Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd Logo
Upload

Welcome to Scribd!

  • 169 views

    3b Extensive-Form Games

    Uploaded by

    Muhammad Ramzan
    1. The document introduces extensive-form games and uses the example of a sharing game between two children to illustrate key concepts. 2. Extensive-form games make the temporal order of actions and decisions explicit in a game tree structure, whereas normal-form games do not assume any temporal order. Both representations can model the same game. 3. Backward induction is a technique for finding subgame-perfect equilibria in extensive-form games by recursively determining the optimal actions working backwards from the end of the game tree. However, it fails to predict behavior in some games like the Centipede Game.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd

    3b Extensive-Form Games

    Uploaded by

    Muhammad Ramzan
    169 views17 pages
    1. The document introduces extensive-form games and uses the example of a sharing game between two children to illustrate key concepts. 2. Extensive-form games make the temporal order of actions and decisions explicit in a game tree structure, whereas normal-form games do not assume any temporal order. Both representations can model the same game. 3. Backward induction is a technique for finding subgame-perfect equilibria in extensive-form games by recursively determining the optimal actions working backwards from the end of the game tree. However, it fails to predict behavior in some games like the Centipede Game.

    Original Description:

    game theory

    Copyright

    © Attribution Non-Commercial (BY-NC)

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    1. The document introduces extensive-form games and uses the example of a sharing game between two children to illustrate key concepts. 2. Extensive-form games make the temporal order of actions and decisions explicit in a game tree structure, whereas normal-form games do not assume any temporal order. Both representations can model the same game. 3. Backward induction is a technique for finding subgame-perfect equilibria in extensive-form games by recursively determining the optimal actions working backwards from the end of the game tree. However, it fails to predict behavior in some games like the Centipede Game.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    169 views17 pages

    3b Extensive-Form Games

    Uploaded by

    Muhammad Ramzan
    1. The document introduces extensive-form games and uses the example of a sharing game between two children to illustrate key concepts. 2. Extensive-form games make the temporal order of actions and decisions explicit in a game tree structure, whereas normal-form games do not assume any temporal order. Both representations can model the same game. 3. Backward induction is a technique for finding subgame-perfect equilibria in extensive-form games by recursively determining the optimal actions working backwards from the end of the game tree. However, it fails to predict behavior in some games like the Centipede Game.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 17
    At a glance
    Powered by AI
    The key takeaways are about extensive-form games, pure strategies, Nash equilibria, subgame-perfect equilibria, and backward induction.

    An extensive-form game is a representation of a game that makes the temporal structure explicit and doesn't assume agents act simultaneously.

    A pure strategy for an agent in an extensive-form game specifies which action the agent will take at every decision node where it is their turn to choose an action.

    Introduction to Game Theory

    3b. Extensive-Form Games


    Dana Nau University of Maryland

    Nau: Game Theory 1

    The Sharing Game


    !! Suppose agents 1 and 2 are two children !! Someone offers them two cookies, but only if they can agree how to share

    them
    !! Agent 1 chooses one of the following options:
    !! Agent 1 gets 2 cookies, agent 2 gets 0 cookies !! They each get 1 cookie !! Agent 1 gets 0 cookies, agent 2 gets 2 cookies

    !! Agent 2 chooses to accept or reject the split:


    !! Accept =>

    they each get their cookies(s)


    !! Otherwise, neither gets any
    no (0,0) 2

    2-0

    1 1-1 2 no (0,0) yes (1,1)

    0-2 2 no (0,0) yes (2,0)

    yes (2,0)

    Nau: Game Theory 2

    Extensive Form
    !! The sharing game is a game in extensive form
    !! A game representation that makes the temporal structure explicit !! Doesnt assume agents act simultaneously

    !! Extensive form can be converted to normal form, so previous results carry over
    !! But there are additional results that depend on the temporal structure

    !! In a perfect-information game, the extensive form is a game tree:


    !! Nonterminal node = place where an agent chooses an action !! Edge = an available action or move !! Terminal node = a final outcome

    2-0 2 no (0,0) yes (2,0) no (0,0)

    1 1-1 2 yes (1,1)

    0-2 2 no (0,0) yes (0,2)

