Botev bandwidths do not match those returned from original matlab code

[oiegorov]

Hi,

Thanks for a python port of Botev's optimal bandwidth code!

I noticed that bandwidths returned by your code are different from those obtained with Botev's original matlab code though. The bug seems to be on line 67 in kde.py file: you start with maximum stage l=7, but Botev starts with l-1, i.e., i=6. Replacing this line with for s in range(l-1, 1, -1): makes results match.

Cheers, Oleg