||Reference value (R) (IEEE 32-bit floating-point value)
||Binary scale factor (E)
||Decimal scale factor (D)
|20||Number of bits used for each packed value|
|21-24||Pre-processing parameter (B) (IEEE 32-bit floating-point value)|
(1) The template is approriately designed for data sets with all non-negative values and a wide variability range (more then 5 orders of magnitude). It must not be used for data sets with negative values or smaller variability range.
(2) A logarithm pre-processing algorithm is used to fit the variability range into one or two order of magnitudes before using the simple packing algorithm. It requires a parameter (B) to assure that all values passed to the logarithm function are positive. Thus scaled values are Z=ln (Y+B), where Y are the original values, ln is the natural logarithm (or Napierian) function and B is chosen so that Y=B>0.
(3) Best pratice follows for choosing the B pre-processing parameter.
(a) If the data set minimum value is positive, B can be safely put to zero.
(b) If the data set minimum is zero, all values must be scaled to become greater than zero and B can be equal to the minimum positive value in the data set.
(4) Data shall be packed using Data template 7.