    ! At each terminal node h, each agent i has a utility ui(h)

    Nau: Game Theory 3

    Pure Strategies
    !! Pure strategy for agent i in a perfect-information game:
    !! specifies which action to take at every node where its is choice

    Sharing game:
    !! Agent 1 has 3 pure strategies:
    !! S1 = {2-0, 1-1, 0-2}

    !! Agent 2 has 8 pure strategies:


    !! S2 = {(yes, yes, yes), (yes, yes, no), (yes, no, yes), (yes, no, no), (no, yes, yes), (no, yes, no), (no, no, yes), (no, no, no)}

    2-0 2

    1 1-1 2 no (0,0) yes (1,1)

    0-2 2 no (0,0) yes (0,2)

    no (0,0)

    yes (2,0)

    Nau: Game Theory 4

    Extensive form vs. normal form


    !! Every game tree corresponds to an
    C (3,8)

    1 A 2 D (8,3) E (5,5) G (2,10) B 2 F 1 H (1,0)

    equivalent normal-form game


    !! The first step is to get all of the agents

    pure strategies
    !! An agents complete strategy must specify an

    action at every node where its the agents move


    !! Example: the game tree shown here
    !! Agent 1 has four pure strategies:

    ! s1 = {(A, G), (A, H), (B, G), (B, H)} ! Must include (A, G) and (A, H), even though action A makes the G-versus-H choice moot
    !! Agent 2 also has four pure strategies:

    ! s2 = {(C, E), (C, F), (D, E), (D, F)}


    Nau: Game Theory 5

    Extensive form vs. normal form


    !! Once we have all of the pure strategies,
    C (3,8)

    1 A 2 D (8,3) E (5,5) G B 2 F 1 H (1,0)

    we can rewrite the game in normal form


    !! Converting to normal form

    introduces redundancy
    !! 16 outcomes in the payoff matrix,

    (2,10)

    versus 5 outcomes in the game tree


    !! Payoff (3,8) occurs

    ! once in the game tree ! four times in the payoff matrix


    !! This can cause an exponential blowup

    Nau: Game Theory 6

    Nash Equilibrium
    2

    B 2 D (8,3) E (5,5) G (2,10) F 1 H (1,0)

    !! Theorem. Every perfect-information game in

    C (3,8)

    extensive form has a pure-strategy Nash equilibrium


    !! This theorem has been attributed to

    Zermelo (1913), but theres some controversy about that


    !! Intuition:
    !! Agents take turns, and everyone sees

    whats happened so far before making a move


    !! So never need to introduce randomness into

    action selection to find an equilibrium


    !! In our example, there are three pure-strategy

    Nash equilibria

    Nau: Game Theory 7

    Nash Equilibrium
    2

    B 2 D (8,3) E (5,5) G (2,10) F 1 H (1,0)

    !! The concept of a Nash equilibrium can be

    C (3,8)

    too weak for use in extensive-form games


    !! Recall that our example has three

    pure-strategy Nash equilibria:


    !! {(A,G), (C,F)} !! {(A,H), (C,F)} !! {(B,H), (C,E)}

    !! Here is {(B,H), (C,E)} with

    the game in extensive form

    Nau: Game Theory 8

    Nash Equilibrium
    2

    B 2 D (8,3) E (5,5) G (2,10) F 1 H (1,0)

    !! If agent 1 used (B,G) instead of (B,H)


    !! Then agent 2s best response

    C (3,8)

    would be (C,F), not (C,E)


    !! When agent 1 plays B
    !! The only reason for agent 2 to choose E

    is if agent 1 has already committed to H rather than G


    !! This behavior by agent 1 is a threat:
    !! By committing to choose H, which is harmful to agent 2,

    agent 1 can make agent 2 avoid that part of the tree


    !! Thus agent 1 gets a payoff of 5 instead of 2

    !! But is agent 1s threat credible?


