A natural way to think of player movement in arbitrary
games is the path they take between dierent game states.
A play trace is then a path that a player takes in this high-
dimensional game state space. In order to visualize these
paths, we use Classical Multidimensional Scaling (CMDS)
[3, 17] to represent observed game states in two dimensions.
CMDS takes an input matrix that species the distance be-
tween every pair of states and outputs a set of points, which
are positioned to minimize a loss function on all interpoint
distances. Therefore, the transformation will place states
that are similar close together and states that are dissimilar
far apart, making it easy to see the similarity of states that
are visited by many players. We used the MDSJ library for
Java [18].
Dierent metrics for calculating distances between states
will result in dierent congurations of points after CMDS is
applied. In general, the distance metric should be dierent
depending on the type of game. Additionally, the distance
metric can be adjusted depending on what features of the
game the designer wishes to analyze or what features he
or she wants the state graph to have. For example, if the
distance metric has a component that compares how many
steps it takes to reach a goal state, then it will naturally
cause goal states to cluster together. States from which it
is dicult to reach a goal state will appear far away. This
allows the designer to identify players who are not making
progress and investigate why they are having trouble.
An example of Playtracer's output can be seen in Figure 1,
which shows the state space for one level of an educational
game we have been developing. Playtracer takes in a list of
all of the states that the player visited and a distance metric
that calculates the distance between states, and creates a
graph where the states are vertices and player movements
are directed edges. Here, the yellow state is the start state
and the green state is the goal state. To identify which states
are most commonly visited, the size of a state is proportional
to how many players reach that state.