How do you manage an inter-column dependency?

[npatki]

I'm filing this issue on behalf of @Pavan-Kalyan1432, who first asked the question in this comment.

Problem description

How to manage inter column dependency...
For example we have 3 columns date of birth, date of death and age... In the synthetic data it is not coming properly. Give me the answer for both single table and multi table

[npatki]

Hi @Pavan-Kalyan1432,

I am assuming that birth and date of death are both datetime columns, whereas age is a numerical column. It seems your data has the following logical rules:

1. date of death must occur after birth
2. age must be exactly equal to the # of years between birth and date of death

Note that SDV synthesizers use AI to learn from your data, which is inherently probabilistic. So if you have any hard-and-fast rules like this (that all rows must follow), a synthesizer will not produce it 100% of the time using just the default options. This is to be expected.

Using Constraints

To resolve a hard-and-fast rule like this, I would recommend you use constraints. Constraints can be applied to both single and multi-table datasets. Some resources are below:

• Demo about using constraints
• Inequality constraint -- this could be useful to enforce that date of death must occur after birth
• Custom constraint -- you would probably need to add custom logic for the computation in the age column.
  • Alternatively, since age can be computed using the other two columns, there is really no need to input into SDV in the first place. You can just leave it out (drop the column) and recreate it in the synthetic data afterwards.