[Python] Complex Recursive Pydantic class initialization: how to have native Pydantic type...

I want to use the Pydantic to model a "cartesian table" represented as a nested dictionary. Here is an example of what I can expect as input to the NestedCounts model:

{
"x1": {
"y1": {
"z1": 5,
"z2": 5
},
"y2": {
"z1": 7,
"z2": 3
},
"y3": {
"z1": 8,
"z2": 2
}
},
"x2": {
"y1": {
"z1": 1,
"z2": 9
},
"y2": {
"z1": 2,
"z2": 8
},
"y3": {
"z1": 4,
"z2": 6
}
}
}


The keys at each level will always be strings. The values at the final level will be int, and the values at a non-final level will be a recursive case.

What is the best way to model this using Pydantic? I want to model it in such a way that the native type checking is supported when instantiating a model.

Here is what I am using for now:

from __future__ import annotations
from pydantic import BaseModel

class NestedCounts(BaseModel):
counts: dict[str, int|NestedCounts]

@classmethod
def from_dict(cls, data: dict) -> NestedCounts:
if all(isinstance(v, int) for v in data.values()):
return cls(counts=data)
elif all(isinstance(v, dict) for v in data.values()):
return cls(counts={k: NestedCounts.from_dict(v) for k, v in data.items()})
else:
raise ValueError(f"Invalid dependent variable probabilities. All values must be ints or all values must be dictionaries:\n{data}")


Is there a better way to model such input schema? I want to make sure that whatever I implement allows Pydantic's native type checking.

Continue reading...