Support for latitude/longitude pairs (GPS)

[npatki]

Problem Description

I'm filing this feature request on behalf of other users who want to model latitude/longitude pairs in their data. The latitude and longitude are present as two separate columns. Together, the pair determines a GPS location.

Expected behavior

The SDV synthesizers should be able to understand the latitude and longitude as a location pair. They should create synthetic data that are generally in the same locations as the original data.

Additional context

Some users have tried adding constraints for this, but as evidenced by #1591, custom constraints won't exactly work.

The overall feature can behave similarly to addresses, as seen in #1593.

[npatki]

This feature has now been added, available in Beta for SDV users on our paid plans.

Usage

1. In the metadata, annotate which pairs of latitude/longitude columns belong together. Use the add_column_relationship API method to do this. One of the columns must be sdtype latitude while the other must be sdtype longitude. You can add multiple pairs.

metadata.add_column_relationship(
  table_name='rides',
  type='gps',
  column_names=['start_lat', 'start_lon'])

metadata.add_column_relationship(
  table_name='rides',
  type='gps',
  column_names=['end_lat', 'end_lon'])

2. That's it! SDV synthesizers will now accommodate these lat/lon pairs when modeling and creating synthetic data.

synthesizer = HSASynthesizer(metadata)
synthesizer.fit()
synthetic_data = synthesizer.sample()

Note that this feature is only available for SDV users on our paid plans.