30_visdom可视化、TensorboardX及其案例、安装visdom、使用visdom的案例

本是古典 何须时尚 2022-10-29 10:25 430阅读 0赞

1.25.visdom可视化

1.25.1.TensorboardX

Pytorch也能用的tensorboard，此外Pytorch还有visdom可视化。

首先上项目地址：https://github.com/lanpa/tensorboardX

1.25.2.安装

  1. (base) C:\Users\toto>pip install tensorboardX
  2. Collecting tensorboardX
  3.   Downloading tensorboardX-2.1-py2.py3-none-any.whl (308 kB)
  4.      |████████████████████████████████| 308 kB 726 kB/s
  5. Requirement already satisfied: six in d:\installed\anaconda3\lib\site-packages (from tensorboardX) (1.12.0)
  6. Requirement already satisfied: protobuf>=3.8.0 in d:\installed\anaconda3\lib\site-packages (from tensorboardX) (3.12.1)
  7. Requirement already satisfied: numpy in d:\installed\anaconda3\lib\site-packages (from tensorboardX) (1.20.1)
  8. Requirement already satisfied: setuptools in d:\installed\anaconda3\lib\site-packages (from protobuf>=3.8.0->tensorboardX) (41.4.0)
  9. Installing collected packages: tensorboardX
  10. Successfully installed tensorboardX-2.1
  12. (base) C:\Users\toto>

或者从源码中build

  1. pip install 'git+https://github.com/lanpa/tensorboardX'

您可以选择安装crc32c来加速。

  1. (base) C:\Users\toto>pip install crc32c
  2. Collecting crc32c
  3.   Downloading crc32c-2.2-cp37-cp37m-win_amd64.whl (31 kB)
  4. Installing collected packages: crc32c
  5. Successfully installed crc32c-2.2
  7. (base) C:\Users\toto>

从tensorboardX 2.1开始，你需要为add_audio()函数安装soundfile(200倍加速)。

  1. (base) C:\Users\toto>pip install soundfile
  2. Collecting soundfile
  3.   Downloading SoundFile-0.10.3.post1-py2.py3.cp26.cp27.cp32.cp33.cp34.cp35.cp36.pp27.pp32.pp33-none-win_amd64.whl (689 kB)
  4.      |████████████████████████████████| 689 kB 117 kB/s
  5. Requirement already satisfied: cffi>=1.0 in d:\installed\anaconda3\lib\site-packages (from soundfile) (1.12.3)
  6. Requirement already satisfied: pycparser in d:\installed\anaconda3\lib\site-packages (from cffi>=1.0->soundfile) (2.19)
  7. Installing collected packages: soundfile
  8. Successfully installed soundfile-0.10.3.post1
  10. (base) C:\Users\toto>

1.25.3.例子

克隆文件：https://github.com/lanpa/tensorboardX/tree/master/examples
运行demo脚本：python examples/demo.py
运行TensorBoard可以用：tensorboard —logdir runs

在下面可能运行：

  1. (base) C:\Users\toto>tensorboard --logdir runs
  2. 2021-02-12 13:35:13.345643: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library cudart64_101.dll
  3. Traceback (most recent call last):
  4. ValueError: Duplicate plugins for name projector

