Network Training :: Applications (Neural Network Toolbox)

Neural Network Toolbox

Network Training

The above neuron is trained incrementally with adapt. Here is the code to train the network on input/target signals P and T.

```
[net,a,e]=adapt(net,P,T);
```

Network Testing

Once the network is adapted, we can plot its output signal and compare it to the target signal.

Initially, it takes the network 1.5 seconds (30 samples) to track the target signal. Then, the predictions are accurate until the fourth second when the target signal suddenly changes frequency. However, the adaptive network learns to track the new signal in an even shorter interval as it has already learned a behavior (a sine wave) similar to the new signal.

A plot of the error signal makes these effects easier to see.

Network Initialization Thoughts and Conclusions