which, while preserving the streaming nature of the algorithm (for
     parallel computation purposes) is numerically much more stable. Note
     that it does *not* require two passes as a common fix to numerical
     instability does.
ENH: removed the static function that calculates centered moments and
     the regression coefficient from the raw moments since the latter
     are no longer calculated.

     which, while preserving the streaming nature of the algorithm (for
     parallel computation purposes) is numerically much more stable. Note
     that it does *not* require two passes as a common fix to numerical
     instability does.
ENH: removed the static function that calculates centered moments from
     raw moments since the latter are no longer calculated.
ENH: added G1 skewness estimator.
ENH: updated tests accordingly.

ENH: Added vtkPRandomGraphSource, which generates random, distributed graphs. Fixed copy-on-write semantics for distributed vtkGraph.

BUG: This filter would always create the vtkPolyData with the points' coordinates stored as floats, regardless of the input point coordinate type possibly reducing accuracy.  Changed it so that the point coordinates type of the outputted vtkPolyData is the same as the input type.

ENH: Fixed BUG #6533. Addded support for composite datasets to D3. D3 now iterates over composite datasets ensuring that all processes correctly processes empty blocks as well.

BUG:  vtkVariantCompare had a serious bug in it. Deleted all instances of it and replacing with vtkVariantLessThan which now properly supports heterogeneous types.

BUG:Fixed double *GetBounds() on vtkAbstractMapper3D subclasses. In vtkAbstractMapper3D, GetCenter() and GetLength() call GetBounds and rely on the side effect that it will update this->Bounds, the subclasses implementations were not changing this->Bounds to uninitialized (1,-1,1,-1,1,-1) and were returning some arbitrary (-1,1,-1,1,-1,1) valid bounds instead when the input is null. It could be problematic if the scene has one mapper with a valid input with bounds out of (-1,1,-1,1-1,1) and another mapper with an empty input. Fixed another bug in vtkCompositePolyDataMapper2.cxx where input can be used before checking it is not null.

     option: parallelism calculations will be done via explicit update
     functions.

     meta information with the Assess option.

      BoxPlot test from the build.

     Options (Learn, Validate, and Assess) which can be executed in
     combination. The flow of RequestData() has subsequently been
     altered. Output ports are modified or simply copied from the input,
     depending on the chosen option values.
ENH: statistics engines now have three output ports: 1 for data (unchanged)
     and 2 for meta information. The first meta port (port 1) contains
     the model itself, modified or not in learn mode. The second meta
     port (port 2) contains meta information about the model itself, such
     as quality of fit to the data, information entropy, etc.
ENH: updated all concrete statistics classes and tests accordingly.

BUG: When setting a value in vtkInformation, do not call Modified if the value does not actually change.

ENH: Add export for vtk_TIFFMergeFieldInfo. While I'm in there, alpabetize the list and eliminate duplicate entries. (There were 5 duplicates since the list was not alphabetized.)

     pairs of table columns (as permitted by bivariate statistics)
ENH: updated test accordingly
PERF: removed unnecessary code