Create a custom column type that is analogous to the custom validator. This is distinct from plugins in the sense that the CustomColumnGenerator doesn't create a new type. Instead, it allows the user to provide the generate method as a function.
Some questions:
-
How do we handle the resource_provider? e.g., when the user wants to use an LLM in their custom column generator. My quick thought is that the user should specify required resources (mainly for bookkeeping and validation). When they implement the generate method, maybe they use generate(self, data: pd.DataFrame) -> DataFrame. Then we document that self has access to the resource provider.
-
What about the config for the custom column? Do we allow arbitrary parameters?
Create a custom column type that is analogous to the custom validator. This is distinct from plugins in the sense that the CustomColumnGenerator doesn't create a new type. Instead, it allows the user to provide the generate method as a function.
Some questions:
How do we handle the
resource_provider? e.g., when the user wants to use an LLM in their custom column generator. My quick thought is that the user should specify required resources (mainly for bookkeeping and validation). When they implement thegeneratemethod, maybe they usegenerate(self, data: pd.DataFrame) -> DataFrame. Then we document thatselfhas access to the resource provider.What about the config for the custom column? Do we allow arbitrary parameters?