Documentation /  Neural Networks /  Vector Quantization /

Too Few Codebook VectorsFurther Reading

11.3 Change Step Length

The option StepLength works for VQ networks exactly as it does for the unsupervised networks. If the default does not suffice, you can supply another function that takes the iteration number as input and gives the step length as output. You can also submit a numerical value, which gives a constant step length. These possibilities are illustrated here.

Read in the Neural Networks package.

In[1]:=

A set of test data is used to illustrate the StepLength option.

Load the data vectors and output indicating class.

In[2]:=

The input data vectors are stored in the matrix x and the output in y. The data format follows the general standard of the Neural Networks package as described in Section 3.2, Package Conventions.

Define a function to give the step length as a function of the iteration number.

In[3]:=

The initialized VQ network is then trained with the new step length function.

Initialize and train a VQ network.

In[4]:=

Instead of supplying a function, you can also submit a constant step length.

Train with a constant step length.

In[5]:=

Too Few Codebook VectorsFurther Reading


Any questions about topics on this page? Click here to get an individual response.Buy NowMore Information