The Array object can take many concrete forms. It may be a single Array (1D) of booleans, or a three-digit list of unique prices. As a built-in work isinstance() will indicate, every Array is an example of np.ndarrayregardless of make-up or type of items stored in Array, Ie, the dtype. Similarly, many types described in type are just as saying np.ndarray:
import numpy as np
def process(
x: np.ndarray,
y: np.ndarray,
) -> np.ndarray: ...Such types are not enough: Many communication areas have a solid expectation of status or dtype of the previous array. A lot of code will fail if the 3D list is expected when the Array 1d is expected, or a list of many days is transferred where it is expected to have other lists.
To fully implement generic generic np.ndarraythe formation of jokes and dtype Features now can be fully defined:
def process(
x: np.ndarray[tuple[int], np.dtype[np.bool_]],
y: np.ndarray[tuple[int, int, int], np.dtype[np.uint8]],
) -> np.ndarray[tuple[int], np.dtype[np.float64]]: ...For such information, the latest types of static evaluation tools like mypy including pyright You can find issues before the code runs. In addition, Run-Time are special Tutpurotors, such as Staticframe's sf.CallGuardcan re-use the same verification of verification period.
Generic species in Python
The normal built-in containers such as list including dict It can be made concrete with the clarification, in each interface, the types contained. Work can reveal that it takes a list of str reference list[str]; or a dict of str above bool can be specified with dict[str, bool].
Generic np.ndarray
A np.ndarray in-okinin a list of single item of item (or dtype). This page np.ndarray Generic takes boundaries of two type: First means formation with tupleThe second describes some kind of generic generic np.dtype. There np.ndarray Took two types of type, the definition of the first parameter, formation, not fully specified until the benefit of 2.1.
SHAPE type parameter
When you create a list of interface like np.empty or np.fullThe issue is provided as repair. Tuple length means the size of this list; The size of each position means the size of that kind. So is the formation (10,) Is 1D list of 10 items; Format (10, 100, 1000) Is a list of 10 size 100 in 1000.
When using a tuple to describe the formation in np.ndarray Generic, currently only the number of variety that can be used to view the type. Therefore, a tuple[int] can specify 1D list; a tuple[int, int, int] can specify a list of 3D; a tuple[int, ...]Specifies zero tendencies or more, says list of three. Possible for the future to write-check an np.ndarray in certain measurements for each (using Literal) But this is not widely supported.
This page dtype Type a parameter
INCYPY dtype The item describes the types of features and, of some types, other features such as size (unicode and forms of disputes) or unit (of np.datetime64 Types). This page dtype Itself generic, takes part “generic” type as parameter. The smallest types specify certain features of the item, for example np.uint8, np.float64or np.bool_. Through these smaller species, the Income provides other common types, such as np.integer, np.inexactor np.number.
Live np.ndarray Concrete
The following examples indicate concrete np.ndarray Definitions:
1D Booleans List:
np.ndarray[tuple[int], np.dtype[np.bool_]]3D Review List of 8 Revenue:
np.ndarray[tuple[int, int, int], np.dtype[np.uint8]]The two-sided Array (2D) of Unicode wires:
np.ndarray[tuple[int, int], np.dtype[np.str_]]1d list of any form of numbers:
np.ndarray[tuple[int], np.dtype[np.number]]Static type checks with mypy
As soon as generic np.ndarray made of concrete, mypy Or the same same selected checks can, find some code methods, point to the amounts that do not match the display.
For example, the work below requires 1D list of signs signed. As shown below, unsubscribed numbers, or size of disgusting factors except one, fail mypy checks.
def process1(x: np.ndarray[tuple[int], np.dtype[np.signedinteger]]): ...
a1 = np.empty(100, dtype=np.int16)
process1(a1) # mypy passes
a2 = np.empty(100, dtype=np.uint8)
process1(a2) # mypy fails
# error: Argument 1 to "process1" has incompatible type
# "ndarray[tuple[int], dtype[unsignedinteger[_8Bit]]]";
# expected "ndarray[tuple[int], dtype[signedinteger[Any]]]" [arg-type]
a3 = np.empty((100, 100, 100), dtype=np.int64)
process1(a3) # mypy fails
# error: Argument 1 to "process1" has incompatible type
# "ndarray[tuple[int, int, int], dtype[signedinteger[_64Bit]]]";
# expected "ndarray[tuple[int], dtype[signedinteger[Any]]]"Runtime verification with sf.CallGuard
Not all the duties of the vocabulary can describe the situation according to the situation or dtype of the resulting array. For this reason, the static analysis will not hold all unwanted frames. It is better than creating refreshment code in all many tasks, the nature of the additions can be reused by Run-Time verification with specialized tools.
Staticframe CallGuard Interface gives two decorates, check including warnwhich increases different warnings, respectively, in verification errors. These decores will confirm the nature's annotations against the symptoms of performance timings.
For example, by adding sf.CallGuard.check In the work below, the Arrays failed to confirm the Expressive CallGuard Different:
import static_frame as sf
@sf.CallGuard.check
def process2(x: np.ndarray[tuple[int], np.dtype[np.signedinteger]]): ...
b1 = np.empty(100, dtype=np.uint8)
process2(b1)
# static_frame.core.type_clinic.ClinicError:
# In args of (x: ndarray[tuple[int], dtype[signedinteger]]) -> Any
# └── In arg x
# └── ndarray[tuple[int], dtype[signedinteger]]
# └── dtype[signedinteger]
# └── Expected signedinteger, provided uint8 invalid
b2 = np.empty((10, 100), dtype=np.int8)
process2(b2)
# static_frame.core.type_clinic.ClinicError:
# In args of (x: ndarray[tuple[int], dtype[signedinteger]]) -> Any
# └── In arg x
# └── ndarray[tuple[int], dtype[signedinteger]]
# └── tuple[int]
# └── Expected tuple length of 1, provided tuple length of 2Store
Much can be done to improve nunpy type. For example, np.object_ The type can be made in such a kind such a kind of Python types contained in the list of something that can be described. For example, 1D list of pairs of pair of pair of pair of paired pairs as:
np.ndarray[tuple[int], np.dtype[np.object_[tuple[int, int]]]]In addition, units of np.datetime64 cannot be naturally specified. Example, Defts Defters can be separated by Nanosecond units with periments such as np.dtype[np.datetime64[Literal['D']]] or np.dtype[np.datetime64[Literal['ns']]].
Even in the limitations, the full-specified Nutter version holds errors and improve the quality of the code. As shown, the TULA analysis can identify the negative structure or dtypeand verification with sf.CallGuard It can provide solid time.
