15 | 可视化工具：如何实现训练的可视化监控？

viz = Visdom() viz.line([[0., 0.]], [0], win='train', opts=dict(title='loss_accuracy', legend=['loss', 'acc'])) for n_iter in range(10): loss = 0.2 * np.random.randn() + 1 accuracy = 0.1 * np.random.randn() + 1 viz.line([[loss, accuracy]], [n_iter], win='train', update='append') time.sleep(0.5)

运行下面指令报错: Error while finding module specification for 'visdom.server' python -m visdom.server 解决方案： Upgrade pip, python -m pip install --upgrade pip They try, python -m pip install visdom 再次运行最开始的命令即可。 ref: https://stackoverflow.com/questions/60623224/modulenotfounderror-no-module-named-visdom-when-importing-visdom

MacOS上运行tensorboard命令启动服务时，遇到以下错误： ValueError: Duplicate plugins for name projector 通过以下卸载重安装办法等到解决： pip uninstall tb-nightly tensorboardX tensorboard pip install tensorboard

站在巨人的肩膀上，发现Visdom 在chrome上运行正常，Safari 不动。 from visdom import Visdom import numpy as np import time # 将窗口类实例化 viz = Visdom() # 创建窗口并初始化 viz.line([[0., 0.]], [0], win='loss and accuracy', opts=dict(title='train_loss')) for n_iter in range(10): # 随机获取loss值 loss = 0.2 * np.random.randn() + 1 # 随机产生accuracy accuracy = 0.1 * np.random.randn() # 更新窗口图像,同时更新loss和accuracy viz.line(Y=[[loss, accuracy]], X=[n_iter], win='loss and accuracy', update='append') time.sleep(0.5)

老师您好，在tensorboard --logdir=runs 这一步出现了问题，求解 File "C:\Users\86189\AppData\Local\Temp/ipykernel_82416/59224237.py", line 1 tensorboard --logdir=runs ^ SyntaxError: cannot assign to operator

老师您好，为什么我在启动tensorboard后，运行程序以后报了如下错误呢，是什么原因造成的呢？ [WinError 10061] 由于目标计算机积极拒绝，无法连接。 on_close() takes 1 positional argument but 3 were given

老师，我的环境里有tensorboardx，可是为什么无法导入呢？报错如下： ModuleNotFoundError Traceback (most recent call last) Input In [7], in <module> ----> 1 from torch.utils.tensorboard import SummaryWriter 2 # PyTorch 1.8之前的版本请使用： 3 # from tensorboardX import SummaryWriter 4 import numpy as np File D:\Anaconda3\envs\PythonEnvironment1\lib\site-packages\torch\utils\tensorboard\__init__.py:1, in <module> ----> 1 import tensorboard 2 from setuptools import distutils 4 LooseVersion = distutils.version.LooseVersion ModuleNotFoundError: No module named 'tensorboard'

老师推荐使用哪种可视化工具了？

# 将窗口类实例化 viz = Visdom() # 创建窗口并初始化 viz.line([0.], [0], win='loss and accuracy', opts=dict(title='train_loss')) for n_iter in range(10): # 随机获取loss值 loss = 0.2 * np.random.randn() + 1 # 随机产生accuracy accuracy = 0.1 * np.random.randn() # 更新窗口图像,同时更新loss和accuracy viz.line(Y=[loss, accuracy], X=[n_iter], win='loss and accuracy', update='append') time.sleep(0.5)

学习打卡