Some of our web visitors ask what's the difference between Clustan/PC and ClustanGraphics? This page attempts an answer. (We apologize for any confusion.) First, we need to explain that Clustan has been around much longer than ClustanGraphics. Clustan is a Unix-based package that we ported to Windows as Clustan/PC in 1991. Clustan 3.4 now runs under Windows 95, 98, 2000, ME, XP and NT. Second, because it runs under Unix, Clustan/PC does not yet have a full Windows GUI. To run Clustan/PC, you either write a Clustan program or enter commands via an interactive command language interpreter. See Cluster Tutorial for an example. On-line Help is provided when entering commands, but not pull-down menus or screen graphics. Clustan/PC graphics are produced as encapsulated postscript files of publication quality. But you need a printer with a postscript option to print these graphics. Not everyone uses postscript. So we introduced ClustanGraphics in 1997 to draw on-screen bitmap graphics that you can view, print and copy into reports and presentations. We have since added lots of clustering functionality around these graphics to satisfy user demand. We describe ClustanGraphics as our "entry level" product, because it's very easy to use and fully exploits Windows. However, this means it can only run under Windows 95, 98, 2000, ME, XP or NT. If you're already using Windows you should find the look and feel of ClustanGraphics instantly familiar and easy to use straight away. Just point, click and cluster! The following comparative summary aims to clarify the differences between ClustanGraphics and Clustan.
Read Data reads and transforms a data matrix. The corresponding Clustan procedure is Read Data. Read Proximities reads a proximity matrix. The corresponding Clustan procedure is Read Similarity Matrix. Compute Proximities computes a proximity matrix from a data matrix. The corresponding Clustan procedure is Calculate Similarity Matrix. Whereas Clustan offers 40 similarity coefficients, ClustanGraphics only offers 4. Cluster Proximities clusters a proximity matrix by 11 methods of hierarchical cluster analysis. The corresponding Clustan procedure is Hierarchy. ClustanGraphics can detect and test for the sensitivity of tied proximities, and offers 2 extra methods. Cluster Data clusters directly on your data, by 2 methods of hierarchical cluster analysis, and is thus suitable for large datasets. The corresponding Clustan procedure is Cluster. Whereas Clustan allows for missing values, this is not yet available in ClustanGraphics. Classify Cases uses a classification produced by a clustering method to identify new cases. It is therefore suitable for classifying datasets of unlimited size. The corrresponding procedure in Clustan is Classify. Both Clustan and ClustanGraphics can handle missing values in Classify. k-Means optimizes a classification by iterative k-means. The corresponding procedure in Clustan is Relocate. Cluster Scatterplots displays a classification for any two internal or external variables. The corresponding procedure in Clustan is Scatter. Best Cut, seeks optimum partition of a tree using significance tests. The corresponding procedure in Clustan is Rules.
Re-Order Proximities orders a tree in optimum case sequence. Cluster Exemplars finds the typical members of each cluster. Shade Proximities draws a shaded representation of the proximity matrix. Weights allows cases or variables to be assigned differential weights when computing proximities and in clustering. Cluster Profiles lists and charts key cluster characteristics. Outlier Analysis identifies and deletes outlier cases in a classification. Truncate Tree truncates a hierarchical classification, computing cluster means and proximities for the resulting clustering. Tree Styles offers a choice of styles when drawing trees. Tools Palette allows trees to be re-sized or formatted. Hybrid clustering, reduces a large tree to kernel clusters for single linkage. Cluster Membership tables are compiled by case label or cluster code. Point-and-click Popup Menu simplifies cluster evaluation. Genetic Data Analysis features include character sequence data, Jukes-Kantor distances and Primates case study. Viewing and Exporting data and clustering results added. Re-Open any of last 4 ClustanGraphics projects with a single click. ClustanGraphics Primer is supplied (60pp, 40 graphics).
Centroid performs hierarchical cluster analysis on a proximity matrix by the centroid method. Divide performs hierarchical divisive clustering on binary variables, similar to decision tree analysis. Normix finds maximum likelihood estimates of a multivariate normal mixture with equal or unequal covariance matrices. Invariant minimizes Wilks' Lambda or Hotelling's Trace. Mode finds the modes in a sample density. Kdend seeks Jardine-Sibson Bk overlapping clusters. Dndrite divides the minimum spanning tree to minimize the euclidean sum of squares. Euclid is a fuzzy clustering method which minimizes the euclidean sum of squares. Print results outputs cluster diagnostic results. Plink plots hierarchical clustering trees. Compare compares hierarchical classifications. The Clustan User Manual (250 pp) is supplied in a ring binder and slip case.
Clustan/PC and ClustanGraphics are supplied on CD-ROM, with a standard Windows 95/98/NT Startup installation. To order ClustanGraphics on-line click To order Clustan/PC on-line, click If you got this far .. Congratulations! You deserve a wee dram! The question is - which one? |