Add set_address_columns method

[amontanez24]

Problem Description

As a user, I'd like a way to specify that multiple columns in my dataset are used to makeup the same address. This way, the synthetic data created maintains a relationship between those columns and makes a valid address.

Expected behavior

In the BaseSynthesizer and BaseMultiTableSynthesizer classes, add a method called add_address_columns

• This method should be used to set the transformer to use on the columns provided.

Parameters

• (required, multi table only) table_name : String that is the name of the table. This must be one of the tables specified in the metadata.
• (required): column_names: A list of one or more column names. These must be specified in the metadata for the table.
• anonymization_level: String that is the type of anonymization the user wants to see. Must be one of:
  • (default) 'full': Anonymize all components of the address (no learning required)
  • 'street_address': Only anonymized the precise street address (and secondary address). Learn everything else.

Behavior

Follow the logic below for transformer assignment

Parameter	Assigned Transformer
'full'	RandomLocationGenerator(locales=<synthesizer locales>)
'street'	RegionalAnonymizer(locales=<synthesizer locales>)

Validation

• (multi table only) The table_name must be found in the metadata
  Error: Unknown table name 'userss'. Please choose a table name from the metadata.
• The column_names must be found in the metadata for the table
  Error: Unknown column names ('A', 'B', 'C'). Please choose column names listed in the metadata for your table.
• The sdtypes for the columns must be compatible with the address transformer. That is they must be one of : 'country', 'country_code', 'administrative_unit', etc.
  Error: Column 'city_name' has invalid sdtype 'categorical'. Please provide a column that is compatible with address data.
• You cannot have 2 or more of the same sdtype within an address.
  Error: Columns 'state_name' and 'state' have the same sdtype 'administrative_unit'. Your address data cannot have duplicate fields.
• If the user has already fit the data, show a warning. The user will need to re-fit in order to get this to work.
  Warning: Please refit your synthesizer for the address changes to appear in your synthetic data.

synthesizer = HSASynthesizer(metadata)
synthesizer.set_address_columns(
    table_name='users',
    column_names=['line1', 'line2', 'city', 'state', 'zip_code', 'country'],
    anonymization_level='full'
)

Additional context

• This method is meant to replace the update_transformers method for address columns. It will happen after the HyperTransformer config is created.
• We will want to keep track of columns that are treated as an address, so we may need to store that information in a list/dict somewhere. This will help for error handling when users call update_transformers on one of these columns or try to add them to constraints.