Tensorboard的可视化损失函数（loss）、准确率（Accuracy）、梯度（grad）、权值（data）的源码，以Pytorch手写数字集为例讲解
一、源码
import torch
import torch.nn as nn
import torch.utils.data as Data
from torchvision.datasets import mnist
import torchvision.transforms as transforms
from torch.autograd import Variable
from torch.utils.tensorboard import SummaryWriter
#############################Download Data################60000张训练，10000张测试
train_dataset =mnist.MNIST(root='./mnist/', train=True, transform=transforms.ToTensor())
test_dataset = mnist.MNIST(root='./mnist/',train=True,transform=transforms.ToTensor())
train_loader = Data.DataLoader(dataset=train_dataset,batch_size=50,shuffle=True)
test_loader =Data.DataLoader ( dataset=test_dataset,batch_size=50,shuffle=False)
###########################Define CNN module################################
class CNN(nn.Module):
def __init__(self):
super(CNN, self).__init__()
# 定义卷基层
self.layer1 = nn.Sequential(
nn.Conv2d(1, 16, kernel_size=3,stride=1), # b,16,26,26
nn.BatchNorm2d(16),
nn.ReLU())
self.layer2= nn.Sequential(
nn.Conv2d(16, 32, kernel_size=3,stride=1), # b.32,24,24
nn.BatchNorm2d(32),
nn.ReLU(),
nn.MaxPool2d(kernel_size=2,stride=2)) # b,32,12,12
self.layer3=nn.Sequential(
nn.Conv2d(32,64,kernel_size=3,stride=1), # b,64,10,10
nn.BatchNorm2d(64),
nn.ReLU())
self.layer4=nn.Sequential(
nn.Conv2d(64,128,kernel_size=3,stride=1), # b,128,8,8
nn.BatchNorm2d(128),
nn.ReLU(),
nn.MaxPool2d(kernel_size=2,stride=2)) # b,128,4,4
self.fc=nn.Sequential(
nn.Linear(128*4*4,1024),
nn.ReLU(),
nn.Linear(1024,128),
nn.ReLU(),
nn.Linear(128,10))
def forward(self, x):
x=self.layer1(x)
x=self.layer2(x)
x=self.layer3(x)
x=self.layer4(x)
x = x.view(x.size(0), -1) # reshape 拉平
x=self.fc(x)
return x
########################loss and optimizer##################################
cnn = CNN()
if torch.cuda.is_available():
cnn = cnn.cuda()
criterion= nn.CrossEntropyLoss() #交叉熵
optimizer = torch.optim.Adam(cnn.parameters(), lr=0.001) #Adam 优化方式
# 构建 SummaryWriter
writer = SummaryWriter(comment='test_your_comment', filename_suffix="_test_your_filename_suffix")
##########################train#################################
for epoch in range(5):
train_loss=0
train_acc=0
for step, (x, label) in enumerate(train_loader):
x = Variable(x).cuda() #50,1,28,28
label = Variable(label).cuda()
###forward######
out=cnn(x) #50,10
loss=criterion(out,label) # 计算损失函数
###backforward#######
optimizer.zero_grad() # 梯度归零
loss.backward() # 反向传播
optimizer.step() # 梯度优化
train_loss+=loss.item()
###计算准确率#####
_, pred = out.max(1)
# print(out.max(1))
num_correct = (pred == label).sum().item()
acc = num_correct /x.shape[0]
train_acc += acc
aver_loss = train_loss/len(train_loader)
aver_acc = train_acc / len(train_loader)
print('Epoch: {}, Train Loss: {:.6f}, Train Acc: {:.6f}'
.format(epoch, train_loss / len(train_loader), train_acc / len(train_loader)))
#########记录数据，保存于event file，这里记录了每一个epoch的损失和准确度########
writer.add_scalars("Loss", {
"Train": aver_loss}, epoch)
writer.add_scalars("Accuracy", {
"Train": aver_acc}, epoch)
############## 每个epoch，记录梯度，权值#######################################
for name, param in cnn.named_parameters(): #返回模型的参数
writer.add_histogram(name + '_grad', param.grad, epoch) #参数的梯度
writer.add_histogram(name + '_data', param, epoch) #参数的权值
二、打开tensorboard方式
1.打开pycharm中的Terminal
2.输入命令 tensorboard –-logdir=+"路径"即可，定位到runs文件
位置
该代码执行完之后会出现一个runs文件夹
3.打开网页即可
4.就会显示如下界面
我们在代码中定义的每个epoch的loss和Accuarcy
定义的每个网络层的权值分布情况
该图显示的是我们定义每个epoch权重的梯度和权值