IBM Support

Release Notes - IBM SPSS Modeler 18.2

Release Notes


Abstract

IBM® SPSS® Modeler is a data mining toolset that helps you build predictive models quickly and intuitively. System requirements, installation, and known problems are addressed.

Content

For a list of new features in this SPSS Modeler release, see New Features in IBM SPSS Modeler 18.2 in the Knowledge Center. To view a list of issues corrected in this release, see the fix list .
System requirements and installation

For information regarding requirements and installation, see: http://www.ibm.com/software/analytics/spss/products/modeler/requirements.html .
Note that Windows 32-bit is no longer a supported platform.
AIX is a supported environment for 18.2

Windows - run as administrator
If you are installing on Windows Server 2012, Windows Vista, Windows 7, Windows 8, or Windows 10 operating system, you must use "Run as Administrator" by first right-clicking on the executable.

IBM Cognos TM1


IBM Cognos TM1 10.2 Fix Pack 1, or later, is required before Modeler can import and export Cognos TM1 data.

Ubuntu 14 and Ubuntu 16 on Linux


On Linux, there can be issues with installing Modeler Server on Ubuntu. For an explanation of the workaround, see: 
https://flexeracommunity.force.com/customer/articles/en_US/PROBLEM/Malformed-uxxxx-Encoding-Error-Generated-when-Launching-Installer

Visual Studio 2015


Visual Studio 2015 is supported for Modeler Solution Publisher.


Single sign-on for data sources

You can connect to databases from IBM SPSS Modeler using single sign-on. If you want to create a database connection using single sign-on, you must first use your ODBC management software to properly configure a data source and single sign-on token. Then when connecting to a database in IBM SPSS Modeler, IBM SPSS Modeler will use that same single sign-on token and the user will not be prompted to log on to the data source.


However, if the data source was not configured properly for single sign-on, IBM SPSS Modeler will prompt the user to log on to the data source. The user will still be able to access the data source after providing valid credentials.

For complete details about configuring ODBC data sources on your system with single sign-on enabled, see your database vendor documentation. Following is an example of the general steps that may be involved:

  1. Configure your database so it can support Kerberos single sign-on.
  2. On the IBM SPSS Modeler Server machine, create an ODBC data source and test it. The DSN connection should not require a user ID and password.
  3. Connect to IBM SPSS Modeler Server using single sign-on and begin using the ODBC data source created and validate in step 2.


Single sign-on Kerberos authentication

If you plan to use single sign-on for SPSS Modeler on either Windows 7 SP1 or Windows Server 2008 R2 SP1, you may be affected by a known issue. As an example, you may see an error message that resembles the following: STATUS_TRUSTED_RELATIONSHIP_FAILURE


If you are affected, see the hotfix available at http://support.microsoft.com/kb/2526946 . for an explanation and solution.

Known issues and limitations
At time of publication, the following issues were known:

  • Because of an issue reported with IBM JDK, Data Views can't be loaded in Modeler Client when connected to a zLinux Server.
  • The Netezza bulkloader nzload does not accept SJIS encoding. So if you are handling data with SJIS encoding and export to to Netezza via a bulkloader script, you must manually set the encoding as UTF8 in Modeler's options.cfg file (add "encoding, UTF-8" to the file).
  • The Evaluation node doesn't support the use of commas in field names. If you have field names containing commas, you must either remove the commas or surround the field name in quotes.
  • Real time scoring for the K-Means-AS node, Isotonic-AS node, and XGBoost-AS node with the embedded Analytic Server adapter is only supported on JBoss EAP and WebSphere deployment at this time. Weblogic is not supported, and if it's deployed on Windows, you must set the environment variable HADOOP_HOME=%CDS_INSTALL_PATH%\components\modeler\as\lib\spark\hadoop -- otherwise the embedded Spark engine will not start correctly and scoring will fail. We recommend you set this environment variable for Windows regardless of which application server you're using, in case there are any unexpected problems during real-time scoring for these three model types.
  • The E-plot (Beta) node can't run in an IBM SPSS Collaboration and Deployment Services job on a headless Linux environment because the third-party component JXBrowser does not support headless Linux.
  • In the E-plot (Beta) node, you can't choose the same data for the X axis and Y axis.
  • The Spark nodes (XGBoost-AS, K-Means, and Isotonic) don't support specifying an integer/double column as Flag/Nominal for building a model. To do this, you must convert the column value to 0/1 or 0,1,2,3,4...
  • The Spark nodes (XGBoost-AS, K-Means, and Isotonic) are only supported on Windows 64, Mac 64, and Linux 64.
  • On machines with a Taiwanese locale, when upgrading from Modeler 17.x, old streams (created in 17.x) may need to be unlocked due to a JRE upgrade. For example, you may need to unlock Super Nodes in Modeler 17.x first and then lock it in 18.2 after upgrading.

