Data structures to represent data in Python

August 16, 2022

• https://www.youtube.com/watch?v=T-TwcmT6Rcw&t=1390s
• https://docs.python.org/3/library/dataclasses.html
• https://stackoverflow.com/questions/51671699/data-classes-vs-typing-namedtuple-primary-use-cases
• https://shayallenhill.com/python-struct-options/
• https://realpython.com/python-data-classes/

Comments from Nic:

I see it as a hierarchy of needs where I choose the "simplest" for the use case required:

• tuple - frozen list
• collections.namedtuple - tuple with labels for positions (I use this for when I want to return multiple objects that don't really go together normally, eg pytest example test case parameters)
• typing.NamedTuple - namedtuple with type hints
• dataclasses - for cases where the object can be thought of as a genuine data model
• pydantic - for when you have a data model and want validation, conversion and serialisation included

Given that we use pydantic extensively already and I like validation the gap in use cases between typing.NamedTuple and pydantic is quite small in my mind which is why I don't default to using them much.