Make tensorflow optional when using tensorboard

[charlesbvll]

Issue

Description

Currently when using tensorboard, the app will crash in an unhelpful way if tensorflow is not installed.

Related issues/PRs

N/A

Proposal

Explanation

If tensorflow is not installed we display a warning to the user, and instead use the regular strategy that was passed to the wrapper so the execution can continue.

Checklist

• Implement proposed change
• Make CI checks pass
• Ping maintainers on Slack (channel #contributions)

Any other comments?

N/A

[jafermarq]

afaik tenforflow is not needed to be installed to use Tensorboard. On a clean env i can do just pip install tensorboard and the following will work just fine:

import tensorboard as tb
w = tb.summary.Writer('test') # creates dir `test` in current path
w.add_scalar(data=0.99, tag='meh', step=0)
w.add_scalar(data=1.99, tag='meh', step=1)
w.flush()
w.close()
quit()

Then i can launch the dashboard and see the two datapoints: tensorboard --logdir test

Maybe some functionalities do indeed require support from TF? (or PyTorch, which also can interface w/ Tensorboard)?

[charlesbvll]

@jafermarq We actually don't use Tensorboard directly in the code, we only use tensorflow.summary , hence why I was doing the check on Tensorflow. Maybe the code could be rewritten without tensorflow, but I'm not sure...