Constraints for generating synthetic data

Question

Constraints for generating synthetic data

Closed this issue 2 years ago · 3 comments

I'd like to make synthtic data with constraints like following, however, I don't know to implement it.

constraint conditions

Column A value is greter than Column B value on the record level.
And, both Column A value and Column B value is positive number.

I know that Synthpop::syn function provides can accept constraints via rules and rvalues argument, but I don't think to be able to configured the above constraints.

Any one, please let me to any suggestions.

Answer 1 · 2023-05-02T06:27:30.000Z

Hi, I assume you are making a synthetic data set based on a real data set? If so, when using the default synthesizing method (cart), the synthetic data will only contain values that are in the real data. So if Column A and B are both strictly positive in the real data, they will also be strictly positive in the synthetic data. The other constraint (Column A>Column B) is a little more complicated. You're right that the rules and rvalues arguments aren't fit for purpose here. I would advise to try and run the synthhpop::syn() first. Sometimes the default settings will find constraints like these in the real data and copy them in the synthetic data. If that doesn't work, there is another approach. You can redefine your dataset a little by adding a new Column C between Column A and Column B. Column C is equal to Column A-Column B. This column should be strictly positive in the real data, and so it will also be strictly positive in the synthetic data. You can then reconstruct Column B by calculating Column A-Column C, either by using the rules and rvalues arguments or manually outside the synthesis. It's a little hacky, but ought it to work. Hope this helps. Best, Lotte

…

On Mon, May 1, 2023, 15:34 Ippei Akiya ***@***.***> wrote: I'd like to make synthtic data with constraints like following, however, I don't know to implement it. constraint conditions Column A value is greter than Column B value. And, both Column A value and Column B value is positive number. I know that Synthpop::syn function provides can accept constraints via rules and rvalues argument, but I don't think to be able to configured the above constraints. Any one, please let me to any suggestions. — Reply to this email directly, view it on GitHub <#28>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AKYULHDAKLZ2BYKSWMGDEY3XD63YDANCNFSM6AAAAAAXRY3X2M> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

Answer 2 · 2023-05-02T07:43:51.000Z

You could try this. 1 synthesise column b first constrained to be >0 then either synthesise (a-b) constrained to be >0 or a constrained to be >b Hope this helps Gillian M Raab Research Fellow (part-time) Scottish Centre for Administrative Data Research My core working days are Tuesdays and Thursdays Though I sometimes swap them for other days 07748 678 551

…

________________________________ From: Ippei Akiya ***@***.***> Sent: 01 May 2023 14:34 To: bnowok/synthpop ***@***.***> Cc: Subscribed ***@***.***> Subject: [bnowok/synthpop] Constraints for generating synthetic data (Issue #28) This email was sent to you by someone outside the University. You should only click on links or attachments if you are certain that the email is genuine and the content is safe. I'd like to make synthtic data with constraints like following, however, I don't know to implement it. constraint conditions Column A value is greter than Column B value. And, both Column A value and Column B value is positive number. I know that Synthpop::syn function provides can accept constraints via rules and rvalues argument, but I don't think to be able to configured the above constraints. Any one, please let me to any suggestions. — Reply to this email directly, view it on GitHub<#28>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AE3HB7G7NCZEC7BSFQUUHK3XD63YDANCNFSM6AAAAAAXRY3X2M>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***> The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. Is e buidheann carthannais a th’ ann an Oilthigh Dhùn Èideann, clàraichte an Alba, àireamh clàraidh SC005336.

Answer 3 · 2023-05-03T05:22:50.000Z

Thanks Lotte and Gillian.

I tried Lotte's hacky approch that make intermediate Column C.
As a result, this way is seed perfect in CART and Ranger method that absolutely keeps Column A > Column B in 76400 records.

Thank you again for your suggestions.
Ippei