package name conflict about tensorboardx

[edgarriba]

Thanks for the answer. The first solution looks good, however the second one should be the way to go since overlapping package names in general is a bad practice. Besides, if you want to give support to other frameworks you will have to rename the package because it won't be pytorch specific anymore.

[lanpa]

Hi, the point of this package is to make use of tensorboard's feature, instead of getting rid of TF. I think most researcher have both pytorch and tensorflow in a single machine. :) Additionally, to use function such as add_embedding, you still need tensorflow installed at this time.

And there will be a more generic version in the future (supports Theano, numpy), so I would like to have a module name without a specific postfix. Please suggest if you have good idea.

A easy workaround of conflict will be https://stackoverflow.com/questions/27532112/how-to-handle-python-packages-with-conflicting-names, or rename the tensorboard folder to tensorboard-pytorch folder and build by yourself.