Scityping: type hints and serializers for scientific types in Python

SciTyping provides a collection of type annotations specifiers for Python types common in the fields of scientific computing, data science and machine learning. Most types come with a pair of JSON serializer/deserializer functions, which can be used both to archive and transfer data. In contrast to pickling, the resulting JSON data is both future safe and human readable, which makes especially useful for reproducible research.

This is similar to the way Pydantic adds serializable types for Python apps (especially webapps) – indeed, SciTyping is designed to be compatible with Pydantic, so you can use it to extend Pydantic with support for scientific data types.

Table of contents

• Getting started
  • Compatibility with Pydantic
  • Usage without Pydantic
  • Origin and motivation
  • An extensible serialization hierarchy
  • Performance
  • Running tests
• List of supported types
  • Base types
  • Functions
  • NumPy types
  • SciPy distributions
  • Units
  • PyTorch
• Defining serializers for preexisting types
  • Standard: By subclassing the type with the same name
  • By subclassing the type with a different name
  • For types which don’t allow subclassing: register against ABC
  • Registering the same serializer for multiple types
• Implementation logic
  • Validation logic
  • Serialization logic
  • Serialization of NumPy arrays
• API reference
  • scityping.base
  • scityping.base_types
  • scityping.json
  • scityping.pydantic
  • scityping.utils
  • scityping.numpy
  • scityping.scipy
  • scityping.functions
  • scityping.pint
  • scityping.quantities
  • scityping.torch