Implemented an algorith to ensure a reasonable error for histogram with a small number of raw events

Release notes

Details

When estimating background from low stat MC or from low stat DATA CR, there could be a big underestimation of the statistical error if events fluctuate downward, to mitigate this effect the error is recomputed in a new histogram with zero entries and an error corresponding to the 68% C.L. limit of a poisson distribution with the given number of observed event, the current statistical error is subtracted in quadrature from this value. If for example 0 events are observed, the limit is slightly larger than 1. This value is multiplied by an average weight that needs to be provided by the user. The average weight could be a fakeFactor value for example, or the luminosity scale factor, or the average weight of the final histogram if only few bins are affected by this issue.

An example on how to use this functionality is below, here Fake is the original data driven estimate for a fake background, eFake is a new sample that will have zero entries and the error computed as described above.

       +Fake {
            <Type = B, Path = "bkg/$(channelSys)/$(campaign)/ddFakes">
        }


        +eFake {
            <Type = B, Path = "bkg/$(channelSys)/$(campaign)/ddFakes", inflateStatUncertaintyLowYield.averageWeight = 0.079>
        }