例子：

  1. # -*- coding: UTF-8 -*-
  3. import torch
  4. import torchvision.utils as vutils
  5. import numpy as np
  6. import torchvision.models as models
  7. from torchvision import datasets
  8. from tensorboardX import SummaryWriter
  10. resnet18 = models.resnet18(False)
  11. writer = SummaryWriter()
  12. sample_rate = 44100
  13. freqs = [262, 294, 330, 349, 392, 440, 440, 440, 440, 440, 440]
  15. for n_iter in range(100):
  17.     dummy_s1 = torch.rand(1)
  18.     dummy_s2 = torch.rand(1)
  19.     # data grouping by `slash`
  20.     writer.add_scalar('data/scalar1', dummy_s1[0], n_iter)
  21.     writer.add_scalar('data/scalar2', dummy_s2[0], n_iter)
  23.     writer.add_scalars('data/scalar_group', { 'xsinx': n_iter * np.sin(n_iter),
  24.                                              'xcosx': n_iter * np.cos(n_iter),
  25.                                              'arctanx': np.arctan(n_iter)}, n_iter)
  27.     dummy_img = torch.rand(32, 3, 64, 64)  # output from network
  28.     if n_iter % 10 == 0:
  29.         x = vutils.make_grid(dummy_img, normalize=True, scale_each=True)
  30.         writer.add_image('Image', x, n_iter)
  32.         dummy_audio = torch.zeros(sample_rate * 2)
  33.         for i in range(x.size(0)):
  34.             # amplitude of sound should in [-1, 1]
  35.             dummy_audio[i] = np.cos(freqs[n_iter // 10] * np.pi * float(i) / float(sample_rate))
  36.         writer.add_audio('myAudio', dummy_audio, n_iter, sample_rate=sample_rate)
  38.         writer.add_text('Text', 'text logged at step:' + str(n_iter), n_iter)
  40.         for name, param in resnet18.named_parameters():
  41.             writer.add_histogram(name, param.clone().cpu().data.numpy(), n_iter)
  43.         # needs tensorboard 0.4RC or later
  44.         writer.add_pr_curve('xoxo', np.random.randint(2, size=100), np.random.rand(100), n_iter)
  46. dataset = datasets.MNIST('mnist', train=False, download=True)
  47. images = dataset.test_data[:100].float()
  48. label = dataset.test_labels[:100]
  50. features = images.view(100, 784)
  51. writer.add_embedding(features, metadata=label, label_img=images.unsqueeze(1))
  53. # export scalar data to JSON for external processing
  54. writer.export_scalars_to_json("./all_scalars.json")
  55. writer.close()

1.25.4.安装visdom

  1. (base) C:\Users\toto>pip install visdom
  2. Collecting visdom
  3.   Downloading visdom-0.1.8.9.tar.gz (676 kB)
  4.      |████████████████████████████████| 676 kB 27 kB/s
  5. Requirement already satisfied: numpy>=1.8 in d:\installed\anaconda3\lib\site-packages (from visdom) (1.18.5)
  6. Requirement already satisfied: scipy in d:\installed\anaconda3\lib\site-packages (from visdom) (1.4.1)
  7. Requirement already satisfied: requests in d:\installed\anaconda3\lib\site-packages (from visdom) (2.22.0)
  8. Requirement already satisfied: tornado in d:\installed\anaconda3\lib\site-packages (from visdom) (6.0.3)
  9. Requirement already satisfied: pyzmq in d:\installed\anaconda3\lib\site-packages (from visdom) (18.1.0)
  10. Requirement already satisfied: six in d:\installed\anaconda3\lib\site-packages (from visdom) (1.15.0)
  11. Collecting jsonpatch
  12.   Downloading jsonpatch-1.28-py2.py3-none-any.whl (12 kB)
  13. Collecting torchfile
  14.   Downloading torchfile-0.1.0.tar.gz (5.2 kB)
  15. Collecting websocket-client
  16.   Downloading websocket_client-0.57.0-py2.py3-none-any.whl (200 kB)
  17.      |████████████████████████████████| 200 kB 20 kB/s
  18. Requirement already satisfied: pillow in d:\installed\anaconda3\lib\site-packages (from visdom) (6.2.0)
  19. Requirement already satisfied: chardet<3.1.0,>=3.0.2 in d:\installed\anaconda3\lib\site-packages (from requests->visdom) (3.0.4)
  20. Requirement already satisfied: urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1 in d:\installed\anaconda3\lib\site-packages (from requests->visdom) (1.24.2)
  21. Requirement already satisfied: idna<2.9,>=2.5 in d:\installed\anaconda3\lib\site-packages (from requests->visdom) (2.8)
  22. Requirement already satisfied: certifi>=2017.4.17 in d:\installed\anaconda3\lib\site-packages (from requests->visdom) (2019.9.11)
  23. Collecting jsonpointer>=1.9
  24.   Downloading jsonpointer-2.0-py2.py3-none-any.whl (7.6 kB)
  25. Building wheels for collected packages: visdom, torchfile
  26.   Building wheel for visdom (setup.py) ... done
  27.   Created wheel for visdom: filename=visdom-0.1.8.9-py3-none-any.whl size=655249 sha256=aa110012ea279783c86dcbc7778c90e8d4c021e5d9ac1a98f68c3134b7c52627
  28.   Stored in directory: c:\users\toto\appdata\local\pip\cache\wheels\2d\d1\9b\cde923274eac9cbb6ff0d8c7c72fe30a3da9095a38fd50bbf1
  29.   Building wheel for torchfile (setup.py) ... done
  30.   Created wheel for torchfile: filename=torchfile-0.1.0-py3-none-any.whl size=5711 sha256=ac722c7bbb17ae1c36bef8db4b20cd6978462f240d76c5b6b0a4dfc6719db2b4
  31.   Stored in directory: c:\users\toto\appdata\local\pip\cache\wheels\ac\5c\3a\a80e1c65880945c71fd833408cd1e9a8cb7e2f8f37620bb75b
  32. Successfully built visdom torchfile
  33. Installing collected packages: jsonpointer, jsonpatch, torchfile, websocket-client, visdom
  34. Successfully installed jsonpatch-1.28 jsonpointer-2.0 torchfile-0.1.0 visdom-0.1.8.9 websocket-client-0.57.0
  36. (base) C:\Users\toto>

