I'm doing my research with raw code of ANFIS. I didn't use anfis toolbox in matlab. N now im in the way of testing my ANFIS network.

Ok, Proccess of ANFIS consist of 2 passes. Forward pass n backward pass. Where the forward pass uses least-square method, while the back pass uses gradient descent. In the learning phase, both forward n backward are implemented in every epoch.

What i'm askin is, how about in the testing phase? I think, there'll be one pass which is forward pass. But the problem is, do we still use least-square method? And if it's not, how do i get the value of consequent parameters?