-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathfit.py
182 lines (156 loc) · 6.94 KB
/
fit.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
import os
import sys
import time
import torch
import numpy as np
import torch.nn as nn
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn import metrics
import torch.nn.functional as F
from dataset import get_Dataloader
from log import _logger
class EarlyStopping:
def __init__(self,logger, patience=20, verbose=False, delta=0):
self.patience = patience
self.verbose = verbose
self.counter = 0
self.best_score = None
self.early_stop = False
self.val_loss_min = np.Inf
self.delta = delta
self.logger = logger
def __call__(self, val_loss, model, path):
score = val_loss
if self.best_score is None:
self.best_score = score
self.save_checkpoint(val_loss, model, path)
elif score < self.best_score + self.delta:
self.counter += 1
self.logger.debug(f'EarlyStopping counter: {self.counter} out of {self.patience}')
if self.counter >= self.patience:
self.early_stop = True
else:
self.best_score = score
self.save_checkpoint(val_loss, model, path)
self.counter = 0
def save_checkpoint(self, val_loss, model, path):
if self.verbose:
self.logger.debug(
f'Validation loss decreased ({self.val_loss_min:.6f} --> {val_loss:.6f}). Saving model ...')
torch.save(model.state_dict(), os.path.join(path, '_checkpoint.pth'))
self.val_loss_min = val_loss
class fit(object):
def __init__(self,args,net):
# self.mode = MLPMixer(args.image_size,args.patch_size,args.hidden_dim,args.depth,dropout=args.dropout,num_classes=args.num_classes)
self.mode = net
# self.mode = kaggle_cnn()
# self.mode = MLP()
# self.mode = AlexNet()
# self.mode = convnextv2()
self.opt = torch.optim.Adam(self.mode.parameters(),lr=args.learning_rate)
self.lossfunc = nn.CrossEntropyLoss()
self.args = args
local_time = time.strftime("%Y年%m月%d日 %H时%M分%S秒")
if os.path.isdir('args.save_dir'+ f'{local_time}') is True:
self.log = _logger(args.save_dir + f'{local_time}/log')
else:
os.makedirs(args.save_dir + f'{local_time}')
self.log = _logger(args.save_dir + f'{local_time}/log')
self.log.debug(self.args)
self.save_dir = args.save_dir + f'{local_time}/'
if torch.cuda.is_available():
self.mode.cuda()
self.train_dataloader = get_Dataloader(mode='train',batch_size=args.batch_size)
self.val_dataloader = get_Dataloader(mode='val',batch_size=args.batch_size)
self.test_dataloader = get_Dataloader(mode='test',batch_size=args.batch_size)
self.earlystopping = EarlyStopping(logger=self.log)
def vis(self,conf,train_iter_loss,val_acc,val_loss):
print(val_acc)
_,ax = plt.subplots(2,2,figsize=(20,15))
sns.heatmap(conf, annot=True, cmap='Blues',ax=ax[0][0],fmt='d')
sns.lineplot(data=train_iter_loss,ax=ax[0][1])
ax[0][1].set_xlabel('Epoch')
ax[0][1].set_ylabel('train_loss')
ax[0][1].set_xlim(0,train_iter_loss.shape[0])
sns.lineplot(data=val_acc,ax=ax[1][0])
ax[1][0].set_xlabel('Epoch')
ax[1][0].set_ylabel('ACC')
ax[1][0].set_xlim(0,val_acc.shape[0])
ax[1][0].set_ylim(0,1)
sns.lineplot(data=val_loss,ax=ax[1][1])
ax[1][1].set_xlabel('Epoch')
ax[1][1].set_ylabel('val_loss')
ax[1][1].set_xlim(0,val_loss.shape[0])
plt.savefig(f'{self.save_dir}'+ 'vis.pdf',dpi=1200)
def val(self):
val_loss = []
val_acc = []
self.mode.eval()
for data, label in self.val_dataloader:
data = data.to(self.args.device)
label = label.to(self.args.device)
output = self.mode(data)
loss = self.lossfunc(output,label)
val_loss.append(loss.item())
acc = torch.sum(torch.argmax(output,dim=-1) == label)/output.shape[0]
val_acc.append(acc.item())
return np.average(val_loss) , np.average(val_acc)
def train(self):
train_iter_loss = []
val_acc_list = []
val_loss_list = []
iter = 0
self.log.debug('='*10 +'Train' + '='*10)
for i in range(self.args.epoch):
train_loss = 0
self.mode.train()
for data , label in self.train_dataloader:
iter += 1
self.opt.zero_grad()
data = data.to(self.args.device) # 128,1,28,28
label = label.to(self.args.device)
output = self.mode(data)
iter_loss = self.lossfunc(output,label)
train_loss += self.lossfunc(output,label).item() * data.shape[0]
train_iter_loss.append(iter_loss.item())
iter_loss.backward()
self.opt.step()
val_loss,val_acc = self.val()
val_loss_list.append(val_loss)
val_acc_list.append(val_acc)
self.log.debug(f'eopch:{i + 1} train_loss: {train_loss/len(self.train_dataloader.sampler):.2f} val_loss:{val_loss} val_acc:{val_acc}')
self.earlystopping(val_acc, self.mode, self.save_dir)
if self.earlystopping.early_stop:
print("Early stopping")
np.save(f'{self.save_dir}' + 'train_iter_loss',np.array(train_iter_loss))
np.save(f'{self.save_dir}' + 'val_loss',np.array(val_loss_list))
np.save(f'{self.save_dir}' + 'val_acc',np.array(val_acc_list))
break
np.save(f'{self.save_dir}' + 'train_iter_loss',np.array(train_iter_loss))
np.save(f'{self.save_dir}' + 'val_loss',np.array(val_loss_list))
np.save(f'{self.save_dir}' + 'val_acc',np.array(val_acc_list))
def test(self):
self.mode.load_state_dict(
torch.load(
os.path.join(self.save_dir + '_checkpoint.pth')))
self.mode.eval()
pred_label = []
true_label = []
acc_num = 0
all_num = 0
for data ,label in self.test_dataloader:
data = data.to(self.args.device)
output = self.mode(data)
acc_num += torch.sum(torch.argmax(output.cpu().detach(),axis=-1) == label)
all_num += data.shape[0]
# print(label.shape)
# print(output.shape)
pred_label.extend(list(np.array(np.argmax(output.cpu().detach(),axis=-1)).reshape(-1)))
true_label.extend(list(np.array(label).reshape(-1)))
self.log.debug(f'test_acc:{acc_num / all_num}')
conf = metrics.confusion_matrix(pred_label,true_label)
train_iter_loss = np.load(f'{self.save_dir}' + 'train_iter_loss.npy')
val_loss = np.load(f'{self.save_dir}' + 'val_loss.npy')
val_acc = np.load( f'{self.save_dir}' + 'val_acc.npy')
self.vis(conf,train_iter_loss,val_acc,val_loss)