As in documentation, the anfis function takes in N rows of input features, and an N+1 row consisting of the output determined from the N rows. The input I have consists of 4 features, and one of these features' delayed version is the output. Something of the form:
x(t) = f(x(t-1), x(t-2), x(t-3))
The x here is a four dimensional vector corresponding to the 4 features for every input. I wanted to ask if doing the following would be any different:
Make a huge multidimensional matlab array where each input is a 4 dimensional vector, then feed this to anfis as the input data (3 rows of input). The output data would also be a 4 dimensional vector.
Make an Nx(number of obs) 2D array with elements of x(t-1) and so on represented as row elements, so effectively there are 12 rows as input. Here the input seems to be simple to give, but how do I get a 4 dimensional output?
Will there be any difference in the trained model based on how I format this input?
