Chart fullscreen and back causes weird scaling bug

[mlavrent]

When using Tensorboard, and following the following steps, the bug can be reproduced:

1. Log scalar data to Tensorboard
2. Open the chart
3. Set it to fullscreen
4. Unset the fullscreen

When this is done, the following can be seen:

As a note, this is doesn't get fixed with the built-in update data button. It does get fixed if you refresh the page.

My setup:

• Standard pip install tensorflow
• Windows 10
• Firefox 54

[chihuahua]

I am trying to repro on FF 54 for Ubuntu. Which of these 2 buttons did you click to full screen? Thank you!

[mlavrent]

I used the one on the left. I don't think the one on the right is even present with what I'm using. My tensorflow version is 1.2.1. Also, my mistake, I'm using firefox 54, sorry.

[wchargin]

From your UI, it looks like you're on a pretty old version of TensorBoard. We fixed similar issues in #190, and I can't reproduce this.

When you use the latest version of TensorBoard from master, do you still see the issue?

[mlavrent]

How would I use the latest version from master? I'm pretty new to using this, so I've just been using the one that gets installed through pip.

[wchargin]

1. Install Bazel: https://bazel.build/
2. Install TensorFlow nightly (in a virtualenv): https://github.com/tensorflow/tensorflow#installation
3. Clone this repository, then write bazel run //tensorboard -- --logdir /tmp/data in your terminal.

[mlavrent]

So should I try this in my Ubuntu 16 vm? Or will it work on Windows 10?

[wchargin]

In theory, everything should work under Windows, but I expect that everything will be much simpler if you use a different operating system. (For instance, Bazel's Windows support is "experimental," and you certainly need Bazel.)

[mlavrent]

I'll try it in Ubuntu 16 and let you know.