGD:
args.test_inner_step = args.train_inner_step
# backbone networks
model = get_model(args, device)
print('-' * 100)
print('algorithm: {}, train inner steps: {}, test_inner step:{}, backbone: {}, paras num:{} K'
.format(args.algorithm, args.train_inner_step, args.test_inner_step, args.backbone,
sum(p.numel() for p in model.parameters() if p.requires_grad)/1000))
# 按照Omniglot数据集类型导入数据
'''
train_dataset, val_dataset = get_dataset_omniglot(args)
train_loader = DataLoader(train_dataset, batch_size=args.task_num, shuffle=True, num_workers=args.num_workers)
val_loader = DataLoader(val_dataset, batch_size=args.val_task_num, shuffle=False, num_workers=args.num_workers)
'''
# 按照MiniImageNet数据集类型导入数据
train_dataset, test_dataset = get_dataset_miniimagenet(args)
train_loader = DataLoader(train_dataset, batch_size=args.train_batch_size, shuffle=True, num_workers=args.num_workers)
test_loader = DataLoader(test_dataset, batch_size=args.test_batch_size, shuffle=True, num_workers=args.num_workers)
print('-' * 100)
print('train image num: {}, train image class: {}, test image num: {}, test image class: {}'.format(len([img for imgs in train_dataset.data for img in imgs]), train_dataset.cls_num,
len([img for imgs in test_dataset.data for img in imgs]), test_dataset.cls_num))
print('-' * 100)
if args.algorithm == 'MAML':
model_params = [p for p in model.parameters() if p.requires_grad]
model_optimizer = optim.Adam(model_params, lr=args.lamda)
if args.algorithm == 'Meta-SGD':
meta_sgd = MetaSGD(copy.deepcopy(model)) # 对参数进行深度复制，与原参数储存下不同的内存中。
model_params = [p for p in model.parameters() if p.requires_grad]
alpha_params = [p for p in meta_sgd.model.parameters() if p.requires_grad]
model_optimizer = optim.Adam(model_params, lr=1e-3)
alpha_optimizer = optim.Adam(alpha_params, lr=1e-4)
if args.algorithm == 'Mada-SGD':
#-------model-level + layer-level------------meda_sgd-------------------
mada_sgd = adaptiveSGD(inner_step=args.train_inner_step, layer_num=len(list(model.parameters())))
mada_sgd = adaptiveSGD_V2(inner_step=args.train_inner_step, merge_step=args.merge_inner_step, layer_num=len(list(model.parameters())))
beta_params = [mada_sgd.beta]
eta_params = [mada_sgd.eta]
# -------element-level-----mada_sgd--------
# beta_models = [copy.deepcopy(model) for i in range(args.train_inner_step)] # element-level
# eta_models = [copy.deepcopy(model) for i in range(args.train_inner_step)]
# mada_sgd = adaptiveSGD3(beta_models=beta_models, eta_models=eta_models,
# inner_step=args.train_inner_step, layer_num=len(list(model.parameters())))
# beta_params = [p for model in mada_sgd.beta_models for p in model.parameters() if p.requires_grad]
# eta_params = [p for model in mada_sgd.eta_models for p in model.parameters() if p.requires_grad]
# beta_eta_params = beta_params + eta_params
model_params = [p for p in model.parameters() if p.requires_grad]
model_optimizer = optim.Adam(model_params, lr=9e-4)
beta_optimizer = optim.Adam(beta_params, lr=2e-3)
eta_optimizer = optim.Adam(eta_params, lr=2e-3)
print('model optimizer lr: {}, beta optimizer lr: {}, eta optimizer lr: {}'.format(9e-4, 2e-3, 2e-3))
print('-' * 100)
best_acc = 0
val_acc_curve = []
val_loss_curve = []
beta = []
eta = []
for epoch in range(args.epochs):
# model.train()
# ada_sgd.beta_model.train()
train_bar = tqdm(train_loader)
for batch, (support_images, support_labels, query_images, query_labels) in enumerate(train_bar):
model.train()
train_bar.set_description("epoch {}/{}".format(epoch + 1, args.epochs))
# save beta & eta
aa = copy.deepcopy(mada_sgd.beta.data).reshape(1, -1)
bb = copy.deepcopy(mada_sgd.eta.data).reshape(1, -1)
beta.append(aa)
eta.append(bb)
train_acc = []
val_acc = []
train_loss = []
val_loss = []
val_predict = []
val_label = []
