Date:         Fri, 6 Sep 1996 10:16:53 GMT
Reply-To:     "A.O.Schweikert" <101557.3650@COMPUSERVE.COM>
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         "A.O.Schweikert" <101557.3650@COMPUSERVE.COM>
Organization: CompuServe, Inc. (1-800-689-0736)
Subject:      SAS: TNN3 macro for neural nets

I'm using SAS TNN3-macro for solving a problem with neural
nets.The problem:
I have two sets of input variables (A) and (B).
I have an output variable Y which is a function of (A) and only
(A):  Y=f(A).
And I have a variable Z which is a function of (A) and (B):
Z=g(A,B).
I want to predict Y and Z simultaneously in one net.
 
        How do I construct the net ?
 
I had tried the following structure:
 
Use 2 input layer.sets I(A) I(B).
Use 2 hidden layer sets H1 H2.
Use the two output layers O(Y) O(Z).
Then connect: I(A)-H1 /  H1-O(Y) /
                     I(A)-H2 / I(B)-H2 / H1-O(Z) / H2-O(Z)
 
This gives nice estimates for Z but constants for Y.
 
I'd appreciate your hints.   TIA   Albrecht