ClustanGraphics5 Error Reports 

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

ClustanGraphics5 (2001/2002 versions)

View Data

It was not possible to view the data for a large number of variables without line wrapping, or to store it on a file, so that copying to a spreadsheet became difficult.

This was corrected in ClustanGraphics 5.28 (November 2002), which includes an option to store the data on a text file.

Work-around: Copy the data to a word processor such as Word, and replace the paragraph marks with tab characters.  This has the effect of unwrapping so that there is one case per line.


Cluster Profiles

An exception occurred when an attempt was made to construct a table of cluster profiles where the number of clusters was equal to the number of cases.  Technically, these are not normally clusters; however, they can be where the user has truncated a larger tree.  Similarly, it was not possible to obtain cluster profiles for the whole dataset as a single cluster.

This was corrected in ClustanGraphics 5.28 (November 2002).

Work-around : To avoid the error report, highlight a number of clusters less than n on the tree before selecting Cluster Profiles.


k-Means Analysis ESS Values

In the statistics table for k-Means Analysis, the value for the within-cluster ESS was not correctly reported.  The result is that a cumulative value is reported where more than one run is attempted within the procedure.  The classifications and all other results are unaffected by this fault.

This was corrected in ClustanGraphics 5.21 (November 2001).

Work-around: If you wish to run k-Means analysis more than once, exit and re-enter the procedure for each run. This is only necessary if you need the within-cluster ESS statistics.


Mixed Data Settings

Some users have found the mixed data settings confusing if the data are all continuous or ordinal variables.  If you enter Edit/Data Types with continuous or ordinal variables, on exit the data file was changed to mixed data which then inhibited its use with certain procedures.

This was corrected in ClustanGraphics 5.20 (October 2001).  Details here.

Work-around: If your data are all continuous or ordinal, then do not enter Edit/Data Types.  Use the data transformation and missing value options provided in File/New/Data only.


FocalPoint Seeds

The option to read cluster centres from a file, or paste them from the Clipboard, and then use them as a starting seeds in FocalPoint was not working properly.

This was corrected in ClustanGraphics 5.20 (October 2001).  Details here.

Work-around : We regret that there is no work-around - if you want to read seed points to start FocalPoint then you need the upgrade.  You can read initial cluster centres using Cluster/k-means.


Gower's General Similarity Coefficient

Gower's Similarity Coefficient was not computed correctly for continuous variables that had not been normalized by range.  Details here.

This was corrected in ClustanGraphics 5.16 (August 2001).


Reading Binary Data Files

ClustanGraphics has a Save/Binary File option which can be used to save a large data matrix as a binary file which can then be re-read quickly using Read/New/Binary Data.  The file has a record width of 32 values, and the reading procedure generates an error exception where the number of variables is an exact multiple of 32.  When the fault occurs, a ClustanGraphics error message "Read beyond end of file." is displayed.  We apologize for this obscure error, which was corrected in ClustanGraphics 5.09 (May 2001).

Work-around: We regret that there is no work-around, other than to avoid reading a data matrix for which the number of variables is an exact multiple of 32.


Reading Large Proximity Matrices

Prior to release 5.07 it was not possible to read in a proximity matrix for n>50 due to faulty dimensioning.  This was corrected in ClustanGraphics 5.07 (May 2001).

Work-around: We regret that there is no work-around.


Saving Large Proximity Matrices

Prior to release 5.05 it was not possible to save a large proximity matrix.  This arose because the proximity matrix is constructed in View/Proximities using a list box which has an upper memory limit which is exceeded with more than 100 cases (the proximity matrix is of size nxn, so for 100 cases it occupies 10,000 cells).

This was corrected in ClustanGraphics 5.05 (May 2001) by providing output to a text file as an alternative in View/Proximities.  The options include output in square or lower diagonal formats, and the resulting files can be opened in a spreadsheet or word processor.

Work-around : If you need to output a large proximity matrix, we regret there is no work-around for more than 100 cases.  Consider obtaining an upgrade.


Labels and Weights for Large Applications

Prior to release 5.05, the labels and weights of both cases and variables were stored and edited in list boxes in the View/Labels and View/Weights dialogues.  The list boxes had a fixed memory limit which meant that labels for very large files were incomplete, and the corresponding cluster membership tables were not labelled correctly.  Also, weights for large applications could not be fully displayed or edited.  This has now been corrected such that case labels of any length for any size of data matrix can be read, stored and displayed; and similarly for variables.

This problem does not arise with fewer than 4000 labelled cases or labelled variables.  It was only encountered by users analyzing very large datasets in data mining and genomics.  The restrictions were removed completely in ClustanGraphics 5.05 (May 2001).

Work-around: Do not attempt to store case labels for large datasets in versions prior to 5.05 and hence the cases will be numbered sequentially from 1 to n.  Have a look-up table, e.g. in Excel, where the case numbers are cross-referenced to the case labels.


Cluster Membership Tables for Large Applications

For the reasons given in the prvious fix report, case labels were not completely stored for large applications in the View/Labels dialogue.  This meant that cluster membership tables were incompletely filled when case labels were selected.  This has now been corrected from release 5.05, such that case labels of any length for any size of data matrix can be displayed in a cluster membership table.  However, because the cluster membership table also has a fixed memory limit, cluster membership results should be saved to a text file by selecting the Output button, and this applies when the number of cases exceeds 4000.  The program automatically switches to text file output for large applications.

Work-around: Do not store case labels for large datasets in versions prior to 5.05 and the cases will be numbered sequentially from 1.  Have a look-up table, e.g. in Excel, and paste the cluster membership table into it as additional columns.  The cluster codes will then cross-refer to the case numbers and hence the case labels.


Reading Large Excel Files

Several users noticed that Excel files were slow to read into ClustanGraphics.  This was corrected in ClustanGraphics 5.04 (April 2001).  In addition, it is now possible to read case labels and weights, and variable labels, directly from an Excel spreadsheet.  Details here, including an apology!

Work-around : To read a large Excel file with a previous version of ClustanGraphics, select "Save As" in Excel and specify "Text (Tab delimited)" as the output type.  The saved file can then be read into ClustanGraphics using File/New/Data and "Space delimited" format.


Gower's General Similarity Coefficient

In beta release 5.01, the value of Gower's Similarity Coefficient was not computed correctly for binary variables, in respect of ignoring negative matches.  The same fault applied to the binary simple matching coefficient.  Details here.

This was corrected in ClustanGraphics 5.02 (January 2001).


.