210 likes | 337 Views
General Game Playing Competition Update. Competition Administration. Current Management Manager makes rules, selects games, runs competition Subject to advice of steering committee composed of Stanford folks and past winners Manager is usually Stanford, this year Dresden Criteria
E N D
Competition Administration Current Management Manager makes rules, selects games, runs competition Subject to advice of steering committee composed of Stanford folks and past winners Manager is usually Stanford, this year Dresden Criteria Promote research Match / slightly exceed current capabilities Provide fun for spectators (understandable,suspenseful)
Proposed Changes to Administration ? Longer Start Clocks Change Ladder-Based Final
Proposed Changes to GDL Build in Arithmetic, e.g. +, *, > ? Step counts in messages ? Game ids and player ids Tournament standings ++ Persistent Connections (as an option)
Metagaming Ideal Challenge - Game Analysis takes time Trade-off - cost of finding structure vs savings Discussion Time proportional to size of the game graph? Small game descriptions possible Analysis can sometimes be done in time proportional to these smaller game descriptions Answer Find cases where cost varies with size of description (e.g. proposition set) rather than with size of expanded game graph
Emphasis on Game Analysis Factorable games Disjunctively decomposable Conjunctively decomposable [Cox] Sequential factoring Conditional factoring Symmetric games See [Schiffel] Game Analysis Invention of game-playing concepts (fork, pin) Evaluation functions “Solvable games”
Single Player Game as a State Machine a b s2 s5 s8 a a b a a b d b b s1 s3 s6 s9 s11 d a c a a c b b a s7 s4 s10
Direct Description Since all of the games that we are considering are finite, it is possible in principle to communicate game information in the form of transition graphs. Problem:Size of description. Even though everything is finite, the graphs can be large.
States versus Features In many cases, worlds are best thought of in terms of features, e.g. red or green, left or right, high or low. Actions often affect subset of features. States represent all possible ways the world can be. As such, the number of states is exponential in the number of “features” of the world, and the action tables are correspondingly large. Idea - represent features directly and describe how actions change individual features rather than entire states. (Reference: STRIPS.)
s1 Propositions Decompose states into “propositions”. Benefit - n propositions can represent 2n states. p q s r
Propositional Net Components Propositions Connectives Transitions p q r
p a1 r q s n1 t1 Example
Performance ? (time (genwinnerp tttrel)) 5,478 states 130,444 milliseconds. 142,685,136 bytes of memory allocated. NIL ? (time (genwinnerp tttprop)) 5,478 states 594,555 milliseconds 117,281,008 bytes of memory allocated. NIL ? (time (propwinnerp tttprop)) 5,478 states 10,390 milliseconds. 5,900,904 bytes of memory allocated. NIL
p1 p2 a1 a2 r1 r2 q1 q2 s1 n1 n2 s2 t1 t2 Example
Performance ? (time (genplan propcompbuttons)) 407 states 1,118 milliseconds. 605,728 bytes of memory allocated. (PROG A B A D E D) ? (time (multiplan propcompbuttons)) 14 states 53 milliseconds 22,320 bytes of memory allocated. (PROG A B A D E D) Partition time: 1 millisecond.
Description of Propnet Components role(white) role(black) proposition(cell(X,Y,M)) :- index(X) & index(Y) & mark(M) action(mark(X,Y)) :- index(X) & index(Y) index(1) mark(x) index(2) mark(o) index(3) mark(b)
Partial Information Games Percept Action Model Situation calculus true(q,N) :- does(P,a,M) & true(p,M) & succ(M,N) sees(P,) Notes: Always possible to simulate with another player On current development plan but not scheduled No plans at present for probabilistic track (though players may use probabilities in playing)
Inductive General Game Playing Unlike GGP, players donot receive complete rules at runtime, only definitions of percepts and actions. On the other hand, players do receive legal game histories. Goal of the player is to induce the rules and then play the game effectively. Alternatively, it may skip straight to learning Q-functions. Interesting to see which approach works best. Longer Start Clock
Really General Game Playing Percept-Action Model for Agent Player is provided Enumeration/definition of percepts and actions That is all Discussion Difficult Gets at issues of conceptualization of world