reprospect.utils.types module

Type information for numeric scalar types.

The minimal information needed to describe a numeric type is:

• kind and signedness — integer or floating-point, signed or unsigned (see Kind)

• bits — width in bits (e.g. 16, 32, 64, 128)

This module provides TypeInfo to carry this information.

Note

Several alternatives to writing this module have been considered:

• numpy does not support arbitrarily wide types such as int128.

• https://github.com/jax-ml/ml_dtypes focuses on small types used in machine learning.

class reprospect.utils.types.Kind(*values)View on GitHub

Bases: Enum

FLOAT = 3

INT = 1

UINT = 2

classmethod from_numpy(kind: str) → SelfView on GitHub

See https://numpy.org/doc/stable/reference/generated/numpy.dtype.kind.html.

property signed: boolView on GitHub

class reprospect.utils.types.TypeInfo(bits: int, kind: Kind)View on GitHub

Bases: object

Minimal type information.

__init__(bits: int, kind: Kind) → None

bits: int

Number of bits.

property itemsize: intView on GitHub

kind: Kind

classmethod normalize(*, dtype: TypeInfo | int | DTypeLike) → SelfView on GitHub

property signed: boolView on GitHub