ENH: Add typing for np.random and Generator

[bashtage]

NumPy supplies partial typing support. Currently random's methods and Generator do not have typing information which results in errors in mypy.

Reproducing code example:

Use mypy on any project that uses np.random and you will see output like

linearmodels/tests/iv/results/simulated-test-data.py:16: error: Module 'numpy.random' has no attribute 'multivariate_normal'
linearmodels/tests/iv/results/simulated-test-data.py:16: error: Module 'numpy.random' has no attribute 'seed'
linearmodels/tests/iv/results/simulated-test-data.py:69: error: Module has no attribute "chisquare"
linearmodels/panel/utility.py:496: error: Name 'np.random.RandomState' is not defined
linearmodels/panel/utility.py:534: error: Module has no attribute "RandomState"; maybe "__RandomState_ctor"?

Numpy/Python version information:

1.19.1 3.7.7 (default, May 7 2020, 21:25:33)
[GCC 7.3.0]

[mattip]

Does the situation change on HEAD?

[bashtage]

edit: Remove linalg issues that seem to be spurious.

Random is about the same.

np.random issues

linearmodels\panel\utility.py:506: error: Name 'np.random.RandomState' is not defined
linearmodels\panel\utility.py:544: error: Module has no attribute "RandomState"
linearmodels\tests\iv\test_data.py:69: error: Module has no attribute "randn"
linearmodels\tests\iv\test_data.py:89: error: Module has no attribute "randn"
linearmodels\tests\iv\results\simulated-test-data.py:16: error: Module 'numpy.random' has no attribute 'multivariate_normal'
linearmodels\tests\iv\results\simulated-test-data.py:16: error: Module 'numpy.random' has no attribute 'seed'
linearmodels\tests\iv\results\simulated-test-data.py:69: error: Module has no attribute "chisquare"
linearmodels\tests\system\results\generate_data.py:20: error: Module has no attribute "seed"
linearmodels\tests\system\results\generate_data.py:23: error: Module has no attribute "random_sample"
linearmodels\tests\system\results\generate_data.py:27: error: Module has no attribute "random_sample"
linearmodels\tests\panel\_utility.py:3: error: Module 'numpy.random' has no attribute 'standard_normal'
linearmodels\tests\panel\results\generate-panel-data.py:9: error: Module has no attribute "seed"
linearmodels\tests\panel\results\generate-panel-data.py:11: error: Module has no attribute "randn"
linearmodels\tests\panel\results\generate-panel-data.py:14: error: Module has no attribute "randn"
linearmodels\tests\panel\results\generate-panel-data.py:17: error: Module has no attribute "randn"
linearmodels\tests\panel\results\generate-panel-data.py:18: error: Module has no attribute "chisquare"
linearmodels\tests\panel\results\generate-panel-data.py:49: error: Module has no attribute "choice"
linearmodels\tests\panel\results\generate-panel-data.py:57: error: Module has no attribute "choice"
linearmodels\tests\panel\test_panel_covariance.py:34: error: Module has no attribute "random_sample"
linearmodels\tests\panel\test_panel_covariance.py:35: error: Module has no attribute "random_sample"
linearmodels\tests\panel\test_panel_covariance.py:43: error: Module has no attribute "randint"
linearmodels\tests\panel\test_panel_covariance.py:44: error: Module has no attribute "randint"
linearmodels\tests\panel\test_panel_covariance.py:45: error: Module has no attribute "randint"

flatiter issues

linearmodels\panel\model.py:2145: error: Value of type "flatiter[ndarray]" is not indexable
linearmodels\panel\model.py:2145: error: Argument 1 to "__call__" of "ufunc" has incompatible type "flatiter[ndarray]"; expected "Union[bool, int, float, complex, _SupportsArray, Sequence[Any]]"

[BvB93]

#17104 should more or less fix the issue by adding placeholder stub files for all (untyped) sub-modules.
Note that all objects within aforementioned modules are typed as Any.

More generally speaking, there is a large list of to-be typed objects, (see #16546) and this is simply going to take quite a bit time.
If you'd be willing to help out then that would be very much appreciated.

[BvB93]

#17104 has been merged now; all (public) objects in np.random are currently annotated as Any.

While still very broad, this will silence all "Module has no attribute ..." errors.

[jinwuxia]

#17104 has been merged now; all (public) objects in np.random are currently annotated as Any.

While still very broad, this will silence all "Module has no attribute ..." errors.

I am not sure whether adding these Any types for silencing Mypy errors is a good practice?
My meaning is that type annotation aims to help understand code and static check type errors, but any types take no such role.

I just feel contradictory about this.
But I don't have any other ideas to quickly remove "Module has no attribute ..." errors.

[BvB93]

I am not sure whether adding these Any types for silencing Mypy errors is a good practice?

Considering the typing of numpy is very much a work in progress, I don't see any problem with this:
The Anys will naturally disappear as more annotations will be added over the course of time.

Improvements are of course possible and, in fact, would be very much welcome additions.

[bashtage]

I agree that this is the pragmatic approach that allows for improvement when people who rely on these modules have time to invest.

[BvB93]

Closed by #18433 and #18395.