weighted median

[dhopkins-gweiss]

I think the weight is placed on the ordered values, not on the original values passed in.
E.g.
> roll_medianr(5:3, 3, weight = c(1,0,0))
gives NA NA 3 1xsmallest, 0x4, 0x5
Not NA NA 5 as expected (1x5,0x4,0x3)

My guess is this stems from sorting first without respect to the weights...
` NumericVector copy(x.begin() + offset, x.begin() + offset + n);
std::sort(copy.begin(), copy.end());

double weights_sum = sum(weights);

int k = 0;
double sum = weights_sum - weights[0];

while (sum > weights_sum / 2) {
  ++k;
  sum -= weights[k];
}

return copy[k]; `