Need help in doing one step prediction of chaotic time series using local linearly method(Farmer sidorowich method)

I have integrated lorenz system of equation using ODE45 solver with an integration time step ∆t = 0.01 for 25000 time steps.In order to remove initial transient the first 5000 points were discarded and the remaining points were assumed to belong to the Lorenz attractor. Now I have only 20000 points. Out of these first 16000 points I have to form training set and the remaining as test set .

Now k(=25) nearest neighbours , of each point in the test set, should be taken from the training set.Now I want to do single step PREDICTION for each point in the test set using local linear approach (FARMER SIDOROWICH TECHNIQUE).

[dpb]

And the specific Matlab-related question is???

[anurag]

one step ahead forecast using K nearest neighbour method.