• No results found

In this chapter, we have presented a survey of literature relevant to our work in this thesis.

Firstly we established all terminology relevant to discussing games as a scientific discipline,

then AI concepts with a specific focus upon MCTS. We then moved on to literature re-

lating to entertaining play experiences, and finally provided an overview of Data Mining

Game Domains and Experimentation

Software

3.1

Game Domains

The following Game Domains were selected for experimentation.

3.1.1

Lords of War

Lords Of War1is a two-player strategic card game by publisher Black Box Games. A board

is used for card placement, and the relative positions of the cards on the board are the main

strategic interest. A player wins when they eliminate twenty of an opponent’s cards, or they

eliminate four of their opponent’s Command Cards. Command cards are significantly more

powerful than other cards, but placing them onto the board carries a risk that they may be

eliminated.

The game board is 7 × 6 squares each of which can hold a single card. Cards have be-

tween 0 and 8 attacks, each with a strength value, and a directionality towards an orthogonal

or diagonally adjacent square (see figure 3.1). Attacks from multiple cards can be combined

Figure 3.1: Bestial Raptor, an example card from the Lords of War game.

tacks which can eliminate (or contribute towards the elimination) of opponent’s cards which

are not adjacent. In regular play, cards can only be placed so as to attack enemy cards, how-

ever Support Cards also have additional placement rules allowing them to be placed next to

friendly cards instead of attacking enemy cards.

On each player’s turn, they are required to place exactly one card, then process combat to

identify and remove eliminated cards, then they have a choice of either drawing a new card

from their deck, or retreating a friendly unthreatened card from the board. The complete

rules of Lords of War appear in Appendix C.

A normal game rarely extends beyond 50 turns, as most moves (particularly strong

moves) result in a capture. Once an average human player has made 25 moves, they have

probably captured more than 20 cards, and thus the game would have completed. Of course

the games can end much sooner if command cards are placed carelessly or last much longer

if players play cautiously. Games with MCTS agents last on average between 30 and 60

length, but normally last between 50 and 120 turns. Our experience with Lords of War

has revealed that it commonly has a mid-game branching factor of 25-50, making move

selection challenging.

Lords of War was chosen for research as it represents a complex game in which players

can adopt a number of different strategies and play styles, which provides a wide range of

behaviour for us to analyse.

3.1.2

Android: Netrunner

Android: Netrunneris a two-player strategy card game published by Fantasy Flight Games2,

which includes elements of bluffing and deception. Netrunner is similar to other popu-

lar card games such as Magic:The Gathering, and is described as an LCG (Living Card

Game [55]).

During a standard match of Netrunner, opponents do not have access to the content of

their opponents deck. Access to such information would provide a substantial advantage

to a player, as they would both be able to predict their opponent’s likely strategy, and also

determine which strategies they are poorly defended against.

Due to the nature of the game, the content of an opponent’s deck is critical strategy

information, and a player who is able to accurately model their opponent’s deck is at a

substantial advantage. There are currently more than 600 cards released for Netrunner, so

accurately modelling a deck is a significant challenge. The combination of the wide number

of choices, plus the complex and specific rules for which cards may be included in decks

makes Netrunner deck construction a highly intricate process.

Netrunner has a well documented rules structure for deck building. Every Netrunner

a Side, an amount of influence and a Faction. There are exactly 2 sides (named Runner

and Corp), and each card in Netrunner is associated with one side and cannot be included

in decks associated with the other side. Identities which are from the corp side must also

include a specific number of agenda points, which are provided corp cards (the specifics of

agenda points are not relevant to this work, other than to recognise that there is a required

number of agenda points for some decks to include, which presents an additional restriction

upon decks.) All non-identity cards also have a Faction and a Influence Cost, the latter of

which describes the amount of influence which must be paid to include the card in a deck

which contains an identity of a different faction. The complete rules of NetRunner appear

in Appendix D.

Netrunner was chosen as a target for research due to the large amounts of imperfect

information, and that bluffing and deception are integral to play, meaning that the content of

an opponent deck is of very high importance.