|
Post by westonmoss on Apr 4, 2016 12:00:38 GMT -8
I want to preface this by saying I am still in the early stages of learning about GGP.
But I wanted to ask a question. Has there been any attempts to take a non-game activity (or at least, one not traditionally thought of as a game) and use GDL to describe its rules, such that it can be presented to the GGP player as a game, and see how the AI reacts? If so are there some examples of that (maybe a paper or a case study?)
What are the challenges in doing this? Is it the "finitely many states" requirement? Or the difficulty in presenting complex rules in GDL format?
Examples of this might be: "Here's a game. Here's the amount of retirement money I want to end up with (goal). Here's the ways it can be saved (rules). Try to end up with the most dollar appreciation when time runs out (resource bounds)."
|
|
|
Post by Andrew Rose on Apr 5, 2016 2:27:50 GMT -8
I coded up a few versions of the graph-colouring problem. No reason that various other optimization problems (knapsack, travelling salesman, etc.) - all of which have real-world applications - couldn't also be coded up. However, you're always going to find that custom solvers perform orders of magnitude better than GGP solvers.
|
|
|
Post by alandau on Apr 6, 2016 20:51:27 GMT -8
I've written up a simple programming problem using a register machine model with time bounds -- one player writes the program and the other picks a test input. I can't imagine, though, how I would make a GGP program good at that.
|
|