|
Post by Lars Ericson on Nov 7, 2015 8:17:48 GMT -8
I took the Coursera/Stanford GGP course and just finished the Coursera/Stanford/UBC Game Theory 1 course. I also watched William Spaniel's GT 101 videos.A question at the beginning the GGP course in the Discussion Forum was whether there was any relevance of GT to GGP. In GT1 certain classes of game are described, normal form perfect strategy, extensive form PS, NF mixed strategy, extensive form MS, finitely repeated NF games, infinitely repeated games, imperfect information NF games and coalition games. The utility of GT on its own and primary motivation come from analyzing strategic situations in economics and politics. William Spaniel has a nice new series of GT-influenced videos on international relations. The utility of GGP is around writing programs to play conventional entertainment games such as chess, checkers, go and so on, which have perfect information and a finite sequence of moves. The question is whether anything in the analytic tools of GT would be helpful in writing a GGP player. Some obvious ideas such as dominated strategies are helpful. However the results are typically for restricted classes of games which are not natural entertainment games and which don't fit necessarily very well in the current GGP framework. So net net: Very weak intersection. It doesn't hurt to know both, but they are largely different tool sets for different purposes.
|
|
wat
New Member
Posts: 32
|
Post by wat on May 10, 2016 12:43:04 GMT -8
MCTS is, actually, a technique to calculate an approximation to Nash equilibrium. And so is minimax/alpha-beta search.
Games supported by the GDL 1 specification support all discrete, finite, perfect information games. Also supports simultaneous games, although they rarely appear in competitions. GDL 2 adds support for all imperfect information games.
General players need to learn to play a new game on-the-fly, during meta-gaming. Game Theory provide a strong framework to deal with different situations in a general way.
I would say, strong intersection.
The differences... if you want to build a competitive general player, most of development time will be spent in coding fast GDL interpreters, which is, definitely, not GT. But the rest of the time will be spent in coding strong Nash equilibrium calculators, which is, definitely, GT.
|
|