LPQ Precision Statistics should use Geometric rather than Arithmetic Mean

[SSoelvsten]

With the prediction precision ratio (PPR) we tried to accumulate some average. Yet, essentially we have done so with something akin to an arithmetic mean. What we should have done is to compute a geometric mean.

   EXP(SUMPRODUCT(weights, ln(samples)) / SUM(weights))

Currently we maintain the sum of all weights, the sum of all samples, and the number of samples. Yet, what we need is to also maintain

• [ ] The sum of ln(sample) (without any weight multiplied) for the unweighted PPR.
• [ ] The sum of weight * ln(sample) for the weighted PPR.

If this is done, then we can fix this set of statistics.