-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmy_train_DDP.py
143 lines (114 loc) · 5.51 KB
/
my_train_DDP.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
import torch
from torch.utils import data
from torch import nn
from torch.utils.tensorboard import SummaryWriter
import tqdm
from torchsummary import summary
import my_pfn_load
import my_PFN
import my_tools
import numpy as np
import sys
import os
import time
import DDP_Config
import torch.distributed as dist
import torch.multiprocessing as mp
def main(local_rank, suffix, process, train_set, val_set, test_set):
DDP_Config.init_ddp(local_rank)
########################################################################################################
#Hyperparameters
epochs=20
batchsize=512
lr=0.001
use_bn=True
Phi_sizes=(100, 100, 128)
F_sizes=(100, 100, 100)
num_classes=len(process)
#########################################################################################################
if local_rank == 0:
print(f'''
classes: {num_classes}
epochs: {epochs}
batchsize: {batchsize}
lr: {lr}
use_bn: {use_bn}
Phi_sizes: {Phi_sizes}
F_sizes: {F_sizes}
''')
input_shape_for_one_event=train_set[0][0].shape
feature_dim=input_shape_for_one_event[0]
train_sampler=data.DistributedSampler(train_set)
val_sampler=data.DistributedSampler(val_set)
train_set=data.DataLoader(train_set,batch_size=batchsize,shuffle=False,sampler=train_sampler)
val_set=data.DataLoader(val_set,batch_size=batchsize,shuffle=False,sampler=val_sampler)
test_set=data.DataLoader(test_set,batch_size=batchsize,shuffle=False)
net=my_PFN.ParticleFlowNetwork(num_classes=num_classes, input_dims=feature_dim, Phi_sizes=Phi_sizes,
F_sizes=F_sizes,use_bn=use_bn)
net.apply(my_tools.initialize_pfn)
loss=nn.CrossEntropyLoss(reduction='none') #none is essential！
loss.to(local_rank)
net.to(local_rank)
if local_rank==0:
writer=SummaryWriter(f'log_tensorboard/log_{suffix}')
print(f'TensorBoard log path is {os.getcwd()}/log_tensorboard/log_{suffix}')
writer.add_graph(net,input_to_model=torch.rand(1,*input_shape_for_one_event).to(local_rank))
print(summary(net,input_size=input_shape_for_one_event))
net=nn.parallel.DistributedDataParallel(net,device_ids=[local_rank])
optimizer=torch.optim.NAdam(net.parameters(),lr=lr)#After the "net.to(device=device)"
if local_rank==0:
torch.cuda.synchronize()
start=time.time()
for epoch in range(epochs):
train_set.sampler.set_epoch(epoch) #Essential!!! Otherwize each GPU only get same ntuples every epoch
#train_sampler.set_epoch(epoch) #Same as the above, both are correct
loss_train,acc_train=my_tools.train_procedure_in_each_epoch(net,train_set,loss,optimizer,local_rank)
loss_val,acc_val=my_tools.evaluate_accuracy(net,loss,val_set,test=False)
if local_rank==0:
writer.add_scalars('loss_and_acc_in_train_and_val',{'loss_train':loss_train,
'acc_train':acc_train,
'loss_val':loss_val,
'acc_val':acc_val},epoch)
print(f'''epoch: {epoch} | acc_train: {acc_train:.3f} | acc_val: {acc_val:.3f} | loss_train: {loss_train:.3f} | loss_val: {loss_val:.3f} |
''')
if local_rank==0:
torch.cuda.synchronize()
end=time.time()
print('Total time in training is ',end-start)
if local_rank==0:
my_tools.save_net(net,suffix)
score,true_label,loss_test,acc_test=my_tools.evaluate_accuracy(net,loss,test_set,test=True)
print('{:-^80}'.format(f'loss in test is {loss_test:.3f}, accuracy in test is {acc_test:.3f},'))
score_label=torch.argmax(score,dim=1)
true_label=true_label.cpu().numpy()
score_label=score_label.cpu().numpy()
my_tools.plot_confusion_matrix(suffix,num_classes,true_label,score_label,classes=process)
dist.destroy_process_group()
###################################################################################################################################
if __name__=='__main__':
my_tools.cheems()#means start...
try:
suffix=sys.argv[1]
except:
default_suffix=time.strftime('%Y-%m-%d_%H_%M_%S',time.localtime())
print(f"Seems like you didn't type a suffix, so the [tag_'localtime({default_suffix})'] has been used as the suffix")
suffix='tag_'+default_suffix
gpu_nums=torch.cuda.device_count()
print(f"The number of GPU(s) is {gpu_nums}!")
#########################################################################################################
process=['Hbb','Hcc', 'Hgg', 'Hww', 'Hzz', 'Pll', 'Pww_l', 'Pzz_l', 'Pzz_sl']
num_data_each_class=300_00
train_val_test=[0.8,0.1,0.1]
#########################################################################################################
fimename=[i +'.root' for i in process]
dataset=my_pfn_load.load(filename=fimename,num_data=num_data_each_class)
train_set,val_set,test_set=data.random_split(dataset=dataset,lengths=train_val_test)
print(f'''
suffix: {suffix}
gpu_nums: {gpu_nums}
process: {process}
tra_val_test: {train_val_test}
numdata_ec {num_data_each_class}
''')
mp.spawn(main,args=(suffix ,process, train_set, val_set, test_set), nprocs=torch.cuda.device_count())
print('Program Is Over !!!')