pydantic-cereal 0.0.6

Advanced serialization for Pydantic models

pydantic-cereal

Advanced serialization for Pydantic models

Pydantic is the most widely used data validation library for Python. It uses type hints/type annotations to define data models and has quite a nice "feel" to it. Pydantic V2 was released in June 2023 and brings many changes and improvements, including a new Rust-based engine for serializing and validating data.

This package, pydantic-cereal, is a small extension package that enables users to serialize Pydantic models with "arbitrary" (non-JSON-fiendly) types to "arbitrary" file-system-like locations. It uses fsspec to support generic file systems. Writing a custom writer (serializer) and reader (loader) with fsspec URIs is quite straightforward. You can also use universal-pathlib's UPath with pydantic-cereal.

📘 See the full documentation here. 📘

Usage Example

See the minimal pure-Python example to learn how to wrap your own type. Below is a preview of this example.

from fsspec import AbstractFileSystem
from pydantic import BaseModel, ConfigDict

from pydantic_cereal import Cereal

cereal = Cereal()  # This is a global variable


# Create and "register" a custom type

class MyType(object):
    """My custom type, which isn't a Pydantic model."""

    def __init__(self, value: str):
        self.value = str(value)

    def __repr__(self) -> str:
        return f"MyType({self.value})"


def my_reader(fs: AbstractFileSystem, path: str) -> MyType:
    """Read a MyType from an fsspec URI."""
    return MyType(value=fs.read_text(path))  # type: ignore


def my_writer(obj: MyType, fs: AbstractFileSystem, path: str) -> None:
    """Write a MyType object to an fsspec URI."""
    fs.write_text(path, obj.value)

MyWrappedType = cereal.wrap_type(MyType, reader=my_reader, writer=my_writer)


# Use type within Pydantic model

class MyModel(BaseModel):
    """My custom Pydantic model."""

    config = ConfigDict(arbitrary_types_allowed=True)  # Pydantic configuration
    fld: MyWrappedType


mdl = MyModel(fld=MyType("my_field"))

# We can save the whole model to an fsspec URI, such as this MemoryFileSystem
uri = "memory://my_model"
cereal.write_model(mdl, uri)

# And we can read it back later
obj = cereal.read_model(uri)
assert isinstance(obj, MyModel)
assert isinstance(obj.fld, MyType)

For wrapping 3rd-party libraries, see the Pandas dataframe example.