A DSEL for experimental game theory (also called evolutionary game theory and behavioral economics). Supports defining games and strategies, then executing them repeatedly in order to collect and observe the results.

Hagl provides built-in support for standard game representations, such as normal and extensive form, and constructs for defining games in terms of the manipulation of a shared state (as in tic-tac-toe). New game representations can be easily added by instantiating a type class.

A monadic strategy DSL supports concise and vaguely English-like definitions of strategies for playing these games, usually iteratively.

While Hagl provides some basic game analyses, its primary focus is simulation and experimentation. Games can be executed and strategies can be pitted against each other for repeated play and in various kinds of tournaments.

To play with the examples, load them into GHCi by running ghci Hagl.Examples from this directory.

There are examples of Normal Form, Extensive Form, and State-Driven games.

For normal form games, you can view the payoff matrix by just evaluating the game in GHCi.

For example:

>> pd
    |  C  |  D 
  C | 2,2 | 0,3
  D | 3,0 | 1,1

You can view the game tree for any discrete game by using the gameTree function.

>> gameTree pd
Player 1
+- C -> Player 2
|  +- C -> [2,2]
|  `- D -> [0,3]
`- D -> Player 2
   +- C -> [3,0]
   `- D -> [1,1]

The function execGame is used to execute a game. It takes a game, a set of players to play the game, and a function to execute within the game execution monad.

Execution functions include:

• step - Process a single node in the game tree.
• once - Run through the game a single time to completion.
• times n - Run the full game n times.

There are also many printing functions available for inspecting the execution state. These can be found in Hagl.Print.hs. Some examples include:

• printMovesFromHere - Print the available moves from this location in the game tree.
• printTranscripts - Print the transcript of all completed games.
• printScore - Print the current score.

The execution and printing functions can be executed sequentially via bind operations. The best way to illustrate how this all works together is a few examples.

The following processes the first node in the game tree (Player 1's decision), then prints the moves available to Player 2.

>> execGame pd [tft, pavlov] (step >> printMovesFromHere)

The following runs the game four times, prints the transcript of all iterations, and then prints the current (in this case, final) score.

>> execGame pd [tft, suspicious] (times 4 >> printTranscripts >> printScore)

And finally, the following runs the game three times, prints the score, then runs 100 more iterations and prints the final score.

>> execGame pd [ccd, grim] (times 3 >> printScore >> times 100 >> printScore)

There is also a "tournament" facility that eases running many combinations of players and comparing their final scores. The tournament running functions can be found in Hagl.Tournament.hs.

All tournament functions return their results as a list of player names paired with scores. This can be printed in a nice way by passing this result to the printResults function, as shown below.

>> roundRobin pd 2 [tft, titForTwoTats, grim, suspicious, pavlov] (times 100 >> printScore) >>= printResults

For each pair of players from the list, this will run the iterated prisoner's dilemma 100 times, and print the score for those players. At the end, the total scores of all players involved will be printed in sorted order.