2-14 extensive games - Swarthmore Collegebryce/cs63/s18/... · What can we model so far? With...

8
Extensive Form Games 2/14/18

Transcript of 2-14 extensive games - Swarthmore Collegebryce/cs63/s18/... · What can we model so far? With...

  • ExtensiveFormGames2/14/18

  • Alpha-BetaPruningExercise

    +

    − −

    ++ ++ ++

    − − − − − − − − −

    5 6 7 4 5 3 6 8 6 9 4 7 6 9 9 8 6 25

  • DiscussionQuestions

    1. Howcanweorganizethesearchtomaximizetheamountofworksavedbypruning?A. Explorebestnodesfirst(estimatewithheuristic).

    2. Howcanwemodifyalpha/betapruningtoworkonnon-zero-sumgames?A. Wecan’t.

    3. Howcanwemodifyalpha/betapruningtoworkon3-(ormore)-playergames?A. Wecan’t.

  • IterativeDeepeninginMin/Max

    • Wegenerallyimposeadepthlimitonalpha/betamin/max,basedonhowmuchtimewewanttoallow.

    • Ifwehavetimeleftover,wecouldre-runwithalargerdepthlimit. Ifwerunoutoftime,wecanjustreturnthemovefromthelastrun.

  • Whatcanwemodelsofar?Withminimax,wecansolve:• two-player,zero-sum,completeinformation,sequentialmovegames• lotsofclassicboardgames:chess,checkers,connect4…• notmuchelse,andthesearchspacefortheaboveisoftentoobig

    Withbackwardsinduction(sofar)wecansolve:• completeinformation,sequentialmovegames• simplemodelsofeconomiccompetition,afewothereconomicapplications

  • Gametheoryisn’tjustaboutplayinggames.

    • We’dliketomodeldecisionmakinginallsortsofmulti-agentinteractions.

    • Todoso,weneedtocomeupwithamodel(suchasagametree)thatdescribestheinteraction.

    • Weneedtoidentifytheplayers,whentheymakedecisions,andhowwhattheywanttoachieve.

  • ExampleApplication:ResourceSharingAlsoknownascake-cutting,asin“Icut,youchoose”.• Oneagentproposesadivisionofadesirableresource,theotheracceptsorrejectsthatdivision.• Here,wemodeltheagentsasgettingutility+1foreachunitofresource,butthey’realsospitefulsotheyfeeladisutilityof0.5iftheyreceivelessthanhalf.• MANYvariationsonthismodelarepossible.

    1

    2 2 2 2

    3,-.5 0,0 2,.5 0,0 .5,2 0,0 -.5,3 0,0

    3,02,1 1,2

    0,3

    A R A R A R A R

  • Whatcan’twemodelsofar?

    RandomOutcomes

    SimultaneousMoves

    IncompleteInformation