dmarienko / chaos Goto Github PK

Hi there!
It is a good work on ssa realization here. And I have a few questions about it.
Which version of ssa algorithm has been implemented in your project? What book or work did you follow?

Time serie from 6000 elements with :param dim: embedding dimensions - 720 and :param k: components indexes for reconstruction - 48 and :param e: minimum value to ensure convergence - 50 and forecasting 10 points ahead takes about 25 seconds to predict. So here 2 questions:

1. Can be algorithm optimized somehow using the same environment? (I assume no, because we use scipy.linalg)
2. More valuable. How should I know which values for the parameters are the best? May be some common practice of time serie elements count\elements values dependency or so?