ClustanGraphics4 Error Reports 

Home
About Clustan
Cluster Analysis
Applications
ClustanGraphics
User Support
Clustan/PC
Orders
What's New
White Papers
Contact Us

ClustanGraphics4 (2000 versions)

Collapsing Scatterplots

Cluster scatterplots can show the distribution of clusters on any combination of two input or scatterplot variables.  In the absence of scatterplot variables, which are read using File/New/Scatterplots, the values of input variables were not being correctly initialized.  This was manifest by a collapsing scatterplot, in which the points typically all appear in the top corner, row or left column of the scatterplot.

This was corrected in ClustanGraphics 4.19 (July 2000).

Work-around: Display your scatterplots only using scatterplot variables.  If necessary, re-read any input variables as scatterplot variables and then use them in plotting.


Reading Large ClustanGraphics Files

ClustanGraphics2 was limited to constructing a hierarchical cluster analysis on a maximum of 16,000 cases.  This restriction was removed in ClustanGraphics3, when it became possible to compute and save trees for any number of cases.  However, a check on the size of an input tree was not removed when re-reading large ClustanGraphics saved files, resulting in an error window reporting that such files cannot be opened.  This restriction was removed in release 4.19 (July 2000)

Work-around: If you want to save trees containing more than 16,000 cases, upgrade to the current release.  Otherwise, re-read your data each time you need to construct a new tree.


Tied Proximities in Cluster Data

When using direct data clustering (Cluster/Data) with a large dataset (e.g. 10,000+ cases) that exhibits tied proximities, it is possible to obtain an incorrectly ordered tree.  Error exceptions can arise if the tree is used as an initial partition or in truncation.  From release 4.17 (July 2000), this error has been corrected.

Work-around: None.  If you encounter error exceptions when selecting a tree partition you need to upgrade to version 4.17 or later.


Empty Clusters

When using outlier analysis, it is possible to obtain empty clusters; this happens, for example, when a very small distance threshold is applied, such that every case in a cluster has a distance from the mean which exceeds the threshold.

From release 4.14 (June 2000), ClustanGraphics eliminates any empty clusters when saving a cluster model, and issues a warning message in the details folder of outlier analysis.

Work-around : Copy the cluster model into an Excel spreadsheet and eliminate any column or row corresponding to an empty cluster.


Cluster Weights

Prior to release 4.14 (June 2000) ClustanGraphics provided no way of saving cluster weights, or sizes.  This was corrected by adding a weights column or row to the cluster model table, which is saved from View|Current Model.

If you do not require weights to be saved with the cluster model, you should paste the output from View|Current Model into an Excel spreadsheet and then delete the column or row which is not required.

Work-around : none.  If you want to output cluster weights you need to upgrade to version 4.14 or later.


Reading Missing Values from Excel Spreadsheets

Missing values were not read correctly from Excel spreadsheets prior to release 4.10 (May 2000).  Missing values can now be coded in a spreadsheet by a special code, such as 999, or as an alpha-numeric code, such as M or "missing".  For further details about missing value treatments in ClustanGraphics click here.

Work-around: Save the data in your Excel spreadsheet to a text file and read it into ClustanGraphics as a space-delimited file; or copy the data to the Clipboard and paste it into ClustanGraphics.


Saving Cluster Membership Table

When a cluster membership table is constructed using Cluster|Members for a very large dataset, the dialogue used to display the table can run out of system resources.  A Save-to-File facility was introduced in release 4.07 which will save any size of cluster membership table to a text file.  This can then be copied into a spreadsheet on analyzed externally using another program.


Case Labels and Weights

ClustanGraphics Release 4.04 (February 2000) coincided with a change of development platform.  Unfortunately, the Edit|Labels and Edit|Weights features ceased to work properly, and Tree|Truncate was also affected.  This was due to a change in the specification of the list box used in the Edit pages, which was not initially noticed.

This error did not result in incorrect clustering; but limited the ability to change labels and case or variable weights.  It was corrected in version 4.05 (March 2000).

Work-around : none.  Users affected by this error are being sent a corrected version.


Reading Large Proximity Matrices

When pasting a proximity matrix, the number of cases that can be read was restricted due to the input field being limited in size.  This restriction was removed in release 4.01 (January 2000).

Work-around: paste large proximity matrices into a word processor as unformatted text and then save them as text files which can be read correctly by ClustanGraphics.


Truncating a Tree

Cluster weights and cluster labels were not correctly updated when truncating a tree. This also affected any further cluster analysis of the truncated tree, where the nodes should have differential cluster weights.  This was corrected in release 4.01 (January 2000).  While the truncated tree was correct, there is no work-around for correcting cluster weights and labels, other than to read them again via Edit|Weights and Edit|Labels.