    !! If agent 2 plays F, would agent 1 really play H rather than G? !! It would reduce agent 1s own utility

    Nau: Game Theory 9

    Subgame-Perfect Equilibrium
    !! Given a perfect-information extensive-form game G, the subgame of G rooted

    at node h is the restriction of G to the descendants of h


    !! Now we can define a refinement of the Nash equilibrium that eliminates

    noncredible threats
    !! A subgame-perfect equilibrium (SPE) is a strategy profile S such that for

    every subgame G! of G, the restriction of S to G! is a Nash equilibrium of G!


    !! Since G itself is is a subgame of G, every SPE is also a Nash equilibrium

    !! Every perfect-information extensive-form game has at least 1 SPE


    !! Can prove this by induction on the height of the game tree

    Nau: Game Theory 10

    Example
    2

    B 2 D (8,3) E (5,5) G (2,10) F 1 H (1,0)

    !! Recall that we have three Nash equilibria:

    C (3,8)

    {(A, G), (C, F)} {(A, H), (C, F)} {(B, H), (C, E)}
    !! Consider this subgame:
    !! For agent 1,

    G strictly dominates H
    !! Thus H cant be part of a Nash equilibrium !! This excludes {(A, H), (C, F)} and {(B, H), (C, E)} !! Just one subgame-perfect equilibrium

    ! {(A, G), (C, F)}


    Nau: Game Theory 11

    Backward Induction
    !! To find subgame-perfect equilibria, we can use backward induction !! Identify the equilibria in the bottom-most nodes
    !! Assume theyll be played if the
    1 (3,8) A 2 (3,8) C (3,8) D (8,3) E (5,5) G (2,10) B 2 (2,10) F 1 (2,10) H (1,0)

    game ever reaches these nodes


    !! For each node x, recursively compute

    a vector vx = (vx1, , vxn) that gives every agents equilibrium utility


    !! At each node x,

    ! If i is the agent to move, then is equilibrium action is to move to a child y of x for which is equilibrium utility vyi is highest ! Thus vx = vy

    Nau: Game Theory 12

    Lets Play a Game


    !! I need two volunteers to

    play the game shown here:


    !! One to be Agent 1 !! One to be Agent 2

    !! Whenever its your turn to move, you have two possible moves:
    !! C (continue) and S (stop)

    !! Agent 1 makes the first move !! At each terminal node, the payoffs are as shown

    Nau: Game Theory 13

    A Problem with Backward Induction


    The Centipede Game
    !! Can extend this game

    to any length
    !! The payoffs are constructed in such a way that for each agent, the only SPE

    is always to choose S
    !! This equilibrium isnt intuitively appealing
    !! Seems unlikely that an agent would choose S near the start of the game !! If the agents continue the game for several moves, theyll both get

    higher payoffs
    !! In lab experiments, subjects continue to choose C until close to the end

    of the game

    Nau: Game Theory 14

    A Problem with Backward Induction


    !! Suppose agent 1 chooses C !! If youre agent 2,

    what do you do?


    !! SPE analysis says you should choose S !! But SPE analysis also says you should never have gotten here at all !! How to amend your beliefs and course of action based on this event?

    !! Fundamental problem in game theory


    !! Differing accounts of it, depending on

    ! the probabilistic assumptions made ! what is common knowledge (whether there is common knowledge of rationality) ! how to revise our beliefs in the face of an event with probability 0
    Nau: Game Theory 15

    Backward Induction in Zero-Sum Games


    !! Backward induction works much better in zero-sum games
    !! No zero-sum version of the Centipede Game, because we cant have increasing

    payoffs for both players


    !! Only need one number: agent 1s payoff (= negative of agent 2s payoff) !! Propagate agent 1s payoff up to the root
    !! At each node where its agent 1s move,

    the value is the maximum of the labels of its children


    !! At each node where its agent 2s move,

    the value is the minimum of the labels of its children


    !! The roots label is the value of the game (from the Minimax Theorem)

    !! In practice, it may not be possible to generate the entire game tree


    !! E.g., extensive-form representation of chess has about 10150 nodes

    !! Need a heuristic search algorithm


    Nau: Game Theory 16

    Summary
    !! Extensive-form games
    !! relation to normal-form games !! Nash equilibria !! subgame-perfect equilibria !! backward induction

    ! The Centipede Game


    !! backward induction in zero-sum games

    Nau: Game Theory 17

    You might also like