For each resource, there exists a variant parameter that can be configured either manually, automatically, or through our ff.set_run API. The term “variant” is preferred over “version” as it accurately reflects the nature of ML feature versions—they are not necessarily “improved” versions compared to older ones. Instead, they represent diverse interpretations of the same concept, each potentially more suited for specific use cases.

​
Auto-generated Variants

Should a variant not be explicitly set, Featureform generates one automatically by combining a randomly selected adjective and noun.

​
Setting all Variants in a Run

While experimenting within a notebook, it’s often desirable to assign the same variant to all resources. Achieving this is straightforward using the ff.set_run(variant) API. This action sets the provided string parameter as the variant for all subsequent resources. Any variant parameter within those subsequent resources would supersede this setting.