Tensorflow not included in requirements

[michaelmior]

In the Python 3.8 requirements file, Tensorflow is not included. Furthermore, it doesn't appear that there's a version of Tensorflow that is compatible with all the other packages that are listed there. Switching to something like Pipenv or Poetry would likely make such things much easier to manage.

[alexandrosandre]

Same here. Looks like it was an unfinished attempt to support Python 3.8?!

[madelonhulsebos]

Thanks for sharing this issue! I hope you managed to make it work in the meantime, but I will look into this later. If you have a general solution, a PR would be welcome as well!

[michaelmior]

@madelonhulsebos I would strongly recommend making use of pipenv or Poetry. Primarily because it makes it easy to track the dependencies the project actually needs while still pinning the versions of all transitive dependencies. This makes it easy to update dependencies while still maintaining a consistent set of versions.

[madelonhulsebos]

Thanks @michaelmior, I manage the packages as in the requirements.txt with conda but did not test the 3.8 requirements. Will work on this, thanks!

[michaelmior]

My point is that requirements.txt includes all transitive dependencies which makes upgrading messy. For example absl-py is not directly required by Sherlock, but somewhere else down the chain of dependencies. There's no way to identify this case with just a requirements file which makes much more challenging to upgrade. I had this experience when trying to get Sherlock with Python 3.x. However, I'm happy to keep maintaining my own fork if the current approach works for you :)

[alexandrosandre]

My point is that requirements.txt includes all transitive dependencies which makes upgrading messy. For example absl-py is not directly required by Sherlock, but somewhere else down the chain of dependencies. There's no way to identify this case with just a requirements file which makes much more challenging to upgrade. I had this experience when trying to get Sherlock with Python 3.x. However, I'm happy to keep maintaining my own fork if the current approach works for you :)

https://github.com/mitmedialab/sherlock-project/pull/29/files Addresses this part.