To continue with the example, the k-means algorithm is applied with nClusters = 2 and the output is: BestPartition = 2 < maxDev, so the clustering loop stops. 10 geneity criterion, which is the sum of the squared errors (line 13). As mentioned in Section 4.1.2, the comparisons of the positions take into account a certain amount of margin, which is the reason why the xAllenInterval of the relation between nameField and passwordField is EQUALS although the projection of the coordinates in the X axis is not exactly the same for both widgets. This unique group is assigned the closeness level of 1.
The number of iterations, Num Iterations variable, is by default 20, and we keep the best solution according to the intra-cluster homo- Since all the distances between the nodes are more or less similar, the clustering algorithm groups all the distances in just one cluster. In order to obtain a better clustering, the algorithm is executed multiple times (lines 11 to 16) with di ff erent random starting conditions.
Because k-means is a heuristic algorithm, it is very fast, but it could fall into a local maximum. This condition is that the standard deviation of every cluster is less than maxDev (line 17). We therefore apply the k-means algorithm several times (lines 6 to 20), while increasing the number of clusters in each iteration (line 10) until the stop condition. However, we do not know the number of clusters a priori.
Given that k-means is a divisive algorithm, the number of clusters must be passed as a parameter. In order to perform the clustering of distances, we have selected the k-means algorithm (line 12), with the euclidean distance as a similarity function.
0 kommentar(er)
