Better Handle Variable Types

[jrm5100]

I think a good test-case for this will be creating a datatype for a variant based on some PLINK files:

• BED (00/01/10/11 encoding for each variant for each sample)
• BIM (chromosome, ID, position, coordinate, allele 1, allele 2)

I think the BIM information would be part of a pandas.api.extensions.ExtensionArray type (since the information would apply to an array/column) and the actual genotype information would be part of a pandas.api.extensions.ExtensionDtype type.

[jrm5100]

Currently the variable types are converted from the underlying numerical types.

Binary = Categorical with 2 categories
Categorical = Categorical with > 2 categories
Continuous= Anything numeric that hasn't been made into a 'category' type
Unknown = 'object' type which occurs when two different types are combined, or when there are strings.

• Right now converting to categorical actually converts to binary if there are only two unique values.
• Ideally the starting type would always be "unknown".

The custom datatypes mentioned above could be a good solution (and would allow for more types in the future as well).

[jrm5100]

As of Pandas v1.1, any value can be converted to the "String" type. This could be treated as "Unknown" in CLARITE and loaded by default.