I've generally noted that WSL really don't cut it as Linux substituent. Same goes for regular Windows, last time I tried installing Tensorflow it failed due to scikit not being available (because of no pre-built binaries or something). Sure you can install it with Anaconda but it's still clunky.

I highly recommend using Docker, mounting a volume to it and running all the Linux code inside the container. You can still modify the files both ways but avoid most of the messy incompatibility issues. There's some learning curve to it but I think it will pay off in the long run.