You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently, we recommend the SingleTablePreset (with 'FAST_ML' option) as the first quickstart synthesizer that a user should try. At the time we added this, the intention was to provide the easiest option for users to try creating synthetic data. But since then, there have been changes that make the SingleTablePreset more of a liability.
We've improved the GaussianCopulaSynthesizer so much that its speed is similar to SingleTablePreset
The GaussianCopulaSynthesizer (and other SDV synthesizers) are hooked up to other customizations (constraints, transformers, etc.) that are not possible with SingleTablePreset. So users have to move to these synthesizers anyway.
SingleTablePreset is a special wrapper on top of GaussianCopulaSynthesizer and as such, it costs us more to maintain and update it with features.
The original intention was to add other options (besides FAST_ML) to SingleTablePreset but upon further consideration, we would rather expose these as different synthesizers.
Expected behavior
Deprecate SingleTablePreset. To be backwards compatible, we should still allow users to use the functionality without crashing. However, any function within the SingleTablePreset class (including __init__) should yield a FutureWarning asking the user to switch to GaussianCopulaSynthesizer instead.
>>> synth = SingleTablePreset(metadata, name='FAST_ML')
Future Warning: The 'SingleTablePreset' is deprecated. For equivalent Fast ML functionality, please use the 'GaussianCopulaSynthesizer'.
Additional context
We will also need to update our docs and Quickstart tutorial after this is updated.
The text was updated successfully, but these errors were encountered:
Problem Description
Currently, we recommend the
SingleTablePreset
(with'FAST_ML'
option) as the first quickstart synthesizer that a user should try. At the time we added this, the intention was to provide the easiest option for users to try creating synthetic data. But since then, there have been changes that make theSingleTablePreset
more of a liability.GaussianCopulaSynthesizer
so much that its speed is similar toSingleTablePreset
GaussianCopulaSynthesizer
(and other SDV synthesizers) are hooked up to other customizations (constraints, transformers, etc.) that are not possible withSingleTablePreset
. So users have to move to these synthesizers anyway.SingleTablePreset
is a special wrapper on top ofGaussianCopulaSynthesizer
and as such, it costs us more to maintain and update it with features.FAST_ML
) toSingleTablePreset
but upon further consideration, we would rather expose these as different synthesizers.Expected behavior
Deprecate
SingleTablePreset
. To be backwards compatible, we should still allow users to use the functionality without crashing. However, any function within theSingleTablePreset
class (including__init__
) should yield aFutureWarning
asking the user to switch toGaussianCopulaSynthesizer
instead.Additional context
We will also need to update our docs and Quickstart tutorial after this is updated.
The text was updated successfully, but these errors were encountered: