fitnet regression plot shows no relation between output and target values

I am training an ANN using the fitting tool. I have 24 inputs, 4 outputs, 1 hidden layer with 15 neurons. Some of my target data is negative some is positive. When I use only positive training data I get regression values ~.99 and a perfectly linear relationship between output and target. All of the positive data values are between .97 and .99 and negative values are between -.97 and -.99. When I use positive and negative target data the plot shows two vertical lines perpendicular to the x-axis at -1 and 1 and the values on the y axis vary between -.5 and 1.5. I can add my code if it would help but it's the code produced by the fitting tool wizard. The 24 inputs are x and y coordinates and the target data are quaternions, but I'm only interested in the scalar part for this query. I can upload a snap of the plots too if it would help.

Answer 1

[Greg Heath]

The regression plot will only show output 1 vs input 1.

Obviously, there are 95 other plots for the other I/O combinations, some of which are probably binary.

If interested, you can use the command line to explicitly plot any other I/O combination.

Otherwise just rely on the overall normalized mse

 NMSE = mse(t-y)/mean(var(t',1))

Hope this helps.

Thank you for formally accepting my answer

Greg

Answer 2

[padraic moriarty]

Thanks Greg, I think my ANN may have been trying to classify the output. I also think the ANN may have over-trained and is not able to generalize. I will train again with a broader training set and see how it looks.