1.25.5.后台运行visdom.server

  1. (base) C:\Users\toto>pip install visdom  -i https://pypi.tuna.tsinghua.edu.cn/simple
  2. Collecting visdom
  3.   Downloading visdom-0.1.8.9.tar.gz (676 kB)
  4.      |████████████████████████████████| 676 kB 27 kB/s
  5. Requirement already satisfied: numpy>=1.8 in d:\installed\anaconda3\lib\site-packages (from visdom) (1.18.5)
  6. Requirement already satisfied: scipy in d:\installed\anaconda3\lib\site-packages (from visdom) (1.4.1)
  7. Requirement already satisfied: requests in d:\installed\anaconda3\lib\site-packages (from visdom) (2.22.0)
  8. Requirement already satisfied: tornado in d:\installed\anaconda3\lib\site-packages (from visdom) (6.0.3)
  9. Requirement already satisfied: pyzmq in d:\installed\anaconda3\lib\site-packages (from visdom) (18.1.0)
  10. Requirement already satisfied: six in d:\installed\anaconda3\lib\site-packages (from visdom) (1.15.0)
  11. Collecting jsonpatch
  12.   Downloading jsonpatch-1.28-py2.py3-none-any.whl (12 kB)
  13. Collecting torchfile
  14.   Downloading torchfile-0.1.0.tar.gz (5.2 kB)
  15. Collecting websocket-client
  16. xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
  17. Installing collected packages: jsonpointer, jsonpatch, torchfile, websocket-client, visdom
  18. Successfully installed jsonpatch-1.28 jsonpointer-2.0 torchfile-0.1.0 visdom-0.1.8.9 websocket-client-0.57.0
  20. 安装并更新，不更新不能正常使用，会卡在download script。
  21. (base) C:\Users\toto> pip install --upgrade visdom
  23. (base) C:\Users\toto>python -m visdom.server
  24. D:\installed\Anaconda3\lib\site-packages\visdom\server.py:39: DeprecationWarning: zmq.eventloop.ioloop is deprecated in pyzmq 17. pyzmq now works with default tornado and asyncio eventloops.
  25.   ioloop.install()  # Needs to happen before any tornado imports!
  26. Checking for scripts.

