-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
executable file
·205 lines (173 loc) · 8.6 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
201
202
203
204
205
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader
from model import CNN1,CNN2,CNN3,SegRNN,GRU, LSTM, BiGRU, BiLSTM, iTransformer, PatchTST
from dataset.dataset import ASEDataset
from tqdm import tqdm
import time
from dataset.parse import load_dataset,bar_progress
from sklearn.metrics import f1_score,precision_score,recall_score,accuracy_score
import os
from utils.tools import EarlyStopping, adjust_learning_rate_withWarmup
import json
import wandb
import warnings
time_exp_dic = {'time': 0, 'counter': 0}
class LabelSmoothingLoss(nn.Module):
def __init__(self, smoothing=0.1):
super(LabelSmoothingLoss, self).__init__()
self.smoothing = smoothing
def forward(self, pred, target):
# 将目标标签转换为 one-hot 编码
num_classes = pred.size(1)
with torch.no_grad():
true_dist = torch.zeros_like(pred)
true_dist.fill_(self.smoothing / (num_classes - 1))
true_dist.scatter_(1, target.data.unsqueeze(1), 1.0 - self.smoothing)
return torch.mean(torch.sum(-true_dist * torch.log_softmax(pred, dim=-1), dim=-1))
class FocalLoss(nn.Module):
def __init__(self, alpha=1, gamma=2, reduction='mean'):
super(FocalLoss, self).__init__()
self.alpha = alpha
self.gamma = gamma
self.reduction = reduction
def forward(self, outputs, labels):
ce_loss = nn.CrossEntropyLoss(reduction='none')(outputs, labels)
probs = torch.softmax(outputs, dim=1)
p_t = probs.gather(1, labels.unsqueeze(1)).squeeze()
alpha_t = self.alpha
fl_loss = alpha_t * (1 - p_t) ** self.gamma * ce_loss
if self.reduction == 'mean':
return fl_loss.mean()
elif self.reduction == 'sum':
return fl_loss.sum()
else:
return fl_loss
def train(args,nowtime):
warnings.filterwarnings("ignore")
if args.use_gpu:
device = torch.device('cuda:{}'.format(args.gpu))
else:
device = torch.device('cpu')
# 1 : Dataset in DB Files
train_dataset = ASEDataset([os.path.join(args.datapath, 'trainV.db')],encode_element=False,train=False)
val_dataset = ASEDataset([os.path.join(args.datapath, 'val.db')],encode_element=False,train=False)
test_dataset = ASEDataset([os.path.join(args.datapath, 'rruff.db')],encode_element=False,train=False)
train_loader = DataLoader(train_dataset, batch_size=args.batch_size, shuffle=True, num_workers=args.num_workers)
val_loader = DataLoader(val_dataset, batch_size=args.batch_size, shuffle=True, num_workers=args.num_workers,drop_last=False)
test_loader = DataLoader(test_dataset, batch_size=args.batch_size, shuffle=False, num_workers=args.num_workers,drop_last=False)
"""
2 : Dataset in TFRecord Files: A Small Dataset for Review
# 1. Point to a local or remote Croissant file
# import mlcroissant as mlc
url = "https://huggingface.co/datasets/caobin/SimXRDreview/raw/main/simxrd_croissant.json"
# 2. Inspect metadata
dataset_info = mlc.Dataset(url).metadata.to_json
print(dataset_info)
for file_info in dataset_info['distribution']:
wget.download(file_info['contentUrl'], './', bar=bar_progress)
# 3. Use Croissant dataset in your ML workload
from dataset.parse import load_dataset,bar_progress # defined in our github
train_loader = DataLoader(load_dataset(name='train.tfrecord'), batch_size=args.batch_size, shuffle=True, num_workers=args.num_workers)
val_loader = DataLoader(load_dataset(name='val.tfrecord'), batch_size=args.batch_size, shuffle=True, num_workers=args.num_workers,drop_last=False)
test_loader = DataLoader(load_dataset(name='test.tfrecord'), batch_size=args.batch_size, shuffle=False, num_workers=args.num_workers,drop_last=False)
"""
if args.model == 'cnn1':
model = CNN1.Model(0.7,args)
elif args.model == 'cnn2':
model = CNN2.Model(args)
elif args.model == 'cnn3':
model = CNN3.Model(args)
elif args.model == 'lstm':
model = LSTM.Model(args)
elif args.model == 'gru':
model = GRU.Model(args)
