© William W. Armstrong, 1999

We import the data set we held back, “test.txt”, into the test set window.

Then by using the “Evaluate ALN” button, we evaluate the ALN at all the test points. In order to have ALNBench calculate the error for us, we can alternatively set the learning rate to 0 and train. We see that the test error is 0.37. We know that this result is about as good as we can ever attain. The chart also gives us the impression that the ALN has fitted the information in the data samples, but not the noise.

This concludes our experiment that demonstrates the simple process of applying machine learning to a set of data points containing noise.

BACK TO START (OR KILL WINDOW WITH X)