Spike: Improve Generics for Input Types in Primitives

[gsheni]

Problem

• Certain comparison primitives like Equal, NotEqual, EqualScalar, NotEqualScalar use the ColumnSchema base class in their allowed primitive inputs. This allows them to take accept any Logical Type as input. For the Equal case, the allowed input is defined this way:

    input_types = [ColumnSchema(), ColumnSchema()]

• This means that valid input could be two numeric features, two text features, or even one text and one numeric feature.
• But we only want to compare two numeric features, two text features. Not the third case.
• We therefore need a way to link two primitive input types.

Request

• This spike is to create a design document to possible solutions to this problem.

Related Docs

• https://alteryx.atlassian.net/wiki/spaces/PS/pages/1034519162/Improving+Generic+Variable+Types+in+Primitives
• https://github.com/alteryx/featuretools/issues/845
• https://github.com/alteryx/featuretools/issues/842

[gsheni]

@davesque questions to answer in design doc:

• Should nullability be a property of ColumnSchema?
• Should we move to 1 Logical Type and nullability being a property? (ex - Integer and IntegerNullable)

Related issues:

• https://github.com/alteryx/woodwork/issues/1121
• https://github.com/alteryx/woodwork/issues/1106