Paths to Python modules - how to do it right?

Question

Python

I am new to Python, and with writing my own modules and connecting third-party code, several questions immediately arose.
1) Why in Flask, even if I refer to nested modules, the module name still comes from the root?
For example, at the root is the file start.py, from which I call the function located in logic/mylogic.py. Also in the logic folder there is another folder, and in it are some of my modules, for example a.py. Regarding mylogic.py, one could do import a, but only import logic.a works, even if the call comes from mylogic.py, which is already in the logic folder.
2) If on the Internet, on github, for example, I come across some lib or someone else's code, and I want to use it, and this code inside refers to its own internal modules, then I have to put it only in the root. As soon as I move it to some other subfolder, internal modules stop working, since all their imports again come from the root (same problem as in the previous question). Of course, it would be possible to rename all imports, but in some cases (for example, some tensoflow examples have thousands or more files) this is wildly inconvenient, maybe there is some normal way to avoid depending on the specific folder where the module is located?
Thank you!

Answer 1

In the folder where the imported module is located, you need to create an empty module __init__.py
Well, of course, if there are many sub-folders, then you need to create each one .
Inside the folders, a relative import
from is done. import my_module

Answer 2

If on the Internet, on github, for example, I come across some lib or someone else's code, and I want to use it, and this code inside refers to its own internal modules, then I have to put it only in the root. As soon as I move it to some other subfolder, internal modules stop working, since all their imports again come from the root (same problem as in the previous question). Of course, it would be possible to rename all imports, but in some cases (for example, some tensoflow examples have thousands or more files) this is wildly inconvenient, maybe there is some normal way to avoid depending on the specific folder where the module is located?

No need to put someone else's code anywhere. You need to use virtualenv (pipenv, conda, poetry) and install someone else's code with the pip install tensorflow. It's the same with all other libraries.