What's New in SAS Enterprise Miner 6.1

Overview

SAS Enterprise Miner 6.1 is a major new release of data mining tools for use with SAS 9.2. The scope of improvements includes many analytical and deployment functionality enhancements, as well as changes that were made to integrate the Enterprise Miner tool set with the SAS 9.2 system.

SAS 9.2 Platform

The SAS 9.2 system is an improved platform for managing and deploying analytical and business intelligence applications for both single-user applications and multi-user enterprises.  SAS Enterprise Miner 6.1 contains changes related to the SAS 9.2 system that improve SAS Enterprise Miner installation, security, and administration.

Software Versions and Migration

• SAS Enterprise Miner 6.1 requires the SAS 9.2 Platform release.
• SAS Enterprise Miner 5.3 will not operate with SAS 9.2.
• If you have existing SAS Enterprise Miner 5.3 project information stored in your SAS Metadata Server, the project information will be converted from SAS 9.1.3 format to SAS 9.2 format during the SAS 9.2 / SAS Enterprise Miner 6.1 installation.
• If you have existing SAS Enterprise Miner 5.3 project data folders that are stored on SAS Workspace Servers, the project data folders do not require conversion for use with SAS 9.2 and SAS Enterprise Miner 6.1. All SAS Enterprise Miner 5.3 project data folders, files, tables, views, and catalogs that are stored on SAS Workspace Servers are compatible for use with SAS 9.2 and SAS Enterprise Miner 6.1.
• SAS Enterprise Miner 6.1 users can open existing SAS Enterprise Miner 5.3 projects without any manual conversion process.
• SAS Enterprise Miner 6.1 projects cannot be converted for use with SAS Enterprise Miner 5.3.
• SAS Enterprise Miner 4.3 users who wish to upgrade project data for use with SAS Enterprise Miner 6.1 can use the SAS Enterprise Miner project conversion macro. The project conversion macro upgrades SAS Enterprise Miner 4.3 project structures to SAS Enterprise Miner 5.3 project structures. SAS Enterprise Miner 6.1 opens SAS Enterprise Miner 5.3 project structures by the SAS Enterprise Miner Project conversion macro.

Projects

• SAS Enterprise Miner 6.1 project information is now stored and managed in the SAS Metadata Folders. SAS Enterprise Miner 6.1 users create projects in a specific folder location.
• The default location for new SAS Enterprise Miner 6.1 projects is My Folder. The My Folder location is unique for every user and is a private location. When a SAS Enterprise Miner 6.1 user creates a project, the user can accept the default project location, or specify a different folder of their own preference. For example, a user or group of users might store mining projects in a common folder where the projects can be shared.
• SAS Enterprise Miner 6.1 users will open projects by using a standard Open File window that displays the SAS Metadata Folders tree structure by default.
• When the SAS Metadata Server is upgraded from SAS 9.1.3 to SAS 9.2, existing SAS Enterprise Miner 5.3 project information that was stored in the SAS Metadata Server is migrated to the Shared Data folder.
• SAS administrators can view SAS Enterprise 6.1 project information via the SAS Management Console.

Models

• SAS Enterprise Miner 6.1 models are stored and managed in the SAS Metadata Folders. SAS Enterprise Miner 6.1 users register models to a specific folder location.
• SAS Enterprise Miner 6.1 users may now open or import models by using a standard Open File window that displays the SAS Metadata Folders tree structure by default.
• When the SAS Metadata Server is upgraded from SAS 9.1.3 to SAS 9.2, existing SAS Enterprise Miner 5.3 models that were stored in the SAS Metadata Server are migrated to the Shared Data folder.
• SAS administrators can view SAS Enterprise 6.1 model information via the SAS Management Console.

SAS Management Console Plug-in

• The SAS Enterprise Miner 6.1 Plug-in for the SAS Management Console is revised for SAS 9.2, but maintains the same range of functionality as in SAS 9.1.3. For more information, see the SAS Enterprise 6.1 Reference Help chapters on Installation and Configuration for more information.
• SAS administrators can use the SAS Management Console to view SAS Enterprise Miner project information.

Java Versions

• SAS administrators retain the ability to deliver SAS Enterprise Miner 6.1 to users via Java Web Start. Java Web Start users should have Java 1.5.12 or a compatible version.
• Installed versions of SAS 9.2 include Java 1.5.12. No further version of Java is required.

Usability

SAS Enterprise Miner 6.1 provides the following improvements in usability:

Summary Statistics in Variable List Tables

• The variable list tables that SAS Enterprise Miner users are familiar with have been improved in SAS Enterprise Miner 6.1. The variable view tables that surface in locations throughout the software now provide users with summary statistics for the table variables.
    
• The summary statistics are computed by the Advanced Advisor function in the Data Source Wizard, by the Input Data node, and by the Stat Explore node. Variable summary statistics are often used to make decisions about how to treat variables in data mining models.

