Design proposal for Fake Data generation

[nhuray]

Hi @JakobGM,

Thanks for that project reconciling Data modelling and validation using Pydantic and Data Transformation using Polars. I think your approach to reconcile those worlds is really interesting ! Kudos 💯

You started to implement APIs for generating examples (Fake data) based on the type defined in the Pydantic model. This is convenient but I think we might go further defining in the pt.Field an attribute example_factory we can pass to generate sample data based on a Factory (similar to the default_factory introduced in Pydantic)

With an example it's probably better to understand:

from typing import Literal, Optional
from mimesis import Generic
import patito as pt

Factory = Generic()


class Employee(pt.Model):
    first_name: str = pt.Field(example_factory=Factory.person.first_name)
    last_name: str = pt.Field(example_factory=Factory.person.last_name)
    age: int = pt.Field(example_factory=lambda: Factory.person.age(18, 65)) 

Here I'm using mimesis as the factory but we can use Faker as well.

Once we can define an example_factory per Field we might imagine to enrich the examples API to generate fake data in bulk:

# Create 50 employees (as a DataFrame)
df = Employee.examples(count=50)

# Create 50 employees (as a sequence of Dictionaries)
employees = Employee.examples(count=50).to_dicts()

Obviously, we should continue support the existing examples API (passing the data to populate the dataframe):

Employee.examples({"age": [25, 32, 49]})

Let me know what you think about that design proposal.

Nicolas