I noticed there's an objective for root mean square error but not for mean absolute error? I'm happy to put together a PR for this if you think this would be useful. Actually what I think would be ...