wkirgsn/thermal-nn

Aux is no valid file name on Windows

Closed this issue · 9 comments

reserschnell commented

Hi Wilhelm,
unfortunately, I cannot checkout your repo on my Windows machine. I'm pretty sure that the issue occurs because you have a folder called aux.
The issue is also described here:
https://stackoverflow.com/questions/61614004/cloning-succeded-but-checkout-failed-due-to-invalid-path-what-is-the-path-probl
Can you maybe rename the folder?

Regards
Andreas

wkirgsn commented

Hey, sorry for getting back to you this late.
I will change that no problem.

I do not cease from being surprised by what weird rules Windows applies to its users.

wkirgsn commented

Just renamed aux to tf2utils, it should work out now, can you confirm?

reserschnell commented

Hi, I've cloned the changes and cloning was working but the notebooks where not running for me. I'll provide a pull request next week and you can check my changes.

  • 👍1
reserschnell commented

Which Python or tensorflow version are you using?
I've tried to use Python 3.11.5 with tensorflow 2.14.0, but it seems that tensorflow.keras.wrappers.scikit_learn is removed in this version.
You can get your tensorflow version with:
pip show tensorflow

wkirgsn commented

it was probably TF 2.1.x, due to the lower bound in 'requirements.txt'.
I was not maintaining the TF version of the code, I have to admit, since I switched to Pytorch.
The latter is way more compact and concise.
I am open to PRs that make the TNN implementation work also for the current tensorflow versions.
However, it might be better to lean on the pytorch implementation while reimplementing the tf code, since I am not sure, whether a TNN implementation in TF still needs to be this complex.

nMaroulis commented

@reserschnell I had the same issue and solved that but installing the scikeras library and replacing the import with 'from scikeras.wrappers import KerasRegressor' and it works fine.

wkirgsn commented

I guess I will look into implementing a modern Tensorflow implementation of the TNN, and then lock the version used in a requirements.txt

reserschnell commented

@nMaroulis thank you for the hint, I can give it a try.
@wkirgsn: It would be really great if you can create a consistent state of the repository. Maybe you can mark it then as a release, that everyone knows that this is a version everyone can start from.

  • 👍1
wkirgsn commented

@reserschnell , @nMaroulis Just published a new initial release, with the tensorflow notebook drastically reduced to a minimal working example, in the same style as the pytorch example.

Closing this issue for now, but can be reopened in case of reoccurring issues