numpydantic/docs/changelog.md

4.9 KiB

Changelog

1.*

1.3.3 - 24-08-13 - Callable type annotations

Problem, when you use a numpydantic "wrap" validator, it gives the annotation as a handler function.

So this is effectively what happens

@field_validator("*", mode="wrap")
@classmethod
def cast_specified_columns(
    cls, val: Any, handler: ValidatorFunctionWrapHandler, info: ValidationInfo
) -> Any:
    # where handler is the callable here
    # so 
    # return handler(val)
    
    return NDArray[Any, Any](val)

where Any, Any is whatever you had put in there.

So this makes it so you can use an annotation as a functional validator. it looks a little bit whacky but idk it makes sense as a PARAMETERIZED TYPE

>>> from numpydantic import NDArray, Shape
>>> import numpy as np

>>> array = np.array([1,2,3], dtype=int)
>>> validated = NDArray[Shape["3"], int](array)
>>> assert validated is array
True

>>> bad_array = np.array([1,2,3,4], dtype=int)
>>> _ = NDArray[Shape["3"], int](bad_array)
    175 """
    176 Raise a ShapeError if the shape is invalid.
    177 
    178 Raises:
    179     :class:`~numpydantic.exceptions.ShapeError`
    180 """
    181 if not valid:
--> 182     raise ShapeError(
    183         f"Invalid shape! expected shape {self.shape.prepared_args}, "
    184         f"got shape {shape}"
    185     )

ShapeError: Invalid shape! expected shape ['3'], got shape (4,)

Performance:

  • Don't import the pandas module if we don't have to, since we are not using it. This shaves ~600ms off import time.

1.3.2 - 24-08-12 - Allow subclasses of dtypes

(also when using objects for dtypes, subclasses of that object are allowed to validate)

1.3.1 - 24-08-12 - Allow arbitrary dtypes, pydantic models as dtypes

Previously we would only allow dtypes if we knew for sure that there was some python base type to generate a schema with.

That seems overly restrictive, so relax the requirements to allow any type to be a dtype. If there are problems with serialization (we assume there will) or handling the object in a given array framework, we leave that up to the person who declared the model to handle :). Let people break things and have fun!

Also support the ability to use a pydantic model as the inner type, which works as expected because pydantic already knows how to generate a schema from its own models.

Only one substantial change, and that is a get_object_dtype method which interfaces can override if there is some fancy way they have of getting types/items from an object array.

1.3.0 - 24-08-05 - Better string dtype handling

API Changes:

  • Split apart the validation methods into smaller chunks to better support overrides by interfaces. Customize getting and raising errors for dtype and shape, as well as separation of concerns between getting, validating, and raising.

Bugfix:

  • #4 - Support dtype checking for strings in zarr and numpy arrays

1.2.3 - 24-07-31 - Vendor nptyping

nptyping vendored into numpydantic.vendor.nptyping - nptyping is no longer maintained, and pins numpy<2. It also has many obnoxious warnings and we have to monkeypatch it so it performs halfway decently. Since we are en-route to deprecating usage of nptyping anyway, in the meantime we have just vendored it in (it is MIT licensed, included) so that we can make those changes ourselves and have to patch less of it. Currently the whole package is vendored with modifications, but will be whittled away until we have replaced it with updated type specification system :)

Bugfix:

  • #2 - Support numpy>=2
  • Remove deprecated numpy dtypes

CI:

  • Add windows and mac tests
  • Add testing with numpy>=2 and <2

DevOps:

  • Make a tox file for local testing, not used in CI.

Tidying:

  • Remove monkeypatch module! we don't need it anymore! everything has either been upstreamed or vendored.

1.2.2 - 24-07-31

Add datetime map to numpy's :class:numpy.datetime64 type

1.2.1 - 24-06-27

Fix a minor bug where {class}~numpydantic.exceptions.DtypeError would not cause pydantic to throw a {class}pydantic.ValidationError because custom validator functions need to raise either AssertionError or ValueError - made DtypeError also inherit from ValueError because that is also technically true.

1.2.0 - 24-06-13 - Shape ranges

  • Add ability to specify shapes as ranges - see shape ranges

1.1.0 - 24-05-24 - Instance Checking

https://github.com/p2p-ld/numpydantic/pull/1

Features:

  • Add __instancecheck__ method to NDArrayMeta to support isinstance() validation
  • Add finer grained errors and parent classes for validation exceptions
  • Add fast matching mode to {meth}.Interface.match that returns the first match without checking for duplicate matches

Bugfix:

  • get all interface classes recursively, instead of just first-layer children
  • fix stubfile generation which badly handled typing imports.