Fix issue17

[pjz]

Potential fix for #17. Since there's only one value present, I assume a normal distribution with a standard deviation that gets lower the more samples there are of that value.

[CamDavidsonPilon]

should this be 0.5? The quantile is the value p s.t. F(q) = p. If F is equal to 0 before 0, and 1 at and after 0, then I would expect this to be 1. (That's kind of confusing because of all the 0 and 1...)

[CamDavidsonPilon]

I was thinking of a solution like this:

def ...

        if len(self) == 1:                                 
            return int(q >= self.C.min_key())

        for i, key in enumerate(self.C.keys()):

The CDF is a degenerate distribution, so the values should return 0 or 1, depending on if q is above what we have seen thus far. What do you think?

[pjz]

You're probably correct ; my thought process was more like "there's one centroid, they asked for the mean, so clearly they should get 0.5". Not only is your reasoning better, but also since quantile and percentile are inverses, it should be true that x == td.quantile(td.percentile(x)) and x = td.percentile(td.quantile(x)), which fails if it's 0.5 as I suggested, but works if it's 1, as you suggested.

[pjz]

The failure here is a bit inexplicable, not even in the code I changed, I think, and only in py3.3: 2.7 and 3.5 pass fine. Can you prod it to re-run the tests?

[CamDavidsonPilon]

Rerunning - apologies, I didn't see you update the code