Need Help on benchmark function for real data about sdgym

[csala]

Hi @ArupDukeBanerjee at the moment SDGym is not intended to be used with your own dataset, but rather to only evaluate and compare the performance of data synthesis methods over a set of well-known datasets.

For the scenario that you mention, we are working on a separated package called SDMetrics that will be made public in the upcoming days.

[ArupDukeBanerjee]

Hi @csala
I just wanted to know one thing about this package. Can it be used for only data generation for real data as you already stated benchmark is yet to come, meanwhile can I use/leverage different generators on my own set of data. Thanks a lot in advance!

Thanks,
Arup

[csala]

@ArupDukeBanerjee Yes, SDGym synthesizers can be used for modeling and sampling your own data, but this is just a secondary effect of having all the synthesizers here implemented with a uniform API.

I would rather recommend you to use the CTGAN package, which is simpler to use and will give you better results in the long term, since it is an actively maintained package with ease of use and sampling quality in mind while SDGym's goal is only to provide benchmark.

[ArupDukeBanerjee]

Hi Carles, Thanks a lot for replying to me. I got your point on benchmarking and also CTGAN is a great package, but while my *data has missing values, it throws errors*. As a part of realistic data generation missing values is also something that needs to be handled. Having said that I mean to say I intend to generate a realistic missing values in my synthetic data, which I believe is not there in CTGAN package. It would be great if you please let me know about the resolution of missing data handling. Thanks a lot! Regards, Arup

…

On Tue, Jun 23, 2020 at 2:18 PM Carles Sala ***@***.***> wrote: @ArupDukeBanerjee <https://github.com/ArupDukeBanerjee> Yes, SDGym synthesizers can be used for modeling and sampling your own data, but this is just a secondary effect of having all the synthesizers here implemented with a uniform API. I would rather recommend you to use the CTGAN <http:///sdv-dev/CTGAN> package, which is simpler to use and will give you better results in the long term, since it is an actively maintained package with ease of use and sampling quality in mind while SDGym's goal is only to provide benchmark. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#23 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AIMRW7VF4DUG3EIWPAYBMNTRYBT7NANCNFSM4LQMTUJQ> .