ImportError: No module named visualization.writer

[muzi-8]

I have installed the mxnet、minpy and tensorflow library, "Visualization with Tensorboard" code, when I run the

"from minpy.visualization.writer import LegacySummaryWriter as SummaryWriter import minpy.visualization.summaryOps as summaryOps" occurs a error, it's "ImportError: No module named visualization.writer" I'm confused. thanks!

[jermainewang]

We are changing the visualization from the official tensorboard to our own tensorboard https://github.com/dmlc/tensorboard . I guess it is broken right now. In the meanwhile, to workaround this. You could check out the repository I mentioned and try to directly use it for visualization.

[zihaolucky]

@muzi-8 our own tensorboard could be installed via pip install tensorboard, and you don't need TF for tensorboard if you use this.

@jermainewang anything I can do for Minpy?

[muzi-8]

@jermainewang Thanks for helping. because the method that comes from officail website is tedious and installation process is time-consuming.So I take the following approach. But all is not think well. @zihaolucky Thanks for your reply, I follow the method which you recommend via "pip install tensorboard ". when I successfully installed the tensorboard moudle,but run the tutorial example, that is "Understanding the vanishing gradient problem through visualization". Jump out of a error. as follows: run the code: "import sys sys.path.append('./mnist/') from train_mnist import *" Error: "ImportError Traceback (most recent call last) in () 1 import sys 2 sys.path.append('./mnist/') ----> 3 from train_mnist import *

/home/muzi/tensorboard/docs/tutorial/mnist/train_mnist.py in () 3 import argparse 4 import os, sys ----> 5 import train_model, train_model_monitor 6 7 def _download(data_dir):

ImportError: No module named train_model_monitor"

According to the information, train_model_monitor is nonexistent. And I don't konw this moudle role .

Looking forward to your help and answer. Thank you very much !

[zihaolucky]

@muzi-8 thanks for your feedback. Just remove the train_model_monitor, it should be okay, I'll remove the useless part later. BTW, the summary API might be slightly different from the TF's, but it's easy to use I think.

You're welcomed to give us feedback at https://github.com/dmlc/tensorboard/issues/19 if you have any issue after installing tensorboard via pip.

[jermainewang]

@zihaolucky Yes. Let me create a separate issue on possible improvements for dmlc/tensorboard and minpy. Thanks.

[zihaolucky]

@jermainewang Cool, @piiswrong and I are just about to bring dmlc/tensorboard to MXNet, https://github.com/dmlc/tensorboard/issues/10, your suggestions must help a lot.