Fixed furthest_sum initialization error

[mpmbq2]

The furthest_sum procedure occasionally generates an initial index value that is exactly the length of I, which causes an IndexError. To fix it, I changed the index selection step to force it to select a valid value from I.

[ulfaslak]

Nice catch. I think you can simplify the line to (no need to cast I as a numpy array):

i = furthest_sum(X[:, I], noc, [np.random.choice(I)])

[ulfaslak]

God, now I look at it, it's hard to understand how I wrote it that way. Must have assumes np.random.rand() can never be one.