The following node is available in the Open Source KNIME predictive analytics and data mining platform version 2.7.1. Discover over 1000 other nodes, as well as enterprise functionality at http://knime.com.
This node is used to construct and view Neighborgrams created from the data input. It does not allow for an interactive clustering (there are more specific nodes, which enable manual or automatic clustering).
The general idea of the Neighborgram data structure is the following: For a selected set of (labeled) objects, construct a neighborhood histogram (called Neighborgram, i.e. a single cell in the node's table view). An indivdual neighborgram summarizes the close vicinity of the respective object according to the selected distance or similarity measure. It usually shows a few hundreds of the closest neighbors (though a different neighbor count can be specified in the dialog), which are colored according their class label (the class coloring needs to be done using a Color Manager node, e.g.). Using the view, the user can select individual neighbors or entire neighborhoods and use other KNIME views to get more details on the selected objects. There are also basic clustering schemes available in the view, which allow the user to navigate between individual neighborgrams and their derived cluster candiates.
The view also supports different "universes", i.e. different measures of similarity. In order to enable this feature, the user needs to use universe marker node beforehand and assign individual columns to universes.
Details on the algorithm have been published in
Michael R. Berthold, Bernd Wiswedel, David E. Patterson
Interactive Exploration of Fuzzy Clusters Using
Neighborgrams,
Fuzzy Sets and Systems, vol. 149, no. 1, pp. 21-37, Elsevier, 2005
0 | Datatable for Neighborgram calculation. |