THE CHALLENGES CONCEPT
First of all, the overall “catalog” in the portal is actually an aggregate view of several sub-catalogs and each sub-catalog has an access-type and contains a distinct set of models (= AI Modules).
A challenge is basically a special kind of catalog in the sense that it has an special view called “Leaderboard” where the entries are ordered by the metrics of execution-runs, best models first.
So, to start a new challenge, the challenge owner has to create a new catalog of type “Challenge” for it: The challenge participants can then do the regular Onboarding of their AI-Modules. It will mean to add a new selection to the catalog creation dialog and the catalog table to represent the catalog type: normal or challenge.
To actually take part in the challenge, the model must be published to the challenge catalog, in this example it would be the Classification Challenge Catalog:
Publication is a two-step process: in the first step, the user submits the module for publication and in the second step, a reviewer must approve the publication. Finally, the participant must create a pipeline that connects their model with the Challenge input data (a “Databroker”) and maybe add other necessary AI-Modules.
-
add catalog type to the database and catalog class (different ticket) -
add catalog type column to the catalog table page -
add catalog type selection to the catalog creation dialog -
change marketplace view of a challenge catalog to list instaed of tiles