Sebastian Krantz
Sebastian Krantz
Ok, thanks! I will also check a bit further still. But I think if it is of this order - 2.8% performance decline for the integer case with no zeros...
I think though I'll implement your method if I can get it to run similarly fast. Your's is more theory consistent. I just tried to translate the h thresholds [here](https://en.wikipedia.org/wiki/Quantile#Estimating_quantiles_from_a_sample)...
collapse version 2.1.0 now implements your method of weighted quantiles (in C): https://sebkrantz.github.io/collapse/reference/fquantile.html Many thanks again for the excellent note!