fix weight in interpolation

[kunigami]

This is essentially: #52 with extra simplifications.

Intuitively if p is close to t (the mid point of the previous cluster), the weight for c_i.mean should be higher. Whereas right now z1 = p - t is smaller the closer p is to t.

Note that when we do this change, the p == t + k is handled too:

z1 = p - t = t + k - t = k
z2 = t + k - p = t + k - t - k = 0

Also noting that k = z1 + z2:

c_i.mean * z1 + c_i_plus_one.mean * z2) / (z1 + z2) = c_i.mean * k / k = c+i.mean

I added a test case to show case this:

 [1, 10, 10, 10, 10, 100, 100, 100, 100, 1000]

it currently returns 132.0 for p=40.

[kunigami]

eh, sorry I didn't realize I would have to update this test again >.<

[CamDavidsonPilon]

🔥

[CamDavidsonPilon]

Cool! Thanks @kunigami!