ML Wiki
Machine Learning Wiki - A collection of ML concepts, algorithms, and resources.

Normal Form Game

Normal Form Game

A ‘‘Normal Form Game’’ (also ‘‘Matrix Form Game’’ or ‘‘Strategic Game’’) if a type of games from the Game Theory

  • main idea: the players move simultaneously
  • compare to Extensive Form Games where players move sequentially

In these games:

  • There is a finite number $n$ of rational players: $N = { 1, 2, …, n }$
  • Each player $i$ has a finite set of actions $A_i$
  • Also each player $i$ has a set of possible consequences $C$, in this case it’s $C \equiv \mathbb{R}$
  • The chosen alternatives form an ‘‘action profile’’ (or ‘‘strategy profile’’) $A: A_1 \times … \times A_n$
  • There’s a consequence function $g_i: A \mapsto C$ that associates each alternative $a \in A_i$ with some consequence
  • The preferences of agents are modeled with an utility function $u_i: A \mapsto \mathbb{R}$

’'’def:’’’ So a ‘‘normal form game’’ (or a ‘‘strategic game’’) is

  • a tuple $\langle N, A, u \rangle$ where
  • $N = { 1, 2, …, n }$ - set of all players
  • $A = {A_1, …, A_n}$ - set of each players’ actions
  • $u = {u_1, …, u_n}$ - set of utility functions that expresses preferences

Utility Function (of payoff function) $u_i$

  • $u_i: A_i \mapsto \mathbb{R}$
  • each agent has a preference relation $S_i$
  • agents are rational: therefore $\forall a,b \in A_i: u_i(a) \geqslant u_i(b) \Rightarrow a \ S_i \ b$
  • in other words, if alternative $a$ gives a better payoff than $b$, $i$ will always prefer $a$ over $b$

Representation

These games are typically represented with a pay-off matrix

  • columns/rows represent actions that players can take
  • each cell shows the outcome of game: it lists utilities that all players will receive

For example,

  • consider this 2-player game, with two players $p_1$ and $p_2$
  • each has two strategies: $A_1 = {a, b}, A_2 = {x, y}$
  • the payoffs are shown in the cells of the matrix | $p_2 \to$
    $p_1 \downarrow$ | $x$ | $y$ | $a$ | $[u_1(a,x), u_2(a,x)]$ | $[u_1(a,y), u_2(a,y)]$ || $b$ | $[u_1(b,x), u_2(b,x)]$ | $[u_1(a,y), u_2(a,y)]$ | Another example
  • 2 players, 3 strategies | $p_2 \to$
    $p_1 \downarrow$ | $L$ | $C$ | $R$ | $T$ | (1,0) | (1,3) | (3,0) || $M$ | (0,2) | (0,1) | (3,0) || $B$ | (0,2) | (2,4) | (5,3) |
  • This game can be solved by Iterative Removal
  • At the end the profile $(C, B)$ will be chosen

Types

There are several types of normal form games:

Sources