add script that trains and evaluates all tasks. Before pushing major updates this can be used to check if we are still getting the same numbers

[Zwoelf12]