deletions | additions       diff --git a/In_this_project_we_adopted__.tex b/In_this_project_we_adopted__.tex  index fdaa6fc..559510c 100644  --- a/In_this_project_we_adopted__.tex  +++ b/In_this_project_we_adopted__.tex 
...
In this project, we adopted the clustering algorithm described in \cite{NIPS2008_3478}. This method relies on a notion of stabilities of a data point also introduced in \cite{NIPS2008_3478}. Depending on the whole set of distances each data point is more or less liable to be a cluster center. Those distances can be penalized so that a point that was a good candidate to be a center become poor one.  The more a data point has to be penalized, the more stable he is. Thus in other words, the  stability of a data point measures how much we need to penalize that a data  point has to be penalized  such that it can no longer be chosen as a center in an optimal solution of the problem. Thanks to a measure of this stability this algorithm managed to cluster data points based on an arbitrary metric since the single input of the algorithm is the matrix of distances of the data points. This quality is essential for our case since it allows us to built a relevant metric for trajectories linked to the dependencies between the players' trajectories throw the time. We will described in more details the metrics used in the next part. Furthermore this algorithm is little sensitive to noise and initialization besides it selects automatically the number of cluster that can be adjusted by a penalization constant which affect the selection of new cluster's centers: the higher this constant the less clusters it tends to find.