Neural networks are giving me a headache!

Firstly, I'm trying to persevere without really understanding their operation. But I'm using them with some success for my problem. I'm just using fitnet. My question is can the output of the neural network be constrained? I'm currently using it to measure the depth of an artifact and I'm getting readings of upto 360% and as low as -40%. Both impossible!

I know I can, and I am, put limits on the output of the neural network, anything above 80% becomes 80% etc. But I'm wondering if I would get better results by constraining the network itself.

Also, can the transform function be changed in fitnet?

Many thanks

Answer 1

[Greg Heath]

 Percentages of What ???

 What are inputs and outputs?
 Matrix dimensions? 
 minmax values?(Fractions instead of percentages are usually better)

 Output values are constrained by using tansig or logsig in the output layer.

Thank you for formally accepting my answer

Greg

[jlt199]

I'm using the NN to approximate the depth of an artifact. The depth is given as percentage of material lost, eg 50% depth means the artifact is half the depth of the material.

So I want to be able to constrain the output to between 0 and 100%.

Cna I use the logsig or tansig functions on the output layer while using fitnet, or do I need to change to a more generic network function.

Thanks