any working example of using TypedDict

[ragrawal]

Hi,
One of my pandas data frame is a list of dictionary. I am trying to figure what is the cleanest way to specify this kind of data within pandera so that it works and mypy is also happy about it. Below is a sample code that currently gives the following error:
TypeError: Cannot interpret 'typing.List[__main__.ReviewDict]' as a data type

import pandera as pa
from pandera.typing import Series
from typing import TypedDict, List

ReviewDict = TypedDict(
    "ReviewDict",
    {
        "heading": str,
        "review": str | None,
        "rating": float | None,
    },
)

class MyDataFrameModel(pa.DataFrameModel):
    my_dict_column: Series[List[ReviewDict]] = pa.Field()

# Example usage:
data = {
    "my_dict_column": [
        [{"review": "Great product!", "heading": "Awesome", "rating": 4.5}],
        [{"review": "Not satisfied.", "heading": "Disappointing", "rating": 2.0}],
        # Add more dictionary entries here...
    ]
}

df = pa.typing.DataFrame(data)
MyDataFrameModel(df)

Attempt 2:
If I import TypedDict from typing_extension then I get the following error: TypeError: TypedDict does not support instance and class checks.

Attempt 3:
if I define my_dict_column: List[ReviewDict] then I get the following error: SchemaInitError: Invalid annotation 'my_dict_column: typing.List[__main__.ReviewDict]'

PS: I am using pandera 0.18.0 and pandas 2.2.0

[thetonus]

Looks like you can't do it the way you are wanting to. Dictionaries are pa.Object types. I suggest using a combination of pandas and pydantic:

Here is an example of how to do it if using a dataframe schema approach:

import pandera as pa
import pydantic

from pandera import Column, DataFrameSchema, Check

class MyModel(pydantic.BaseModel):
    x: int
    y: int
    z: int

schema = DataFrameSchema(
    {
        "foo": Column(pa.Object, Check(lambda x: MyModel(**x))),
    },
    strict=True,
    coerce=True,
)

_ = schema.validate(df)

If using a dataframe model, you can use the above with custom checks to derive a similar solution.

Unfortunately, this approach provides validation but not strong typing.

---

I suggest pydantic v2 for performance reasons.