This is the P2PU Archive. If you want the current site, go to www.p2pu.org!

Mathematics for Game Designers

My recent threads

You haven't posted any discussions yet.

Recently updated threads

When is a graph view appropriate?

Go back to: General discussion

I was just thinking about when a graph is appropriate to use when thinking about the design of a game and when it isn't.

Now, it's obvious (at least to me) that you could always use a graph to model a game (at the very least, each unique positioning of elements in the game can be classed as the game state, and can represent a node in the game, connected to all states it is possible to reach from there), but this isn't always useful - for example, if you used this 'state graph' representation for Go or Chess, you end up with a huge graph that isn't very useful (it's too computationally expensive to derive any useful info from the graph).

Now, you could find a more suitable graph representation of a game. For example, you could model just the movement of pieces as graphs. On some board games, this is quite boring (monopoly, for example, is an extremely simple graph - 36 vertices each of degree 2) and on others this again gets quite complicated (e,g, with chess, you would need a separate graph for each piece with different connections and it would keep on changing depending on the positioning of the other pieces). However, I believe there is a 'sweet' spot in this view where you have games with more complicated movement that monopoly but simpler than chess where graphs are a much more suitable tool for thinking about the game. Some thoughts in this area lead me to the games of Cluedo and Arkham Horror. In the case of cluedo I'd anticipate simplifying it down a bit to remove the hallways to make a useful graph. With Arkham Horror (a very niche game based off the Cthulhu mythos) there is a much more direct graph transaltion - the entire playing board is a graph with city streets connected to each other and 'play areas' and the graph is updated through the game with new connections as portals open to other dimensions. There are also some areas connected to all the streets (e.g. the Sky) and some that aren't connected at all (the outskirts). Even movement in the game isn't done by dice rolls, but by a character specific set amount of 'nodes' that you can move in a turn. To me, looking at this game as a graph is extremely sensible for this game, but for some others it is less of a useful thought process.

Now, in the case of video games, things become a lot more complicated. Many of these are trying to mimic real-time continuous gameplay, and graphs are inherently discrete rather than continuous (unless you want to deal with infinite graphs with each discrete element being an infinitesimal amount of continuous gameplay... nah, I don't want to do that either :) ). But there are places where graphs are a useful simplification of smaller portions of the game. E.g. path finding used to be done with a graph of 'waypoints' on a map. Some games did straight-line travel between these points, other more sophisticated ones did proper route finding (the simple versions had enemies falling off cliffs at times, if the graph node was just slightly off position).
Also, a graph (or rather tree) can be constructed of the space in the game. In a 2d game, you can construct a quad-tree, where the entire map is divided into 4 sections, then each of those is divided into 4, and so on down until you have a way of referring to parts of space and traversing the tree in a more useful manner than if it was just a simple grid. This can be expanded into the 3d case, where you instead construct an octtree and divide each 3d region into 8 sub-regions, and so on down hierarchically.

Also, at a meta level, the entire game can be considered a simple state graph. The game goes from an initial 'loading' state to a 'menu' state. From this the graph can go to different menus, or to the game, or quit the game. Within the game, there is also a paused state, from which you could reach some menu items, or you can leave the game and go back to the main menu, or quit.

I guess the main thing I'm driving at here is what other people think of as useful times to visualise some game behaviour as a graph. Maybe with that in mind, we can start looking at these specific graphs and working out if they have anything in common, or if there are categories within them, and getting useful maths for the various categories.

Or maybe others can think of ways of making what I class as unwieldy graphs (e.g. the state graph of chess) more useful :)

Joe Corneli's picture
Joe Corneli
Tue, 2011-02-08 00:44

Hi David:

Excellent post!

Regarding the unweildy graphs: it might be interesting to look at the ways people *do* study Chess in practise, e.g. what was done by the Deep Blue team?

My guess is that things that are unweildy are often "pattern-matched" into simpler categories. This (and the related topic of "heuristics") I haven't thought to emphasise in the current course -- but it's certainly worth knowing that they are there.

I would also say that in video games, movement is always taking place in a graph -- not with "infinitesimal" pieces, just really, really small ones. Think about the example of PacMan or Asteroids. The movement in Asteroids is on a "mesh" on the surface of a torus. It's meant to make the player feel like the ship is really moving on a torus, but it's just a simulation. In PacMan the map is like a version of "Monopoly" but with a lot of long loops in it!

Anyway, I think your post nicely sums up the spirit of "graphs for game designers". I particularly liked the "Arkham Horror" example.

Joe

David Workman's picture
David Workman
Tue, 2011-02-08 01:14

Thanks Joe... I just had a few thoughts I wanted to get out of my head when I woke up this morning :)

Reading back over it, I think the point I was aiming at was more at when is a graph useful rather than when it's possible, and what made me think that was with the snowman example... the behaviour can be well defined with a graph in the abstract, but if I were to try and encode the actual program using graph data structures (rather than doing a form of differential equation in the code) it would make the program a lot more complicated.

The same is true for something like Asteroids - while the movement can be modelled as a graph if desired, the actual implementation is very far from that (having done a few asteroids programs in the past, I know it's not something I ever needed explicitly in my data structures :) ). The same isn't true of things like game state or path-finding though, which have good implementations that are much more like a graph (although sometimes with some different names for the constructs... state and transition rather than vertex and edge for state machines for example).

Your thought on deep blue is a nice one to investigate. I've read a very little about chess solving in the past, but never in any real depth. I know deep blue was 'trained' which makes me think it involved a neural network at some level (which is another area that I believe maps very well to graph structures). That could make an interesting case study, if there's decent material out there on it :)

Which just led me onto a segway thought - there are actually two different areas of study here. There is the area of study for analysing games using graphs (e.g. move trees in chess) that can then be subjected to mathematical analysis, which I suspect is very much game theory... then there's the area of study for actually using graph structures in game creation - to me, this would be the area of study where an actual graph implementation (rather than differential equation solvers or simple updates of position and velocity variables, for example) makes sense to realise some behaviour. There's overlap, obviously, such as when doing analysis on a game using graph theory leads to deeper insights into the game's behaviour (which may then ironically lead to an implementation that doesn't use graphs... kind of like finding roots for cubic equations requires complex numbers, but ends up with real roots :) )

I'll leave my train of thought there so someone can pull me back on track though ;)

Joe Corneli's picture
Joe Corneli
Tue, 2011-02-08 01:34

Just to hammer a little on Asteroids: the fact that it can be modeled as a graph is good enough to satisfy me from a "game designer" standpoint. Implementation is mostly left as a separate issue here.

That said, if we had a great "graph library", then maybe graphs would be used in more implementations, too. (Not that that would be good or bad.)

AND, one more thing! - if you chase all the way into data structures, you can even find graphs there (pointers, linked lists).

But yes, as you say, there can frequently be other ways to think about things that are more useful!!