Self-Organizing Map Functions :: Advanced Topics (Neural Network Toolbox)

Neural Network Toolbox

Self-Organizing Map Functions

There are two kinds of functions that control how neurons in self-organizing maps respond. They are topology and distance functions.

Topology Functions

Topology functions calculate the positions of a layer's neurons given its dimensions.

Once defined, you can assign your topology function to any layer of a network. For example, the following line of code assigns the topology function yourtopf to the second layer of a network.

net.layers{2}.topologyFcn = 'yourtopf';

Your topology function is used whenever your network is trained or adapts.

[net,tr] = train(NET,P,T,Pi,Ai)
[net,Y,E,Pf,Af] = adapt(NET,P,T,Pi,Ai)

To be a valid topology function your function must calculate positions pos from dimensions dim as follows,

```
pos = yourtopf(dim1,dim2,...,dimN)
```

where:

dimi is the number of neurons along the ith dimension of the layer.
pos is an matrix of S position vectors, where S is the total number of neurons that is defined by the product dim1*dim1*...*dimN.

The toolbox contains an example custom topology function called mytopf. Enter the following lines of code to see how it is used.

help mytopf
pos = mytopf(20,20);
plotsom(pos)

If you type that code, you get the following plot.

Enter the following command to see how mytf is implemented.

```
type mytopf
```

You can use mytopf as a template to create your own topology function.

Distance Functions

Distance functions calculate the distances of a layer's neurons given their position.

Once defined, you can assign your distance function to any layer of a network. For example, the following line of code assigns the topology function yourdistf to the second layer of a network.

net.layers{2}.distanceFcn = 'yourdistf';

Your distance function is used whenever your network is trained or adapts.

[net,tr] = train(NET,P,T,Pi,Ai)
[net,Y,E,Pf,Af] = adapt(NET,P,T,Pi,Ai)

To be a valid distance function, it must calculate distances d from position pos as follows,

```
pos = yourtopf(dim1,dim2,...,dimN)
```

where:

pos is an matrix of S neuron position vectors.
d is an matrix of neuron distances.

The toolbox contains an example custom distance function called mydistf. Enter the following lines of code to see how it is used.

help mydistf
pos = gridtop(4,5);
d = mydistf(pos)

Enter the following command to see how mytf is implemented.

```
type mydistf
```

You can use mydistf as a template to create your own distance function.

Learning Functions Network Object Reference