0.0.26 - wandb sync raises exception when trying to sync TF event files

[douglasrizzo]

I have multiple TensorBoard event files, which were created with PyTorch. I am on Linux. When I run wandb sync and point to a TF event file, I get the error reported below. This is the error I got when using the current-gen wandb client. These are the versions of things I'm running.

• Manjaro Linux 20.0.3 Lysia
• Python 3.7.7
• wandb client 0.0.26
• tensorboard 2.2.2
• tensorflow 2.2.0
• torch 1.5.0

Exception in thread Thread-1:
Traceback (most recent call last):
  File "/home/dodo/.anaconda3/lib/python3.7/threading.py", line 926, in _bootstrap_inner
    self.run()
  File "/home/dodo/.anaconda3/lib/python3.7/site-packages/wandb/sync/sync.py", line 37, in run
    ds.open_for_scan(sync_item)
  File "/home/dodo/.anaconda3/lib/python3.7/site-packages/wandb/internal/datastore.py", line 98, in open_for_scan
    self._read_header()
  File "/home/dodo/.anaconda3/lib/python3.7/site-packages/wandb/internal/datastore.py", line 165, in _read_header
    raise Exception("Invalid header")
Exception: Invalid header

[raubitsj]

Thanks @douglasrizzo for the report.

Currently we only support syncing runs with the wandb sync command that were already executed with the wandb library integration but the files were stored locally. For instance, if you ran while not connected to the internet.

I would like track this bug as a feature request as I think this would be a useful feature to directly sync a TFevent file into wandb. We have the code to do this, we would just need to do more testing to support this.

[douglasrizzo]

Thanks for the quick reply. I read the documentation in this page, which said:

If you have existing experiments you would like to import into wandb, you can run wandb sync log_dir where log_dir is a local directory containing the tfevents files.

I interpreted that as wandb being able to import any TF event files, not just the ones generated with wandb integration. If you don't mind, I'll suggest a small pull request to the documentation to make that a little more clear.