SSA parameters about chaos HOT 1 CLOSED

Hi, thank you for reaching out, actually I used some standard algo description as for basis of this impl (seems I've lost links to docs after moving to new notebook).

As for forecast method speed - yeah it's really non optimized in sense of speed. Sometimes algo converges very slow.

As idea to test I could advice you to look at ssa_predict method. There I use last know value from original series as some initial estimation of next point. It's not worst approach especially when your series has some autocorrelations.

So you could take a look at here:

   # here we use previous value as initial estimation
        x = np.append(x, x[-1]) 
        yq = x[-1]

and could try to implement more sophisticated methods for obtainig initial estimation. As example you could use averaged values from recent observations or even estimate AR(I)MA model ang use it's prediction as some initial values.

As for parameters values - what values do you mean ? If you talk about size of lagging space you might use ssa_cutoff_order method. It may give you some initial imaginary about the lagging space dimension.

Hope it helps.

from chaos.