Configurable Attributes in Variable List Tables

• SAS Enterprise Miner 6.1 is capable of displaying many different variable attribute columns in SAS Enterprise Miner variable list tables. Instead of displaying enormous tables that have many variable attribute columns, SAS Enterprise Miner 6.1 enables users to configure variable list table displays by selecting only the variable attributes that are important to their work.

Quick Text Search for SAS Code Editors and Text Viewers

• The SAS Code editors and text viewers have been enhanced with a quick text search toolbar that highlights and navigates between selected text search results. This is a great aid when searching for text in SAS Code, the SAS Log, and SAS Output listings.
    
• You can launch Quick Text Search from the SAS Enterprise Miner 6.1 main menu, or use <CTRL+T> to access the new tool bar.

Interactive Graphics Samples

• Previous versions of SAS Enterprise Miner provided interactive exploratory graphics that used a quick sample of the values in a variable list table. In SAS Enterprise Miner 6.1, the quick table sample that the software performs to generate interactive graphics has been improved.
    
• The new quick sample method scans only the attribute columns that the user selects, plus any additional Target, ID, Frequency, or Cost variables. This capability reduces the number of columns needed to perform interactive graphic sampling and increases the number of rows of data that are available for graphics.
    
• Variable table list sampling for interactive graphics can now be performed by using a sampling algorithm that is stratified by categorical target variables. This change improves the representation of the sample in the presence of skewed data.

Project Start and Stop Code

• The Project Start Code Editor window is modified to include the SAS log. Convenient access to the SAS log helps users who need to debug or modify their SAS Enterprise Miner project start code.
    
• The Project End Code Editor window has been eliminated.

SAS Library Explorer

• The SAS Library Explorer has been enhanced to view and edit (when appropriate) catalog entries of the types SOURCE, LOG, OUTPUT, and XML.

Model Import and Export

• SAS Enterprise Miner 6.1 users can register models directly to the SAS Metadata Folders tree structure. This feature provides users with more control over the security, access privileges, and organization of models.
    
• SAS Enterprise Miner 6.1 users can import a registered model into an existing data mining process flow diagram by using the Model Import node. The score code of the imported model is applied to the data in the process flow diagram, generating new model assessment statistics.
    
• The Model Repository window has been removed from SAS Enterprise Miner 6.1. The former flat list of registered models has been replaced by a hierarchical view of models in the SAS Metadata Folders. The Model Import node provides SAS Enterprise Miner 6.1 users with a list of available models.
    
• SAS Enterprise Miner 6.1 users can select File Open Model from the main menu to open a file utility window to browse the SAS Metadata Folders tree structure and choose a model for inspection.
    
• SAS Enterprise Miner 6.1 users can also use the Model Import tool to navigate the SAS Metadata Folders tree structure and choose a model for addition to the process flow diagram.

Interactive Decision Tree

• A switch-targets feature has been added to SAS Enterprise Miner 6.1 so that users can select a new dependent variable in a tree leaf and make new splits based on the new target. This is a powerful analytical feature for users who design decision trees for segmentation strategies.
    
• The Interactive Decision Tree is fully integrated into SAS Enterprise Miner 6.1 and requires no separate installation or documentation.
    
• SAS Enterprise Miner 6.1 gives users who start the software using Java Web Start users full use of the Interactive Decision Tree.
    
• The former Tree Desktop Application that was associated with prior releases of SAS Enterprise Miner is not distributed with SAS Enterprise Miner 6.1, but is available on the SAS downloads Web page for legacy purposes.
    
• The Tree Desktop Application will not work with a SAS 9.2 server.

New Nodes

SAS Enterprise Miner 6.1 includes two new data mining nodes. The new nodes are presented using the SEMMA functional groupings of Enterprise Miner.

• Sample —  SAS Enterprise Miner 6.1 adds the following new node to the Sample tab of the Enterprise Miner tool bar:
    
  • File Import node — The File Import node enables users to directly integrate external data files into SAS Enterprise Miner 6.1 process flow diagrams. The external file types supported include dBase .DBF files, Stata .DTA files; Microsoft Excel .XLS files; SAS .JMP files; Paradox .DB files; SPSS .SAV files; Lotus .WK1, .WK3, and .WK4 files; as well as tab-delimited .TXT files; comma-delimited .CSV files; and user-defined delimited .DLM files. Data files to be imported must be located either on the SAS Enterprise Miner client machine or in a network location that is accessible to the SAS Enterprise Miner server or the SAS server system.
      
• Model —  SAS Enterprise Miner 6.1 adds the following new node to the Model tab of the Enterprise Miner tool bar:
  • LARS — The LARS node uses Least Angle Regression and LASSO algorithms from the SAS/STAT procedure GLMSELECT to perform model fitting tasks and sophisticated variable selection for interval target models.

Enhanced Nodes

The following nodes in SAS Enterprise Miner 6.1 were enhanced in functionality or reorganized into new Enterprise Miner tool groups.  The enhanced and changed nodes are presented using the SEMMA functional groupings of  Enterprise Miner.