# Get variables
support_images = support_images.float().to(device)
support_labels = support_labels.long().to(device)
query_images = query_images.float().to(device)
query_labels = query_labels.long().to(device)
if args.algorithm == 'MAML':
loss, acc = maml_loop(model, support_images, support_labels, query_images, query_labels,
args.train_inner_step, model_optimizer, is_train=True)
if args.algorithm == 'Meta-SGD':
loss, acc = meta_sgd_loop(model, support_images, support_labels, query_images, query_labels,
args.train_inner_step, model_optimizer, alpha_optimizer, meta_sgd, is_train=True)
if args.algorithm == 'Mada-SGD':
loss, acc, _ = mada_sgd_loop(model, support_images, support_labels, query_images, query_labels,
args.train_inner_step, model_optimizer, beta_optimizer, eta_optimizer, mada_sgd, is_train=True)
train_loss.append(loss.item())
train_acc.append(acc)
train_bar.set_postfix(loss="{:.4f}".format(loss.item()))
# 每5个batch测试一次
if (batch+1) % 5 == 0:
model.eval()
# ada_sgd.beta_model.eval()
for support_images, support_labels, query_images, query_labels in test_loader:
# Get variables
support_images = support_images.float().to(device)
support_labels = support_labels.long().to(device)
query_images = query_images.float().to(device)
query_labels = query_labels.long().to(device)
if args.algorithm == 'MAML':
loss, acc = maml_loop(model, support_images, support_labels, query_images, query_labels,
args.test_inner_step, model_optimizer, is_train=False)
if args.algorithm == 'Meta-SGD':
loss, acc = meta_sgd_loop(model, support_images, support_labels, query_images, query_labels,
args.test_inner_step, model_optimizer, alpha_optimizer, meta_sgd, is_train=False)
if args.algorithm == 'Mada-SGD':
loss, acc, predict = mada_sgd_loop(model, support_images, support_labels, query_images, query_labels,
args.test_inner_step, model_optimizer, beta_optimizer, eta_optimizer, mada_sgd, is_train=False)
val_predict.append(predict)
# Must use .item() to add total loss, or will occur GPU memory leak.
# Because dynamic graph is created during forward, collect in backward.
val_loss.append(loss.item())
val_acc.append(acc)
val_label.append(query_labels)
if args.adaptiveSGD:
# print('inner step weight factors: {}'.format(mada_sgd.beta))
# print('inner step learning factors: {}'.format(mada_sgd.eta))
# beta = [beta.detach().cpu() for weight in ada_sgd.beta for beta in weight]
# for w in beta:
# np.savetxt('./results/beta.txt', np.array(w).reshape(1, -1))
pass
print("\n=> train_loss: {:.4f} train_acc: {:.4f} val_loss: {:.4f} val_acc: {:.4f}".
format(np.mean(train_loss), np.mean(train_acc), np.mean(val_loss), np.mean(val_acc)))
print("=> Min_Acc: {:.4f} Max_acc: {:.4f} Mean_Acc: {:.4f} Std: {:.6f}".
format(np.min(val_acc), np.max(val_acc), np.mean(val_acc), np.std(val_acc)))
val_acc_curve.append(np.mean(val_acc))
val_loss_curve.append(np.mean(val_loss))
if np.mean(val_acc) > best_acc:
best_acc = np.mean(val_acc)
save_path = ('./results/models/' + args.dataset + '-' + args.backbone + '-' + args.algorithm
+ '-' + str(args.n_way) + 'way' + '-' + str(args.k_shot) + 'shot' + '-' +
'epoch-' + str(epoch+1) + '-' + 'batch-'+ str(batch+1) + '-' + str('%0.4f' % best_acc) + 'acc.pt')
if args.algorithm == 'MAML':
torch.save(model, save_path)
if args.algorithm == 'Meta-SGD':
torch.save({'base_learner': model,
'alpha': meta_sgd}, save_path)
if args.algorithm == 'Mada-SGD':
torch.save({'base_learner': model,
'algorithm': mada_sgd,
'val_acc': val_acc,
'val_label': val_label,
'val_predict': val_predict}, save_path)
beta = torch.cat(beta, dim=0)
eta = torch.cat(eta, dim=0)
scipy.io.savemat('./results/models/beta_eta.mat', {'beta': beta.data.cpu().numpy(), 'eta': eta.data.cpu().numpy()})
scipy.io.savemat('./results/models/val_acc_loss_curve.mat', {'val_acc':val_acc_curve, 'val_loss':val_loss_curve})