A numerical sequence is given: (x0, ..., xq), where xi = f(t+i*h). The implemented model, after training on a sample of L = q-p images (xk, ...,xk+p), where p < q and k = 0, ... q-p-1, whose reference values are xk+p+1, should provide prediction of the p+i-th value (i> 1), for an arbitrary sequence of p+1 values. The model must provide scaling of the values of a given sequence for any range, if the activation function used so requires.
- the ability to specify p explicitly;
- possibility to specify the maximum allowable root-mean-square error of sampling;
- possibility to specify maximum admissible step of training;
- the ability to specify the maximum allowable number of training iterations;
- the possibility of specifying on/off the mode of obligatory zeroing, separately for the first and for every following iteration, of the context neurons both at training as well as during prediction;
- ability to automatically predict n values from p+2 to p+1+n;
- the possibility to display the values of weights and thresholds for each layer for the current iteration;
The sequence X of the k=q+1 length (where q > 0), by which this network will be trained.