The Cluster Count property may not match the actual number of clusters displayed in the Mining Model Editor (273495)


The information in this article applies to:

  • Microsoft SQL Server 2000 Analysis Services
This article was previously published under Q273495

SYMPTOMS

When you create a mining model by using the Microsoft Clustering algorithm and you modify the model in the Mining Model Editor, you will find that the Properties pane allows you to specify the Cluster Count. However, the Cluster Count property value may not represent the actual cluster count found by the algorithm. If you specify a value and you reprocess the model, you may get a different number of clusters, or the property value and the actual number of clusters might not match.

CAUSE



This behavior is by design.

The Cluster Count property indicates how many clusters you want to find when you train the model. If the model cannot find that many clusters in the data, it finds fewer clusters. Consequently, the property value and actual number of clusters do not match.

MORE INFORMATION

Steps to Reproduce the Problem

  1. On the SQL Server server, use the following code to create a table:
    Create table Travel ( age int,  Country  char ( 10 ) )
  2. Import the sample data from the SQL Server Books Online topic "Microsoft Clustering" about using the Data Transformation Services (DTS) wizard from a text file into the table you created in step 1.
  3. Run the SQL Server 2000 Analysis Services Data Mining wizard to create a mining model based on the table. Use the following information as you navigate through the wizard:

    Microsoft Clustering:
    Case column: Age
    Input column: Country

  4. In the editor, click Finish, and then save and process the model. Saving and processing the mining model creates two clusters.

    However, the Cluster Count property on the left hand side displays a 10.

    Edit the mining model, click the Properties button on the left hand side (the pane might be hidden for a new model), and then specify a new value for the Cluster Count as follows:

    Cluster Count = 3 (The number of clusters you want the data mining algorithm to identify.)

    Now, you must save and reprocess the model.

    The mining mode uses three (3) for the Cluster Count as follows: 
    MiningModel 'Country_age' Execute : CREATE MINING MODEL [Country_age'S] ([Age] LONG   KEY  , [Country] TEXT   DISCRETE  PREDICT) USING Microsoft_Clustering (CLUSTER_COUNT=3)
After the mining model is processed, the Cluster Count value is 3; however, the display still shows two clusters.
Modification Type:MinorLast Reviewed:3/18/2006
Keywords:kbprb KB273495 kbAudDeveloper
©2004 Microsoft Corporation. All rights reserved.