Clustering Sequences_Old Kiwi
Grouping similar objects in a multidimensional space. It is useful for constructing new features which are abstractions of the existing features. Some algorithms, like k-means, simply partition the feature space. Other algorithms, like single-link agglomeration, create nested partitionings which form a taxonomy. Another possibility is to learn a graph structure between the clusters, as in the Growing Neural Gas. The quality of the clustering depends crucially on the distance metric in the space. Most techniques are very sensitive to irrelevant features, so they should be combined with feature selection.
