Numpy deprecations

[sbrugman]

Numpy deprecated np.bool and other types in 1.20, which expired in the latest release 1.24.

Deprecated name	Identical to	NumPy scalar type names
numpy.bool	bool	numpy.bool_
numpy.int	int	numpy.int_ (default), numpy.int64, or numpy.int32
numpy.float	float	numpy.float64, numpy.float_, numpy.double (equivalent)
numpy.complex	complex	numpy.complex128, numpy.complex_, numpy.cdouble(equivalent)
numpy.object	object	numpy.object_
numpy.str	str	numpy.str_
numpy.long	int	numpy.int_ (C long), numpy.longlong (largest integer type)
numpy.unicode	str	numpy.unicode_

Would be great to be able to automatically flag and fix this. Seen it pass by in multiple projects:
apache/spark#37817
pandas-dev/pandas#39019
dylan-profiler/visions#184
plotly/plotly.py#4010
https://stackoverflow.com/questions/67779374/deprecationwarning-np-bool

[charliermarsh]

It might be reasonable to add these as an UP rule, since those codes are made up anyway given that pyupgrade isn't actually a Flake8 plugin.

[charliermarsh]

Although it might be incorrect to "market" rules as part of pyupgrade when they're in fact not.

[ngnpope]

Use UPG for "upgrade"? Or UPN for "upgrade numpy"?

[charliermarsh]

For now, if we did something like this, it's probably safest to put it in RUF or add a new category like RX that requires users to turn them on one-by-one.

[sbrugman]

For the future effort of reorganizing rules, we might group them per dependency (if any). Now we have pytest and pandas. For now I'd opt for flake8-numpy or just numpy with NPY. (This allows to add all future rules related to the package there too). RUF is also fine of course.

[charliermarsh]

NPY is reasonable too, yeah.

[ngnpope]

NPY makes so much sense. Good shout.

[sbrugman]

Also relevant for mlflow/mlflow#4048