Partial sorting into unsorted groups

[Beliavsky]

Sometimes you don't need the exact ranks of elements in a vector but what quantile each value is in. For quintiles you would return an integer from 1 to 5 corresponding to each element. Such partial ranking ought to be faster than full ranking, especially when the number of groups is small compared to the size of the vector. The interface could be

pure function quantile(x,ngroups) result(igroup)
real, intent(in) :: x(:) ! data to be grouped
integer, intent(in) :: ngroups ! # of groups
integer :: igroup(size(x)) ! returned values will be between 1 and ngroups, inclusive       
end function quantile

There was a Stack Overflow thread Efficient algorithm of partial sorting into N unsorted groups

In finance, researchers commonly group stocks into quintiles or deciles based on some characteristic, such as price/earnings ratio, and then study future performance by quintile. The exact ranks of stocks are not needed for such studies.

[milancurcic]

I often use the histogram function in NumPy and matplotlib, and this seems somewhat related. The difference is that this function would return the index of the bin that each elements belongs in, whereas histogram returns the number of elements in each bin. Though I haven't had use for this myself, it sounds useful.