[Neuroscience] I can not calculate alpha factor for LSM algorithm

[zsevil]

I want to design a Neural Natwork with 4 inputs .The input are all 4
combination of 0 1 2 and 3.And the output is a scaler one of 0 1 2 or
3.The inputs are not linear independent.
Then I think that i have to use MADALINE network ,but in the
implementation of it , I faced some problems.My firs case is that I am
not sure about sign function ,i have to use this function if my factors
 are between 0 and 1 .


                                if (0<=n && n<2)
               				n=0;
				if (2<=n && n<6)
					n=1;
				if (6<=n && n<10)
					n=2;
				if (10<=n && n<=12)
					n=3;

It is'nt a usual sign function ,I think .can I use it?

And my another problem is that I don't know that what is error,really?
For example  I have a simple 4-4-1 network ,that in first layer has 4
neron and in last layer has 1 neron.
In algorithm steps ,has been wroten that we must count the outputs
those are against of ideal output ,in each level this is error! but we
can't compare arrors in a layer with 4 neron with another layer with 1
neron . then I don't know what is the error,I thought the error is
difference of output and ideal output .

and I can not calculate alpha factor for LSM algorithm with these
output ,and Ihave some other problems.


I want to know is my selected network correct ?
with this problems I can not receive to a convergant answer .

Please guide me !