Vector Quantization Network

Documentation Neural Networks Changing the Neural Network Structure Change the Parameter Values of an Existing Network

13.1.4 Vector Quantization Network

Because a VQ network's codebook vectors are divided into several classes, it is slightly more complicated to change the parameters in a VQ network than in an unsupervised network, as is explained next.

Load the Neural Networks package.

In[1]:=<<NeuralNetworks`

In this example, a VQ network with two inputs, two classes, and six codebook vectors is used.

Initialize a VQ network with six codebook vectors.

In[2]:=

Out[2]=

View some information about the network.

In[3]:=

Out[3]=

The six codebook vectors are divided between the two classes, so that three codebook vectors are used to explain the data in each class. The codebook vectors are contained in the first element of the network.

Look at the codebook vectors.

In[4]:=

Out[4]=

You may modify any codebook vector by giving new numerical values to the corresponding parameters in this structure. Then you insert it at the first position of the network.

If you are interested of a particular class you can extract the codebook vectors of that class and display them as a matrix.

Look at the codebook vectors of the second class.

In[5]:=

Out[5]//MatrixForm=

There is one codebook vector on each row of the matrix. You may modify the values of the vectors by changing the corresponding elements of the matrix and inserting the modified matrix on the appropriate position in the network.