newlin (Neural Network Toolbox)

Neural Network Toolbox

newlin

Create a linear layer

Syntax

net = newlin

net = newlin(PR,S,ID,LR)

Description

Linear layers are often used as adaptive filters for signal processing and prediction.

net = newlin creates a new network with a dialog box.

newlin(PR,S,ID,LR) takes these arguments,

PR -- R x 2 matrix of min and max values for R input elements

S -- Number of elements in the output vector

ID -- Input delay vector, default = [0]

LR -- Learning rate, default = 0.01

and returns a new linear layer.

net = newlin(PR,S,0,P) takes an alternate argument,

P -- Matrix of input vectors

and returns a linear layer with the maximum stable learning rate for learning with inputs P.

Examples

This code creates a single input (range of [-1 1] linear layer with one neuron, input delays of 0 and 1, and a learning rate of 0.01. It is simulated for an input sequence P1.

net = newlin([-1 1],1,[0 1],0.01);
P1 = {0 -1 1 1 0 -1 1 0 0 1};
Y = sim(net,P1)

Here targets T1 are defined and the layer adapts to them. (Since this is the first call to adapt, the default input delay conditions are used.)

T1 = {0 -1 0 2 1 -1 0 1 0 1};
[net,Y,E,Pf] = adapt(net,P1,T1); Y

Here the linear layer continues to adapt for a new sequence using the previous final conditions PF as initial conditions.

P2 = {1 0 -1 -1 1 1 1 0 -1};
T2 = {2 1 -1 -2 0 2 2 1 0};
[net,Y,E,Pf] = adapt(net,P2,T2,Pf); Y

Here we initialize the layer's weights and biases to new values.

```
net = init(net);
```

Here we train the newly initialized layer on the entire sequence for 200 epochs to an error goal of 0.1.

P3 = [P1 P2];
T3 = [T1 T2];
net.trainParam.epochs = 200;
net.trainParam.goal = 0.1;
net = train(net,P3,T3);
Y = sim(net,[P1 P2])

Algorithm

Linear layers consist of a single layer with the dotprod weight function, netsum net input function, and purelin transfer function.

The layer has a weight from the input and a bias.

Weights and biases are initialized with initzero.

Adaption and training are done with trains and trainb, which both update weight and bias values with learnwh. Performance is measured with mse.

See Also

newlind, sim, init, adapt, train, trains, trainb

newhop newlind