Fw: Weird behavior hierarchical cluster

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

Fw: Weird behavior hierarchical cluster

SPSS Support-2

Dear Ruben,

What's going on here is that in Release 16 and a number of earlier releases the icicle plots in CLUSTER were actually pivot tables, with the icicles formed by columns (or rows) of X values. When clustering cases, the body of a vertical icicle plot had N-1 rows and 2N-1 columns (each original case or cluster was in a column, and these N columns has N-1 columns in between them), so you had a pivot table with (N-1)(2N-1) cells in its body. Huge pivot tables tended to cause problems with crashes in the output viewer, and the MXCELLS subcommand in the SET command was designed to try to stop the building of tables that were too large for the available memory. You may be able to get the job to run without warnings in Releases 15 and 16 by allocating more memory via SET MXCELLS, but it may take a long time to see the resulting table, and it may be difficult to move around in it.

In Release 17, these chart-like pivot tables were replaced with true charts. However, if you run out of memory for the chart, you still get the warning about pivot tables and MXCELLS in current releases, so we've filed a bug report and will get that fixed so that an appropriate warning is issued.

David Nichols
Statistical Support
SPSS, an IBM Company

From: Ruben van den Berg <[hidden email]>
To: [hidden email]
Date: 06/07/2010 02:22 AM
Subject: Weird behavior hierarchical cluster
Sent by: "SPSSX(r) Discussion" <[hidden email]>





Dear all,

A colleague approached me with this warning
Warnings
The total number of pivot table cells across split files exceeds 3241472. This limit can be altered by using the SET MXCELLS command. If the working file was being created or modified during the current procedure, any changes were probably lost.
This command is not executed.

 
which was generated by this syntax on V16

CLUSTER FAC1_5 FAC2_5 FAC3_5 FAC4_5 FAC5_5
/METHOD WARD
/MEASURE=SEUCLID
/PRINT SCHEDULE
/PLOT VICICLE.

I reran the syntax on V15 and it didn't run at all. I then reran it on V17 and it ran fine, in very little without any error/warning. The dataset contains some 1.500 cases and no (system or user) missing values. Even though this warning was generated, the results looked just normal. But it did frighten us of course.

I think SPSS no longer supports V15 and V16 but since some colleagues still use these versions, I wondered if anyone could shed any light on our findings.

TIA,

Ruben van den Berg

Consultant Models & Methods

TNS NIPO

Email: [hidden email]

Mobiel: +31 6 24641435

Telefoon: +31 20 522 5738

Internet:
www.tns-nipo.com





New Windows 7: Find the right PC for you. Learn more.