无法让 pytorch 与 tensorboard 一起工作

Question

我正在通过本教程使用 tensorboard 设置 pytorch（v1.3.0 到 conda）https://pytorch.org/tutorials/intermediate/tensorboard_tutorial.html#

但是在台阶上

from torch.utils.tensorboard import SummaryWriter

# default `log_dir` is "runs" - we'll be more specific here
writer = SummaryWriter('runs/fashion_mnist_experiment_1')

我一直收到错误

---------------------------------------------------------------------------
ModuleNotFoundError                       Traceback (most recent call last)
C:\ProgramData\Anaconda3\envs\fastai_v1\lib\site-packages\torch\utils\tensorboard\__init__.py in 
      1 try:
----> 2     from tensorboard.summary.writer.record_writer import RecordWriter  # noqa F401
      3 except ImportError:

ModuleNotFoundError: No module named 'tensorboard.summary'; 'tensorboard' is not a package

During handling of the above exception, another exception occurred:

ImportError                               Traceback (most recent call last)
c:\Users\matt\Documents\code\playground\tensorboard.py in 
----> 1 from torch.utils.tensorboard import SummaryWriter
      2 
      3 # default `log_dir` is "runs" - we'll be more specific here
      4 writer = SummaryWriter('runs/fashion_mnist_experiment_1')

C:\ProgramData\Anaconda3\envs\fastai_v1\lib\site-packages\torch\utils\tensorboard\__init__.py in 
      2     from tensorboard.summary.writer.record_writer import RecordWriter  # noqa F401
      3 except ImportError:
----> 4     raise ImportError('TensorBoard logging requires TensorBoard with Python summary writer installed. '
      5                       'This should be available in 1.14 or above.')
      6 from .writer import FileWriter, SummaryWriter  # noqa F401

ImportError: TensorBoard logging requires TensorBoard with Python summary writer installed. This should be available in 1.14 or above.

有人有什么建议吗？

Answer 1

我遇到了同样的错误，我通过以下步骤解决了：

从 conda 环境中删除了所有 Tensorflow 或 Tensorboard 安装。
然后通过激活相同的 conda 环境，输入 "pip install -U tb-nightly"
然后输入，"pip install -U future"
完成

Answer 2

我觉得是版本问题

就运行这个：

pip install tensorboard==1.14.0

(不是pip install tensorboard==1.14)

或者只安装tensoflow 1.14.0，其中包含tensorboard 1.14.0:

pip install tensorflow==1.14.0

这个版本的 tensorflow 在 pytorch 1.2 中对我有用。

Answer 3

错误日志说，除其他外，

ImportError: TensorBoard logging requires TensorBoard with Python summary writer installed. This should be available in 1.14 or above.

因此，当它尝试导入 TensorBoard 时，它无法这样做，因为它在搜索路径中丢失了它。您可以安装最新版本（无需指定任何版本号），如：

$ conda install -c conda-forge tensorboard

除此之外，您可能还需要安装 protobuf：

$ conda install -c conda-forge protobuf

这些安装应该修复 ImportErrors。

Answer 4

我完成了：

conda install -y tensorboard

之前没有问题，所以我不确定为什么那行不通。最简单了。

Answer 5

我遇到了同样的错误，那次我的conda也没有运行，所以我选择使用tensorboardX，它和tensorboard几乎完全一样（也是它的操作） .
只需使用 pip install tensorboardX
安装即可您可以使用 from tensorboardX import SummaryWriter

导入它

无法让 pytorch 与 tensorboard 一起工作

Can not get pytorch working with tensorboard

python

data-visualization

machine-learning

tensorboard

pytorch