Quantile output not sorted

[cstjean]

On Julia 0.4.5 and the latest release of StatsBase:

y = [0.40003674665581906,0.4085630862624367,0.41662034698690303,0.41662034698690303,0.42189053966652057,0.42189053966652057,0.42553514344518345,0.43985732442991354]
issorted(quantile(y, linspace(0.01, 0.99, 17)))   # false

The quantiles are:

[0.4006335904282823,0.4042892585346196,0.40794492664095694,
0.4114334853955278,0.41488803593114276,
0.41662034698690303,0.416620346986903, # these two not in order
0.4169958482153258,0.41925544332671183,0.4215150384380978,0.42189053966652057,
0.42189053966652057,0.4226741294789331,0.42423675334903477,0.42657350156657636,
0.4327141366637794,0.4388547717609824]

Is that a bug, or is it a conscious decision to define "empty quantiles"?

[andreasnoack]

The difference is smaller than eps() and you are asking for more quantiles than you have data points. Is this a real problem for you?

[simonbyrne]

This is actually a problem in Base, particularly this line.

I think we can fix this by changing

(1-h)*T(v[i]) + h*T(v[i+1])

to

T(v[i]) + h*(T(v[i+1])-T(v[i]))

[cstjean]

It's not a problem for my algorithm, but FWIW it originally happened on a dataset with more unique values than quantiles. I just trimmed it to file the issue.

[cstjean]

Thank you for the quick fix @simonbyrne and @andreasnoack