Support for minimax-regret curation

[MasterXiong]

Dear authors,

Thanks for deleloping this brilliant package to accelerate research in UED!

I'm trying to develop a minimax-regret curator, which seems to be not included in the current repo yet. E.g., for the maze task, the plan is to use algorithms like A* to compute the performance upper bound on each task, so that we can curate with the true regret on each task. I was wondering if you have some suggestions on what may be the best way to implement this? Is it sufficient to just add a new runner to support this new function? Thanks!

[minqi]

Thanks! One way to do this would be to extend ued_scores.py with your custom score function, then simply pass --ued_score='your_function_name' as part of train.py's command line args.