TensorboardX Versions Save

• Add API for Custom scalar
• Add support for logging directly to S3
• Add support for Caffe2 graph
• Pytorch 1.0.0 JIT graph support (alpha-release)

• Supports tensorshape information and show node name in graph visualization. This requires pytorch==0.4
• Adds add_video function. You can play video clip now
  • You can use tensorboard-dumper to dump images.
• Better embedding function

People who use pytorch==0.3.1 should use tensorboardX==1.1 People who use pytorch==0.4 should use tensorboardX==1.2

Major Tensorboard features are ready: audio, scalar, distribution, graph, histogram, image, pr curve, projector, text (markdown).

Some important change since v0.6:

• The package name is changed from tensorboard to tensorboardX to prevent from name collision with official tensorboard. (which leads to import error, ...etc) The name tensorboardX means tensorboard for X. I hope this package can be used by other DL frameworks such as mxnet, chainer as well. This is achieved by wrapping an make_np() call to function arguments. In fact, you can log experiment if you use tensorflow's eager mode.

• Removes dependency for tensorflow and torchvision to make this package much neutral.

• For other changes, see the commit log or HISTORY.rst

All tensorboard functions are implemented.

more details on: https://medium.com/@dexterhuang/tensorboard-for-pytorch-201a228533c5