pydantic_numpy 3.0.0

Pydantic Model integration of the NumPy array

pydantic-numpy

Integrate NumPy into Pydantic, and provide tooling! NumpyModel make it possible to dump and load np.ndarray within model fields!

Install

pip install pydantic-numpy

Usage

For more examples see test_ndarray.py

import pydantic_numpy.dtype as pnd
from pydantic_numpy import NDArray, NDArrayFp32, NumpyModel


class MyPydanticNumpyModel(NumpyModel):
    K: NDArray[float, pnd.float32]
    C: NDArrayFp32  # <- Shorthand for same type as K


# Instantiate from array
cfg = MyPydanticNumpyModel(K=[1, 2])
# Instantiate from numpy file
cfg = MyPydanticNumpyModel(K={"path": "path_to/array.npy"})
# Instantiate from npz file with key
cfg = MyPydanticNumpyModel(K={"path": "path_to/array.npz", "key": "K"})

cfg.K
# np.ndarray[np.float32]

cfg.dump("path_to_dump_dir", "object_id")
cfg.load("path_to_dump_dir", "object_id")

NumpyModel.load requires the original mode, use model_agnostic_load when you have several models that may be the right model.

Data type (dtype) support!

This package also comes with pydantic_numpy.dtype, which adds subtyping support such as NDArray[float, pnd.float32]. All subfields must be from this package as numpy dtypes have no Pydantic support, which is implemented in this package through the generic class workflow.

Considerations

You can install from cheind's repository if you want Python 3.8 support, but this version only support Pydantic V1 and will not work with V2.

Licensing notice

As of version 3.0.0 the license has moved over to BSD-4. The versions prior are under the MIT license.

History

The original idea originates from this discussion, and forked from cheind's repository.