Pytorch与tensorboard观察Loss变化

下载相应库

pip3 install tensorboardX
pip3 install tensorboard
conda install tensorflow
1
2
3

使用tensorboard

调包

from torch.utils.tensorboard import SummaryWriter 
1

初始化

writer = SummaryWriter(The file path)

Writer = SummaryWriter() 
或者
writer = SummaryWriter(os.path.join(basedir, expname))
# 在默认或者定义的文件夹中生成记录的loss文件
1
2
3
4

记录变量值

writer.add_scalar(“The text explains”,Loss,The number of iterations）

writer.add_scalar('Loss', loss, i)
writer.add_scalar('psnr', psnr, i)
1
2

生成下面的文件

查看Loss

在终端输入：
tensorboard --logdir --port

tensorboard --logdir='/home/ubuntu/workspace/xxx/logs/xxx' --port='6006'
1

点击(http://localhost:6006/)进入web端查看数据