group_scalar seems inappropriate as integers are also scalars. renamed to group_float (is actually usually of double precision). think about better name, types should have a t prefix. tgroupFloat?