Include automatic evaluation of existing historic benchmark dataasets.

[reisepass]

Should AI model providers get the option to use the framework and network for testing their model against historic data ?

Pro :

• More test data may give a better generalization error
• A draw for AI model providers that want to use the service to help them guide their model improvement

Con :

• Added complexity
• To solve the problem of over fitting on public test datasets we can't include these results in the model score