VORONOI DIAGRAM BASED DIMENSIONAL ANCHOR ASSESSMENT FOR RADIAL VISUALIZATIONS - Information Visualisation: Techniques, Usability and Evaluation - page 257

Fig. 13.8. Khan SRBCT data with dimensional anchors (top five per class) selected
by mean ratio method. Dimensional anchors are placed using the CDL+AAM
method [9] explained in Section II .
II. Selecting Dimensions with the Mean Ratio
Zhou et al. 's efforts [5] were to illustrate the value in using the mean ratio
as a metric for dimensional reduction. In that application the dimensions
are human genes. The intended use of these techniques is to provide
researchers with data exploration tools to determine which genes may be
most closely linked to any of a variety of cancers and to generate visual
classifiers. The Khan SRBCT data set used [5] is broken up into four
classes corresponding to four childhood tumour categories: Neuroblastoma
(NB), Rhabdomyosarcoma (RMS), Burkitt Lymphomas (BL), and Ewing
Family Sarcoma (EWS). We make one deviation from Zhou et al. 's
methods as we use an algorithmic technique for placing dimensional
anchors. The dimensional anchors in their work were placed arbitrarily. In
the description of this algorithm note that “above” and “below” are made
in the context in which the labels are placed for reading by English
language readers about the RadViz circle. Our “Alternating Anchor
Metric” (AAM) method is used throughout this paper to: (a) sort the
dimensional anchors by the selection metric used (e.g. mean ratio); (b)
uniformly divide the RadViz circle into sectors for each class; (c) place the
highest ranked anchor in the middle of the sector for that class; (d) place
the next highest immediately above it; (e) place the next (third highest)
immediately below that middle point; and (f) continue the process in this