ENH: [typing] `NDarray[plt.Axes]` not possible - Type argument of "NDArray" must be a subtype of "generic" #24738

randolf-scholz · 2023-09-18T16:32:20Z

Proposed new feature or change:

For numpy.typing.NDArray, the argument needs to be a subtype of numpy.generic. However, this leads to problem since we can no longer specify concrete types for object-arrays, for example, mypy raises [type-var] for this code:

import matplotlib.pyplot as plt
from matplotlib.figure import Figure
from matplotlib.axes import Axes
from numpy.typing import NDArray

axes: NDArray[Axes]  # ✘ mypy: raises type-var
fig: Figure
fig, axes = plt.subplots(2, 2, sharex=True, sharey=True)

for ax in axes.flat:
    reveal_type(axes)  # Axes if the previous error is ignored, else ndarray
    ax.set_title("A subplot")

For NDArray of dtype np.object_, it should somehow wrap the given type. One approach could be to make np.object_ generic and change methods overloads so that for object-type, it generally returns the wrapped type.

The text was updated successfully, but these errors were encountered:

ksunden · 2023-09-18T17:20:53Z

In type hinting Matplotlib, I stopped short of specifying the type in subplots (Figure.subplots only says np.ndarray, pyplot.subplots actually just does Any)

In most cases adding that specificity was just too much to do all at once so I left it for follow up. I don't specifically recall trying to specify here, but it is possible I did and ran into this then.

BvB93 · 2023-09-20T22:09:49Z

For NDArray of dtype np.object_, it should somehow wrap the given type. One approach could be to make np.object_ generic and change methods overloads so that for object-type, it generally returns the wrapped type.

It's a feature on the wish list, but yeah this will in all likelihood be the most feasible and sensible approach here. I do foresee one limitation though that I sincerely doubt we'd be to handle satisfactory: type promotions between arrays with different object dtypes (e.g. an object array wrapping builtins.int and one wrapping builtins.float for example). Still, this shouldn't really be an issue as long as you're only mixing np.object_[plt.Axes] object arrays with each other for example.

randolf-scholz · 2023-11-29T12:16:49Z

Is there any workaround for this atm? I also run in this problem when trying to do NDArray[T] where T is a TypeVar bound to a custom Protocol type, which is very annoying...

jorenham · 2024-07-14T23:12:50Z

Note that numpy.typing.NDArray[T: numpy.generic] is an alias for numpy.ndarray[Any, numpy.dtype[T]].

The first type parameter of numpy.ndarray is meant to be used for shape typing.
Currently, it accepts any type, but that is likely to be changed in the future to tuple[int, ...], see #26081.

See #16544 for an in-depth discussion on array shape typing.

FYI, the optype typing library provides a type alias optype.numpy.Array, which similar to numpy.typing.NDArray, but has two type parameters: one for the shape, and one for the scalar type. See the documentation for examples and details.

Summary: visualize_image_attr_multiple can return a List[Axes], adds proper annotations to satisfy mypy Why casting is necessary: numpy/numpy#24738 https://github.com/matplotlib/matplotlib/blob/v3.9.2/lib/matplotlib/pyplot.py#L1583C41-L1584C1 Differential Revision: D64998799

Summary: Pull Request resolved: #1416 visualize_image_attr_multiple can return a List[Axes], adds proper annotations to satisfy mypy Why casting is necessary: numpy/numpy#24738 https://github.com/matplotlib/matplotlib/blob/v3.9.2/lib/matplotlib/pyplot.py#L1583C41-L1584C1 Reviewed By: cyrjano Differential Revision: D64998799 fbshipit-source-id: 7b2249654f5eb784f6c0686160f78f1be61bf7a2

BvB93 added 01 - Enhancement 23 - Wish List 41 - Static typing labels Sep 20, 2023

ksunden mentioned this issue Sep 21, 2023

[Bug/TYP]: misc type errors matplotlib/matplotlib#26858

Open

ksunden mentioned this issue Oct 5, 2023

[TYP] Add overload of pyplot.subplots matplotlib/matplotlib#27001

Merged

1 task

basnijholt mentioned this issue Aug 26, 2024

TYP: Make np.object_ generic for better object array type hinting #25351

Open

craymichael mentioned this issue Oct 25, 2024

Fix mypy issue in visualization.py pytorch/captum#1416

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

ENH: [typing] `NDarray[plt.Axes]` not possible - Type argument of "NDArray" must be a subtype of "generic" #24738

ENH: [typing] `NDarray[plt.Axes]` not possible - Type argument of "NDArray" must be a subtype of "generic" #24738

randolf-scholz commented Sep 18, 2023 •

edited

Loading

ksunden commented Sep 18, 2023

BvB93 commented Sep 20, 2023 •

edited

Loading

randolf-scholz commented Nov 29, 2023

jorenham commented Jul 14, 2024 •

edited

Loading

ENH: [typing] NDarray[plt.Axes] not possible - Type argument of "NDArray" must be a subtype of "generic" #24738

ENH: [typing] NDarray[plt.Axes] not possible - Type argument of "NDArray" must be a subtype of "generic" #24738

Comments

randolf-scholz commented Sep 18, 2023 • edited Loading

Proposed new feature or change:

ksunden commented Sep 18, 2023

BvB93 commented Sep 20, 2023 • edited Loading

randolf-scholz commented Nov 29, 2023

jorenham commented Jul 14, 2024 • edited Loading

ENH: [typing] `NDarray[plt.Axes]` not possible - Type argument of "NDArray" must be a subtype of "generic" #24738

ENH: [typing] `NDarray[plt.Axes]` not possible - Type argument of "NDArray" must be a subtype of "generic" #24738

randolf-scholz commented Sep 18, 2023 •

edited

Loading

BvB93 commented Sep 20, 2023 •

edited

Loading

jorenham commented Jul 14, 2024 •

edited

Loading