This is the second improvement suggested in the google paper. The basic idea is to reduce the range until which LinearCounting is used and to introduce a new range at which a bias is subtracted from the estimate. The thresholds and the bias data used by google can be found in the appendix. It probably makes sense to simply reuse these values and maybe if we have the time calculate our own.
This is the second improvement suggested in the google paper. The basic idea is to reduce the range until which
LinearCounting
is used and to introduce a new range at which a bias is subtracted from the estimate. The thresholds and the bias data used by google can be found in the appendix. It probably makes sense to simply reuse these values and maybe if we have the time calculate our own.