search

Spenhouet / tensorboard-aggregator

Aggregate multiple tensorboard runs to new summary or csv files
MIT License
166 stars 27 forks source link

What if there are no subfolders test/train? #2

Closed ahinoampollack closed 5 years ago

ahinoampollack commented 5 years ago

Describe the bug if there are no subfolders then the code doesn't work. Get an error: /path/test is not a valid path --> since there is no test folder.

Spenhouet commented 5 years ago

Hi @ahinoampollack,

please try the parameter subpaths with the argument "['.']" (with or without outer quotes, not sure if os dependent).

python aggregator.py --subpaths "['.']"

Does that work?

ahinoampollack commented 5 years ago

Yes! it does.

Thank you so much for your time!

I now have a new error though (do you want me to start a new thread?):

For scalar val_loss the step numbering or count doesn't match. Step count for all runs: [600, 100, 187, 300, 300, 500, 400, 100, 400, 284, 600, 500, 300, 100, 500, 200, 200, 200, 200, 600, 400, 200, 600, 400, 400,

since not all of my runs have the same number of steps

Spenhouet commented 5 years ago

Sorry to hear that but this is a limitation of the implementation.

https://github.com/Spenhouet/tensorboard-aggregator#limitations

It is only possible to aggregate runs that have the same number of steps.

ahinoampollack commented 5 years ago

Ok! thanks. I'll use this as a starting code!

Spenhouet commented 5 years ago

I hope you get it to work 👍

If you do and the current functionality still works you could consider a pull request :)

EDIT: I forgot that this issue was raised in the past. There are reasons to not support it in this project. For the reasons take a look here: https://github.com/Spenhouet/tensorboard-aggregator/issues/1#issuecomment-461904929