pytorch visdom安装开启及使用方法

安装

conda activate ps
pip install visdom

激活ps的环境，在指定的ps环境中安装visdom

开启

python -m visdom.server

浏览器输入红框内的网址

使用

1. 简单示例：一条线

from visdom import Visdom

# 创建一个实例
viz=Visdom()

# 创建一个直线，再把最新数据添加到直线上
# y x二维两个轴，win 创建一个小窗口，不指定就默认为大窗口，opts其他信息比如名称
viz.line([1,2,3,4],[1,2,3,4],win="train_loss",opts=dict(title='train_loss'))

# 更一般的情况,因为下面y x数据不存在，只是示例
#  append 添加到原来的后面，不然全部覆盖掉
# viz.line([loss.item()],[global_step],win="train_loss",update='append')

2. 简单示例：2条线

下面主要是[[y1],[y2]],[x] 两条映射，legend就是线条名称

from visdom import Visdom
viz=Visdom()
viz.line([[1,2],[5,6]],[1,2],win="loss_acc",opts=dict(title='train loss & acc',legend=['loss','acc']))

3. 显示图片

from visdom import Visdom
viz=Visdom()
# data 是一个batch
viz.image(data.view(-1,1,28,28),win='x')
viz.text(str(pred.datach().cpu().numpy()),win='pred',opts=dict(title='pred'))

4. 手写数字示例

动画效果图如下

import  torch
import  torch.nn as nn
import  torch.nn.functional as F
import  torch.optim as optim
from    torchvision import datasets, transforms

from visdom import Visdom

batch_size=200
learning_rate=0.01
epochs=10

train_loader = torch.utils.data.DataLoader(
   datasets.MNIST('../data', train=True, download=True,
                  transform=transforms.Compose([
                      transforms.ToTensor(),
                      # transforms.Normalize((0.1307,), (0.3081,))
                  ])),
   batch_size=batch_size, shuffle=True)
test_loader = torch.utils.data.DataLoader(
   datasets.MNIST('../data', train=False, transform=transforms.Compose([
       transforms.ToTensor(),
       # transforms.Normalize((0.1307,), (0.3081,))
   ])),
   batch_size=batch_size, shuffle=True)

class MLP(nn.Module):

def __init__(self):
       super(MLP, self).__init__()

self.model = nn.Sequential(
           nn.Linear(784, 200),
           nn.LeakyReLU(inplace=True),
           nn.Linear(200, 200),
           nn.LeakyReLU(inplace=True),
           nn.Linear(200, 10),
           nn.LeakyReLU(inplace=True),
       )

def forward(self, x):
       x = self.model(x)

return x

device = torch.device('cuda:0')
net = MLP().to(device)
optimizer = optim.SGD(net.parameters(), lr=learning_rate)
criteon = nn.CrossEntropyLoss().to(device)

viz = Visdom()

viz.line([0.], [0.], win='train_loss', opts=dict(title='train loss'))
viz.line([[0.0, 0.0]], [0.], win='test', opts=dict(title='test loss&acc.',
                                                  legend=['loss', 'acc.']))
global_step = 0

for epoch in range(epochs):

for batch_idx, (data, target) in enumerate(train_loader):
       data = data.view(-1, 28*28)
       data, target = data.to(device), target.cuda()

logits = net(data)
       loss = criteon(logits, target)

optimizer.zero_grad()
       loss.backward()
       # print(w1.grad.norm(), w2.grad.norm())
       optimizer.step()

global_step += 1
       viz.line([loss.item()], [global_step], win='train_loss', update='append')

if batch_idx ％ 100 == 0:
           print('Train Epoch: {} [{}/{} ({:.0f}％)]\tLoss: {:.6f}'.format(
               epoch, batch_idx * len(data), len(train_loader.dataset),
                      100. * batch_idx / len(train_loader), loss.item()))

test_loss = 0
   correct = 0
   for data, target in test_loader:
       data = data.view(-1, 28 * 28)
       data, target = data.to(device), target.cuda()
       logits = net(data)
       test_loss += criteon(logits, target).item()

pred = logits.argmax(dim=1)
       correct += pred.eq(target).float().sum().item()

viz.line([[test_loss, correct / len(test_loader.dataset)]],
            [global_step], win='test', update='append')
   viz.images(data.view(-1, 1, 28, 28), win='x')
   viz.text(str(pred.detach().cpu().numpy()), win='pred',
            opts=dict(title='pred'))

test_loss /= len(test_loader.dataset)
   print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}％)\n'.format(
       test_loss, correct, len(test_loader.dataset),
       100. * correct / len(test_loader.dataset)))

来源：https://blog.csdn.net/qq_42146775/article/details/115837929