I've experienced something similar with SQLAlchemy, because of the amount of auto-magical stuff it does. I solved it with `[TYPECHECK] generated-members=query`,

I've experienced something similar with SQLAlchemy, because of the amount of auto-magical stuff it does. I solved it with `[TYPECHECK] generated-members=query`, although this can be slow so blacklisting SQLAlchemy classes is still an option. NumPy's `__init__.py` code is a bit messy and manipulates `__all__` a lot [0], so I can understand why pylint (and other static code analysis tools) struggle with it. Yeah, it's frustrating hitting errors in third party modules, and it's tedious for one person initially to set up pylint correctly, but so worth it for a team. [0] https://github.com/numpy/numpy/blob/master/numpy/__init__.py#L183 https://github.com/numpy/numpy/blob/master/numpy/__init__.py...