Sorted by x; each local fit uses a contiguous window of q = ⌈frac·n⌉
nearest neighbours. Optional delta (default 1% of x-range) skips fits for
nearly-tied x and interpolates — same idea as R lowess / statsmodels.
Complexity ≈ O(n · q · iterations / skip-factor), not O(n² log n).
Cleveland LOWESS (degree-1 tricube, optional robust iterations).
Sorted by x; each local fit uses a contiguous window of
q = ⌈frac·n⌉nearest neighbours. Optionaldelta(default 1% of x-range) skips fits for nearly-tied x and interpolates — same idea as Rlowess/ statsmodels. Complexity ≈ O(n · q · iterations / skip-factor), not O(n² log n).