如果在启动过程中一直报：Downloading scripts, this may take a little while
一直卡着不动，解决办法是：

  • 找到visdom安装位置，我的路径为D:\Anaconda3\Lib\site-packages\visdom，打开sever.py。

  • 修改server.py文件，在函数download_scripts_and_run下注释掉download_scripts()，第1917行。
    在这里插入图片描述
    然后再执行，出现如下错误：

    (base) C:\Users\toto>python -m visdom.server
    D:\installed\anaconda3\lib\site-packages\visdom\server.py:39: DeprecationWarning: zmq.eventloop.ioloop is deprecated in pyzmq 17. pyzmq now works with default tornado and asyncio eventloops.
    ioloop.install() # Needs to happen before any tornado imports!
    It’s Alive!
    Traceback (most recent call last):
    File “D:\installed\anaconda3\lib\runpy.py”, line 194, in _run_module_as_main

    1. return _run_code(code, main_globals, None,

    File “D:\installed\anaconda3\lib\runpy.py”, line 87, in _run_code

    1. exec(code, run_globals)

    File “D:\installed\anaconda3\lib\site-packages\visdom\server.py”, line 1922, in 

    1. download_scripts_and_run()

    File “D:\installed\anaconda3\lib\site-packages\visdom\server.py”, line 1918, in download_scripts_and_run

    1. main()

    File “D:\installed\anaconda3\lib\site-packages\visdom\server.py”, line 1910, in main

    1. start_server(port=FLAGS.port, hostname=FLAGS.hostname, base_url=base_url,

    File “D:\installed\anaconda3\lib\site-packages\visdom\server.py”, line 1791, in start_server

    1. app.listen(port, max_buffer_size=1024 ** 3)

    File “D:\installed\anaconda3\lib\site-packages\tornado\web.py”, line 2116, in listen

    1. server.listen(port, address)

    File “D:\installed\anaconda3\lib\site-packages\tornado\tcpserver.py”, line 152, in listen

    1. self.add_sockets(sockets)

    File “D:\installed\anaconda3\lib\site-packages\tornado\tcpserver.py”, line 165, in add_sockets

    1. self._handlers[sock.fileno()] = add_accept_handler(

    File “D:\installed\anaconda3\lib\site-packages\tornado\netutil.py”, line 279, in add_accept_handler

    1. io_loop.add_handler(sock, accept_handler, IOLoop.READ)

    File “D:\installed\anaconda3\lib\site-packages\tornado\platform\asyncio.py”, line 100, in add_handler

    1. self.asyncio_loop.add_reader(fd, self._handle_events, fd, IOLoop.READ)

    File “D:\installed\anaconda3\lib\asyncio\events.py”, line 501, in add_reader

    1. raise NotImplementedError

    NotImplementedError

    (base) C:\Users\toto>

大概的意思Python3.8的asyncio改变了循环方式，因为这种方式在windows上不支持相应的add_reader APIs，就会抛NotImplementedError。

解决办法是： 找到这个项目使用的python环境的lib\site-packages,做下面的修改，在\Lib\site-packages\tornado\platform\asyncio.py开头添加代码：

  1. import sys            #此处如果已经引入过了，不用再引入了。
  3. if sys.platform == 'win32':
  4.     asyncio.set_event_loop_policy(asyncio.WindowsSelectorEventLoopPolicy())

在这里插入图片描述
然后在执行命令：
在这里插入图片描述
如果期间出现：

  1. Visdom python client failed to establish socket to get messages from the server. This feature is optional 
  2. and can be disabled by initializing Visdom with `use_incoming_socket=False`, which will prevent waiting for this request to timeout.

解决办法：

  1. vis = visdom.Visdom(use_incoming_socket=False)

示例：

  1. 在这里插入代码片

lines: single trace

  1. from visdom import Visdom
  3. viz = Visdom()
  5. viz.line([0.],[0.], win='train_loss', opts=dict(title='train loss'))
  7. viz.line([loss.item()], [global_step], win='train_loss', update='append')

在这里插入图片描述

lines: multi-traces

  1. from visdom import Visdom
  3. viz = Visdom()
  5. viz.line([[0.0, 0.0]], [0.], win='test', opts=dict(title='test loss&acc.',
  6.                                                    legend=['loss', 'acc.']))
  8. viz.line([[test_loss, correct / len(test_loader.dataset)]],
  9.          [global_step], win='test', update='append')

