AI Matrix Payoff game

• Description
• Target
• Author
• Lab Account
• Compilation
• Run
• AI Algorithm
• Additional functionalities

Description

This program is a game where there is a payoff matrix (NxN) filled with Vector2() This matrix is symmetrical The MainAgent will ask the players of the tournament about their movements, each player do the movement without knowing the current enemy movement. After the movement the MainAgent will inform to the players about the movement results

Target

This program is developed targetting to learn to program an intelligent AI with JADE

Author

Carlos Daniel Garrido Suárez xuzon69@gmail.com

Lab Account

psi27

Compilation

Just compile it with like you would compile any java with the classpath jade

Needs java 1.8

Run

You must use

java -cp .:jade.jar jade.Boot "referee:psi27_MainAg;Player1:psi27_Random;Player2:psi27_Fixed;Player3:psi27_Intel1"

The only runnable players class are psi27_Random and psi27_Fixed and psi27_Intel1, the psi27_Player is an abstract base class
The psi27_MainAg class invoke my GUI My Intelligent player is psi27_Intel1

AI Algorithm

First the AI will start the match wihout knowing the matrix so it will choose a random value.

After the first movement now it knows some of the matrix and begin the real algorithm:

• First I decide if I have a dominant move

  • If I have one or more decide wich is better (harms the most the enemy) and thats the choosen movement

  • If no I start the discover matrix algorithm

• Discover matrix algorithm

  • I see what percentage of each movement I don't know and store them

  • If the percentage surpass my factor is a candidate to know

  • I ponderate the know candidates by known payoff difference

  • If I have a candidate that will be the movement, if no, it is time to do the strategy

• Strategy

  • Get the dominants (it have to be null but just in case), Nash Equilibriums, Max Min strategies

  • If I'm winning I temporary choose a Max Min strategy

  • If I'm losing I temporary choose the best winner movement, this movement is the movement where I win or draw the most of all the possible movements

  • Afer all this strategies it is time to look at the past

    • If the enemy player was playing the same movement for a factor turns in the recent past I choose the Better Response for that, if not don't do anything

    • This turns factor is initialized to 4 and if I'm wrong in my assumption I increment that factor

Dominant

My own definition of dominant movement in this game is a movement where you can only win or draw in that movement

Additional functionalities

• The Main Agent is searching for players, if he founds a new player he will see if the player has another service will type that player with the name of that service

• You can turn on / off the log and a detailed log version

• We can change properties directly in the GUI

• We can rename the players as we want