在pytorch下使用tensorboard进行可视化训练

在pytorch下使用tensorboard进行可视化训练

1.安装

pip install tensorflow

pip install tensorboardX

2.相关代码

from tensorboardX import SummaryWriter

writer = SummaryWriter()

tags = ["train_loss", "train_acc", "val_loss", "val_acc", "learning_rate"]
writer.add_scalar(tags[0], train_loss, epoch)
writer.add_scalar(tags[1], train_acc, epoch)
writer.add_scalar(tags[2], val_loss, epoch)
writer.add_scalar(tags[3], val_acc, epoch)
writer.add_scalar(tags[4], optimizer.param_groups[0]["lr"], epoch)

3.可视化结果

windows下在终端输入tensorboard --logdir=./runs

linux下比较麻烦：

首先在shell中进入到项目路径，输入tensorboard --logdir=./runs，复制网址，如（http://localhost:6006/），复制后，右键单击你所用的账户，点击属性