在这里插入图片描述

visual X

  1. from visdom import Visdom
  3. viz = Visdom()
  5. viz.images(data.view(-1, 1, 28, 28), win='x')
  7. viz.text(str(pred.detach().cpu().numpy()), win='pred', opts=dict(title='pred'))

在这里插入图片描述

1.25.6.使用visdom的案例

  1. import  torch
  2. import  torch.nn as nn
  3. import  torch.nn.functional as F
  4. import  torch.optim as optim
  5. from    torchvision import datasets, transforms
  7. from visdom import Visdom
  9. batch_size=200
  10. learning_rate=0.01
  11. epochs=10
  13. train_loader = torch.utils.data.DataLoader(
  14.     datasets.MNIST('../data', train=True, download=True,
  15.                    transform=transforms.Compose([
  16.                        transforms.ToTensor(),
  17.                        # transforms.Normalize((0.1307,), (0.3081,))
  18.                    ])),
  19.     batch_size=batch_size, shuffle=True)
  20. test_loader = torch.utils.data.DataLoader(
  21.     datasets.MNIST('../data', train=False, transform=transforms.Compose([
  22.         transforms.ToTensor(),
  23.         # transforms.Normalize((0.1307,), (0.3081,))
  24.     ])),
  25.     batch_size=batch_size, shuffle=True)
  28. class MLP(nn.Module):
  30.     def __init__(self):
  31.         super(MLP, self).__init__()
  33.         self.model = nn.Sequential(
  34.             nn.Linear(784, 200),
  35.             nn.LeakyReLU(inplace=True),
  36.             nn.Linear(200, 200),
  37.             nn.LeakyReLU(inplace=True),
  38.             nn.Linear(200, 10),
  39.             nn.LeakyReLU(inplace=True),
  40.         )
  42.     def forward(self, x):
  43.         x = self.model(x)
  45.         return x
  47. device = torch.device('cuda:0')
  48. net = MLP().to(device)
  49. optimizer = optim.SGD(net.parameters(), lr=learning_rate)
  50. criteon = nn.CrossEntropyLoss().to(device)
  52. viz = Visdom()
  54. viz.line([0.], [0.], win='train_loss', opts=dict(title='train loss'))
  55. viz.line([[0.0, 0.0]], [0.], win='test', opts=dict(title='test loss&acc.',
  56.                                                    legend=['loss', 'acc.']))
  57. global_step = 0
  59. for epoch in range(epochs):
  61.     for batch_idx, (data, target) in enumerate(train_loader):
  62.         data = data.view(-1, 28*28)
  63.         data, target = data.to(device), target.cuda()
  65.         logits = net(data)
  66.         loss = criteon(logits, target)
  68.         optimizer.zero_grad()
  69.         loss.backward()
  70.         # print(w1.grad.norm(), w2.grad.norm())
  71.         optimizer.step()
  73.         global_step += 1
  74.         viz.line([loss.item()], [global_step], win='train_loss', update='append')
  76.         if batch_idx % 100 == 0:
  77.             print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
  78.                 epoch, batch_idx * len(data), len(train_loader.dataset),
  79.                        100. * batch_idx / len(train_loader), loss.item()))
  82.     test_loss = 0
  83.     correct = 0
  84.     for data, target in test_loader:
  85.         data = data.view(-1, 28 * 28)
  86.         data, target = data.to(device), target.cuda()
  87.         logits = net(data)
  88.         test_loss += criteon(logits, target).item()
  90.         pred = logits.argmax(dim=1)
  91.         correct += pred.eq(target).float().sum().item()
  93.     viz.line([[test_loss, correct / len(test_loader.dataset)]],
  94.              [global_step], win='test', update='append')
  95.     viz.images(data.view(-1, 1, 28, 28), win='x')
  96.     viz.text(str(pred.detach().cpu().numpy()), win='pred',
  97.              opts=dict(title='pred'))
  99.     test_loss /= len(test_loader.dataset)
  100.     print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
  101.         test_loss, correct, len(test_loader.dataset),
  102.         100. * correct / len(test_loader.dataset)))

发表评论

表情：
评论列表 （有 0 条评论，430人围观）

还没有评论，来说两句吧...

相关阅读