I am trying to learn how to visualize layer activations in a convolutional neural network using MXNet and Tensorboard. When following the tutorial, however, I run into an error that seems to be related to numpy.
I followed this tutorial: https://medium.com/apache-mxnet/mxboard-mxnet-data-visualization-2eed6ae31d2c
When I am in the conda prompt and enter the code:
tensorboard --logdir=./logs --host=127.0.0.1 --port=8888
The following error occurs:
ModuleNotFoundError: No module named 'numpy.core._multiarray_umath'
ImportError: numpy.core.multiarray failed to import
I thought I found a solution on the web that claimed that I have installed an older version of numpy (1.14). Which is kind of ironic because MXNet runs witht this older version only. So I am kind of confused...
In general, however, I am totally lost how to visualize the layer activation as others do. So if you know a different way how to do it, I am glad for some help!
Thanks a lot!
It looks like tensorboard is expecting a different version of numpy to the one installed. You could try upgrading the version of numpy (but this could potentially break MXNet functionality), or downgrading the version of tensorboard. As a safe choice, give this a go:
pip install --upgrade tensorflow==1.6.0
pip install --upgrade tensorboard==1.6.0