elif args.model == 'bilstm':
model = BiLSTM.Model(args)
elif args.model == 'bigru':
model = BiGRU.Model(args)
elif args.model == 'SegRNN':
model = SegRNN.Model(args)
elif args.model == 'iTransformer':
model = iTransformer.Model(args)
elif args.model == 'PatchTST':
model = PatchTST.Model(args)
save_path = f'/data/zzn/prl/checkpoints/{args.model}_lr{args.lr}_bs{args.batch_size}_{nowtime}_{args.seed}'
if not os.path.exists('/data/zzn/prl/checkpoints'):
os.mkdir('/data/zzn/prl/checkpoints')
os.makedirs(save_path, exist_ok=True)
with open(save_path+'/args.json', 'w') as f:
json.dump(args.__dict__, f)
model = model.to(device)
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=args.lr)
early_stopping = EarlyStopping(patience=args.patience, verbose=True)
for epoch in range(args.epochs):
train_loss, _ = run_epoch(model, optimizer, criterion, epoch, train_loader, device, args)
val_loss, _ = run_epoch(model, optimizer, criterion, epoch, val_loader,device, args, backprop=False)
test_loss, res = run_epoch(model, optimizer, criterion, epoch, test_loader,device, args, backprop=False)
wandb.log({"epoch": epoch, "val_loss": val_loss,
"test_loss":test_loss, "test_f1":res['macro_f1'],
"test_precision":res['macro_precision'],
"test_recall":res['macro_recall'],"test_acc": res['accuracy'],"infer_time":time_exp_dic['time'] / time_exp_dic['counter']})
early_stopping(epoch,val_loss,test_loss,res, model, save_path)
if early_stopping.early_stop:
print("Early stopping")
break
wandb.log({"epoch": early_stopping.best_epoch, "val_loss": early_stopping.val_loss_min,
"test_loss":early_stopping.best_test_loss, "test_f1": early_stopping.best_test_macrof1,
"test_precision":early_stopping.best_test_precision ,
"test_recall":early_stopping.best_test_recall ,"test_acc": early_stopping.best_test_accuracy})
print("*** Best Val Loss: %.5f \t Best Test Loss: %.5f \t Best Test Accuracy: %.5f \t Best Macro-F1: %.5f\t Best epoch %d" %
(early_stopping.val_loss_min, early_stopping.best_test_loss, early_stopping.best_test_accuracy,early_stopping.best_test_macrof1, early_stopping.best_epoch))
return early_stopping
def run_epoch(model, optimizer, criterion, epoch, loader, device, args, backprop=True):
if backprop:
model.train()
else:
model.eval()
res = {'epoch': epoch, 'loss': 0, 'accuracy': 0, 'macro_f1': 0 ,'counter': 0,'macro_precision':0,"macro_recall":0}
all_labels = []
all_predicted = []
for batch_index, data in enumerate(tqdm(loader)):
intensity,labels,element = data['intensity'].to(device), data['id'].to(device), data['element'].to(device)
intensity = intensity.unsqueeze(1)
element = element.unsqueeze(1)
batch_size=intensity.shape[0]
if backprop:
optimizer.zero_grad()
torch.cuda.synchronize()
t1 = time.time()
outputs = model(intensity)
torch.cuda.synchronize()
t2 = time.time()
time_exp_dic['time'] += t2 - t1
time_exp_dic['counter'] += 1
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
else:
outputs = model(intensity)
loss = criterion(outputs, labels)
_, predicted = torch.max(outputs, 1)
res['loss'] += loss.item()*batch_size
res['counter'] += batch_size
all_labels.extend(labels.cpu().numpy())
all_predicted.extend(predicted.cpu().numpy())
if args.model == 'CPICANN':
adjust_learning_rate_withWarmup(optimizer, epoch + batch_index / len(loader), args)
accuracy = accuracy_score(all_labels, all_predicted)
macro_f1 = f1_score(all_labels, all_predicted, average='macro')
macro_precision = precision_score(all_labels, all_predicted, average='macro')
macro_recall = recall_score(all_labels, all_predicted, average='macro')
res['accuracy'] = accuracy
res['macro_f1'] = macro_f1
res['macro_precision'] = macro_precision
res['macro_recall'] = macro_recall
if not backprop:
prefix = "==> "
else:
prefix = " "
print('%s epoch %d avg loss: %.5f' % (prefix, epoch, res['loss'] / res['counter']))
return res['loss'] / res['counter'] , res