cMap
Screen Shots
While cMap does not produce any interesting (graphical) output itself, another helper program is able to visualize the neural net.
By examining the final values of the reference vectors contained in the neurons of the map, we can create a landscape representation of the clusters computed by the map. The landscape metaphor is easily understood--the document clusters are represented by the "valleys" in the landscape, and the boundaries between the clusters are represented by the "mountains". Another important aspect of the visualization is that the heights of the mountains represent the degree of dissimilarity between neighboring clusters. For example, two clusters that are next to each other on the map are somewhat similar to each other by virtue of their placement; a low mountain range between them shows that the two neighboring clusters are slightly dissimilar, however, if there is a high mountain range between them, it would indicate that their dissimilarity is quite large, though not so large as to move the clusters apart from each other.
Three clusters
This is a graph of a sample data set containing 500 points grouped into three clusters:
|
As can be seen, there are many points tightly grouped together, and a few points floating around that are outliers.
The map produced by cMap for this dataset can be seen here:
|
One can see that the lines drawn on the map (in red) indicate that there are definitely three groups of points in the dataset.
Five clusters
This next graph is of a sample data set containing 1000 points grouped into five clusters, and the cMap output:
|
|
Again, there are many points tightly grouped together, and a few points floating around that are outliers.
Ten clusters
Here is 2000 points grouped into ten clusters, and the cMap output:
|
|