  • Weblogic and Collaboration and Deployment Services Deployment Portal

When using Weblogic as your Application Server, the module pasw.scoring sometimes does not load while running real-time or batch scores with a Modeler stream in Deployment Portal. To successfully load this module, you must manually update the java library path. Contact IBM Support for details.

  • Java heap space error in Modeler Server

For some tasks, when processing a large number of records, it is possible that a Java heap space error can be displayed.

If this happens, you can configure the JVM setting to increase the memory. as follows:

  1. Close the stream that generated the error.
  2. In the <Modeler Server installation path>\config folder, open the jvm.cfg
  3. Configure the setting, the default is 256; for example: options, "-Xmx1024m"
  4. Save the configuration and reopen the stream.
  • datetime_now function returns null value against Teradata


When using the DataDirect Teradata driver, the SQL_COLUMN_TYPE may return a null value. The alternative is to use the native Teradata driver.

Database import and export

  • SQL Server support with the Data Access Pack driver.

The ODBC configuration for SQL Server must have the "Enable Quoted Identifiers" ODBC connection attribute set to “Yes” (the default for this driver is "No"). On UNIX this attribute is configured in the system information file (odbc.ini) using the "QuotedId" option.

 

  • In Database Caching with IBM DB2.

When attempting to cache a node in a stream which reads data from a DB2 database, you may see the error message “A default table space could not be found with a pagesize of at least 4096 that authorization ID TEST is authorized to use”. To configure DB2 to enable in-database caching to work properly in SPSS Modeler, the database administrator should create a "user temporary" tablespace and grant access to this tablespace to the relevant DB2 accounts. We recommend using a pagesize of 32768 in the new tablespace, as this will increase the limit on the number of fields that can be successfully cached.

 

  • Database errors with IBM DB2 for z/OS.

When running streams against DB2 for z/OS, you may experience database errors if the timeout for idle database connections is enabled and set too low. In DB2 for z/OS version 8, the default has changed from no timeout to 2 minutes. The solution is to increase the value of the DB2 system parameter IDLE THREAD TIMEOUT (IDTHTOIN), or reset the value to 0.

 

  • Scoring some models with confidences enabled using generated SQL returns a database error message on DB2 z/OS.

Scoring a subset of algorithms, with confidences enabled, using generated SQL can return an error on execution. The issue is specific to DB2 for z/OS; to fix this, use the Modeler Server Scoring Adapter for DB2 on z/OS.

 

  • Database bulk loaders.

In order to use the bulk loading feature of the Database export node, you need to install Python on the same machine as SPSS Modeler (or if using SPSS Modeler Server, on the same machine as the server). The "python_exe_path" parameter must be set in the options.cfg file. You can install Python from the SPSS Modeler Client, SPSS Modeler Server or SPSS Modeler Solution Publisher product DVDs.

 

  • Some aggregation results can differ between SQL pushback and native modes with Oracle.

When running a stream containing an Aggregate node, the values returned for 1st and 3rd Quartiles when pushing back SQL to an Oracle database may differ from those returned in native mode.

 

Modeling nodes

  • Record ID field.

Passing a non-numeric Record ID field into a modeling algorithm may cause a stream to execute slowly. The Record ID field is not a requirement for modeling, so we recommend filtering out the field.

  • Logistic Regression.

Binomial Logistic Regression does not allow strings longer than 8 characters. You can avoid this problem by encoding strings, for example by using the Anonymize node, before passing them to the algorithm.

If temporary disk space is low, Binomial Logistic Regression can fail to build, and reports an error. When building from a large data set (10GB or more), the same amount of free disk space is needed. You can use the environment variable SPSSTMPDIR to set the location of the temporary directory.

  • Cox regression.

On scoring a Cox regression model, an error is reported if empty strings in categorical variables are used as input to model building. Avoid using empty strings as input.

Integration with IBM SPSS Collaboration and Deployment Services

  • Storing SPSS Modeler streams in SPSS Collaboration and Deployment Services.

No unlock checkbox is available when storing streams in the Collaboration and Deployment Services repository. Unlock is the default when storing streams. To lock or unlock an object, choose Tools->Repository->Explore, navigate to the object, and right-click on its name to display the context menu.

