lyap_r checking number of points

[alangiontra]

Where does the check for "enough data points to continue" come from? What I've noticed:

Instead of using a KNN search (as the Rosenstein algorithm does) the lyap_r function replaces distances with Inf, which, although, makes sense when looking for the minimum value that is at least min_tsep away in index. However, this code seems to ignore the case where min_tsep is the same size as ntraj. If the minimum separation is too large, MATLAB's_ lyapunovExponent, defaults the minimum separation to ntraj. When this happens, there are no nearest neighbors, so to find a neighbor, the algorithm takes the furthest neighbor. This case is not reflected in this code.

I have found that as long as min_tsep <= ((len(data) - (embed_dim - 1) * lag) - expan_rng + 1) - 2) /2, then the algorithm will run. In fact, nolds and MATLAB match for these cases, but not when min_tsep is greater.

The default for nolds is 25% of the length of the data, which ensure the above condition, but MATLAB's KNN overwrites the minimum separation to (len(data) - (embed_dim - 1) * lag) - expan_rng + 1), which is the max index.