ClustanGraphics3 Error Reports 

About Clustan
Cluster Analysis
User Support
What's New
White Papers
Contact Us

ClustanGraphics3 (1999 versions)

Reading Excel Spreadsheets

Prior to release 4.0 (January 2000), ClustanGraphics failed to initialise case weights correctly when an error occurred while reading from an Excel spreadsheet.

Work-around:  Always ensure that the spreadsheet has been read correctly.  Check that the number of cases and variables are correct, and click Details for any warning messages that may have been issued.  Provided that the spreadsheet was read correctly, the rest of the analysis will proceed normally.

ClustanGraphics Save File Precision

Prior to release 3.17 (December 1999), ClustanGraphics saved data and proximities with 3 decimal digit accuracy.  This has now been extended to 6-digit precision, because certain data sets (those with small decimal values below 1) were losing precision due to rounding.  No changes are required to existing ClustanGraphics saved files, which should re-open as normal.  However, to take full advantage of the extra precision, datasets should be re-read from new, and proximities re-computed.

Standardization to Unit Range

Classify When using File|New Data, the option to standardize variables to unit range was not working correctly.  This was corrected in release 3.17 (December 1999).  The other standardization options (z-scores, rescale and no transformation) work correctly. 

Work-around: The only workaround for previous versions is to standardize variables to unit ranges outside of ClustanGraphics, for example using Excel. 

Classify Cluster Level n

Classify did not allow new cases to be compared with all the cases in a dataset or tree, i.e. the partition corresponding to n clusters at the very bottom of the tree.  This was corrected in release 3.15 (August 1999).  A cluster model can now be read in as a new data matrix and then used as a reference for classifying new cases.

k-Means Cluster Level n

Similar to the above error, k-Means Analysis did not allow the partition corresponding to all the cases in a dataset or tree to be selected.  This was corrected in release 3.15 (August 1999).  A k-Means cluster model can now be read in as a new data matrix and then selected with one case per cluster.  This allows k-Means Analysis and statistics to be computed for an input cluster model.

Pasting data into ClustanGraphics

When pasting a data matrix or scatterplot variables, the number of variables that can be read was restricted due to the input field being limited in size.  This restriction was removed in release 3.10 (from May 1999).

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

Differential Variable Weighting

The computation of proximities using differential variable weights was provided from release 3.09 onwards (April 1999).  Prior to that, variables were assumed to be weighted equally, after standardization or range transformations had been applied.

Work-around: multiply the variables by the appropriate weights outside ClustanGraphics - for example, using a spreadsheet - and then read them into ClustanGraphics with no data transformation.

Reading Character-Sequence Data

Character sequence data was incorrectly read in releases 3.07-3.08 (April 1999) as a consequence of initializing differential variable weights (see above).  We regret that there is no work-around in these releases, other than to re-code character sequence data as numerical values.  This was corrected from release 3.09 (April 1999) onwards.

Truncate Tree

When using Truncate Tree with a data matrix or scatterplot variables, the cluster means were incorrectly computed up to release 3.01 (February 1999). 

ClustanGraphics3 release 3.00 (January 1999) and the beta versions are affected by the above reported errors.  If you need to obtain an upgrade, please contact Clustan by email.