 

  • Incomplete output for Gains graph from Evaluation node.

In Deployment Manager, on running a job containing a stream with an Evaluation node set to produce a Gains graph, the graph output may be incomplete if the system is running under Oracle Weblogic 11g using the Oracle JRockit JRE. To avoid this problem, use the IBM JRE.

 

Integration with IBM SPSS Statistics

  • Generating non-English output.

When generating output in languages other than English, it is advisable to specify the language in the Syntax.

 

  • The 'Launch application' option of a Statistics Export node does not open the data set when working in distributed mode.

When running Modeler and Statistics together in Server mode, writing the data out and launching a Statistics session does not automatically open a Statistics client showing the data set read into the active data set. The workaround is to manually open the data file in Statistics client once it is launched.

 

Multi-byte data

  • Inserting multibyte data into Teradata from SPSS Modeler Server.

To insert multibyte data into a Teradata database from SPSS Modeler Server, use the following configuration:

1. Run the server in Unicode.

2. Set the Teradata user default character set to UNICODE using tdadmin.

3. Configure CharacterSet=UTF8 (UNIX DSN), or set the Session Character Set to UTF-8 (Windows DSN).

4. Ensure that there are only ASCII characters in the column names.

  • String Collation.

In this release, string sorting and comparison use the ICU 4.8.1 collation service for the system locale. Japanese collation does not distinguish half-width from full-width Katakana.

Accessibility

  • Graphs.

The screen reader is not able to read graphs, so these are not accessible to visually-impaired users.

 

SPSS Modeler Text Analytics

  • Cancelling an extraction can take a long time when working with very large data sets.

Multiple Interactive Workbench sessions can cause sluggish behavior. Text Analytics and Modeler share a common Java run-time engine when an interactive workbench session is launched. Depending on the number of Interactive Workbench sessions you invoke during a Modeler session - even if opening and closing the same session - system memory may cause the application to become sluggish. This effect may be especially pronounced if you are working with large data or have a machine with less than the recommended RAM setting of 4GB. If you notice your machine is slow to respond, it is recommended that you save all your work, shut down Modeler and re-launch the application. Running Text Analytics on a machine with less than the recommended memory - particularly when working with large data sets or for prolonged periods of time - may cause Java to run out of memory and shut down. It is strongly suggested you upgrade to the recommended memory setting or larger (or use Text Analytics Server) if you work with large data.

 

  • Scoring on rules is different in Text Analytics depending on whether you load a text analysis package (TAP) from Text Analytics directly or whether you load a TAP from Text Analytics for Surveys

Use TAPs made within Text Analytics as the ones made in Text Analytics for Surveys may be created using a different version of the linguistic resources.

 

  • Linux on x64 - openMP support requires the customer to install a separate package.


Linux Red Hat x64 support.
For Red Hat Linux, openMP support requires the package "libgomp-4.4.4-13.el5", which is available from the RedHat website: https://rhn.redhat.com/network/software/search.pxt

Linux SuSe x64 support.


For SuSe11, openMP support requires the package "libgomp43 4.3.3_20081022", which is available from the SuSe website: http://www.suse.com/LinuxPackages/packageRouter.jsp?product=server&version=11&service_pack=&architecture=i386&package_name=index_group .

The package is for the GNU compiler collection OpenMP runtime library, and is available from the section titled: "Development/Libraries/Parallel".

  • Client memory exhausted after many repeated Interactive Workbench extractions.

Modeler Client can run out of memory after multiple Text Analytics Interactive Workbench sessions have been run without restarting the application. Monitor the memory usage in the status line and, if running low, close and re-open Modeler Client.

Related information

Technical Support is available to maintenance customers. Customers may contact Technical Support for assistance in using IBM Corp. products or for installation help for one of the supported hardware environments. To reach Technical Support, see the IBM Corp. web site at http://www.ibm.com/support . We recommend that you check the support site for updates.

© Copyright IBM Corporation 1994, 2018.

Original Publication Date

04 December 2018

[{"Business Unit":{"code":"BU001","label":"Analytics Private Cloud"},"Product":{"code":"SS3RA7","label":"SPSS Modeler"},"Component":"","Platform":[{"code":"PF016","label":"Linux"},{"code":"PF017","label":"Mac OS"},{"code":"PF033","label":"Windows"}],"Version":"18.2","Edition":""}]

Document Information

Modified date:
14 January 2019

UID

ibm10720105