Neural Network Clustering

Hello, everyone. I am sorry if this question has been asked before, but I couldn’t find the answer. I am brand new to Matlab, so please keep hat in mind.:) I am using the clustering element of the Neural Net Toolbox and have a question. I would like to cluster along a “dominant column” instead of having all dimensions of my feature vector being equivalent. That is, let’s say that I have a matrix of ten rows and ten columns. The “dominant column” can have values from zero to three. What I would like to do is to have four clusters emerge from this. One cluster will always have the dominant column set to zero, another cluster would have the dominant column set to one, the third would have it set to two, and the fourth would have it set to three.

How can I make sure that all my clusters are formed around the unique values of the dominant column?

Respectfully, Kevin

[Greg Heath]

I don't understand. Please give an example.

Greg

[Kevin]

Imagine the following data set: (0 3) (0 4) (0 5) (1 6) (1 7) (1 8) (2 9) (2 10) (2 11)

I would like to form three clusters

In the first cluster would be the rows (0 3) (0 4) (0 5)

In the second cluster would be rows (1 6) (1 7) (1 8)

and in the third cluster would be rows (2 9) (2 10) (2 11)

[Greg Heath]

I do not see why the points would not be clustered as you wish.

None of the 2nd coordinates overlap.

Hope this helps.

Greg

[Kevin]

Ah, My apologies for the poor example. Perhaps a better example would be:

(0, 0) (0, 1) (0, 2) (1, 0) (1, 1) (1, 2) (2, 0) (2, 1) (2, 2)

I would like the clusters to match the X coordinate. So, the clusters would appear in the order the points were listed. That is, Cluster 1: (0, 0) (0, 1) (0, 2)

Cluster 2: (1, 0) (1, 1) (1, 2)

and Cluster 3: (2, 0) (2, 1) (2, 2)

So the clusters cluster around the X coordinate, what I am calling the "dominant column". I call it this because while in this example the dominant column is the first column, it could be the third, or the sixth, or the twentieth column of the data set.

Answer 1

[Greg Heath]

This is supervised clustering since you know what vectors belong to each cluster. It falls into the category of classification and can be implemented with patternnet,tansig,softmax and trainscg.

For practice try to run the classification and pattern recognition examples and demos in the documentation.

Hope this helps.

Greg