TensorFlow 2: TensorBoard import error

wandb --version && python --version && uname

• Weights and Biases version: 0.9.3
• Python version: 3.6.9
• Operating System: Linux

Description

I am trying to sync the client with TensorBoard in a Docker container with the TF2.2 base image. However, the client is unable to import TensorBoard, although I can import this package just fine. Related issue.

What I Did

I have created a minimal repository to reproduce the issue.

git clone https://github.com/pascal-hwky/wandb-tb-bug.git
cd wandb-tb-bug
./run.sh

Sample output:

wandb: Dry run mode, not syncing to the cloud.
wandb: Tracking run with wandb version 0.9.3

...

wandb: ERROR Couldn't import tensorboard, not streaming events. Run `pip install tensorboard`
wandb: Program ended successfully.
wandb: You can sync this run to the cloud by running: 
wandb: wandb sync wandb/dryrun-20200716_123910-351imp5c

Any help is appreciated, as I cannot currently use my favorite experiment tracking framework with the Object Detection API. Running in Docker is a requirement for me, I cannot use virtual environments as suggested in the previous issue.

[vanpelt]

Apologies for the delay. The issue is being caused by sys.path issue messing up our ability to import tensorboard. We’ve fixed this in our latest library. You can install it with pip install wandb -U --pre. I’ve verified this works in the example docker image provided.

[tyomhak]

Hi there, thank you for reporting. I’m filing an internal ticket for this. We’ll look into this and get back to you.