• Sample — The following changes have been made to the Sample tools in Enterprise Miner 6.1:
  • Append node — The Append node enables you to concatenate two data sets together. In SAS Enterprise Miner 6.1, the Append node is able to combine training, validation, and test data sets into a single training data set for the purpose of computing full data statistics.
      
• Explore — The following changes have been made to the Explore tools in Enterprise Miner 6.1:
  • Association node — The Association node is used to identify frequently occuring association and sequence patterns in transactional data. In SAS Enterprise Miner 6.1, the Association node improves by using a new SAS data mining procedure called MBSCORE. MBSCORE produces faster and more accurate output than previous versions of the Association node.
      
  • Stat Explore node — The Stat Explore node is used to generate summary statistics for data exploration. In SAS Enterprise Miner 6.1, the Stat Explore node computes summary statistics on validation and test data as well as the train data. Most Stat Explore results plots have been updated to show validation and test results. Stat Explore provides new plots that can compare variable distributions across multiple categorical targets and by-group segments.
      
  • Graph Explore node — The Graph Explore node is an advanced visualization tool for interactive data exploration. In SAS Enterprise Miner 6.1, the Graph Explore node can generate samples that are stratified by categorical target variables.
      
• Modify — The following changes have been made to the Modify tools in Enterprise Miner 6.1:
  • Drop node — The Drop node is used to remove variables from metadata, SAS tables, and SAS views. In SAS Enterprise Miner 6.1, the Drop node works on data sources other than train tables. For example, in SAS Enterprise Miner 6.1, the Drop node can be used on transaction tables.
      
• Model — The following changes have been made to the Model node tools in Enterprise Miner 6.1:
  • AutoNeural node — The AutoNeural node is used to automatically search for a Neural Network topology. The SAS Enterprise Miner 6.1 AutoNeural node adds a Target Layer Error Function property that permits a wider variety of distributions to be fitted. The AutoNeural node also adds a new final training phase that further refines the model after the topology has been selected.
      
  • Decision Tree node — The Decision Tree node builds statistical decision trees for predictive modeling. The SAS Enterprise Miner 6.1 Decision Tree node contains a new integrated interactive Decision Tree model building utility. Multiple target variables are supported for Interactive Decision Tree designs. Only one target is be used for model assessment statistic calculations. You can also use a Model Import node to select a different target variable and to generate model assessment statistics. Lastly, the default value for the SAS Enterprise Miner 6.1 Decision Tree sample sizes has been changed to 20,000.
      
  • Model Assessment Statistics — The model assessment statistics modules in SAS Enterprise Miner modeling nodes and in the model comparison node compute rank order statistics such as lift, captured response, and ROC. In SAS Enterprise Miner 6.1, a new algorithm provides faster and more accurate model assessment results. Some users might observe minor differences in the model assessment measurements when the analyzed data contains large proportions of observations that have tied probabilities. See the SAS Enterprise Miner Reference Help chapter on the Model Comparison node for more information about model assessment statistics.
      
  • Model Import node — The Model Import node imports registered models and models that were not created using SAS Enterprise Miner into the SAS Enterprise Miner 6.1 environment. The score code of the saved model is applied to the data that is used in the process flow diagram and new model assessment statistics are generated.
    You can use the Model Import node to compare registered models to newly developed models, or to apply registered model score code to new data sets. The Model Import node and the File Import node can be used together to enable users to compare models across different projects and data sources.
      
  • Neural Network node — The Neural Network node creates feed forward networks for predictive models. The SAS Enterprise Miner 6.1 Neural Network node contains a new Weight Decay property that has an initial value of 0.0. Non-zero values for the Weight Decay property will penalize the growth of weights in the neural network, and sometimes they are used to limit overfitting in the absence of validation data. The Neural Network node Properties Panel has also been reorganized for improved usability.
      
  • Rule Induction node — The Rule Induction node builds predictive models based on incrementally identifying true cases in the data. In SAS Enterprise Miner 6.1, the default maximum number of target levels to be modeled in the Rule Induction node is increased from 32 to 1024. The increase in the maximum number of target levels facilitates the modeling of problems with high cardinality.
      
• Assess —  The following changes have been made to the Assessment node tools in Enterprise Miner 6.1:
  • Model Comparison node — The Model Comparison node generates comparative statistics and then automatically or manually selects a champion model from the contender models. In SAS Enterprise Miner 6.1, the Model Comparison node can compute or recompute statistics for train, validation, and test data sets. This capability is useful when model data has been modified and new model fit statistics are needed. Use the Model Comparison node together with the Append node to partition training data in order to perform model selection, and then recombine the data and compute fit statistics for the full data. The full data fit statistics are useful for model comparison purposes.
      
  • Score node — The Score node aggregates score code from the process flow diagram to create a single, deployable score code object. In SAS Enterprise Miner 6.1, the Score node scans and manipulates the SAS score code that the process flow diagram generates in order to eliminate intermediate code that produces terms that are not deployed in the final model function. The internally manipulated code is called optimized score code. The Score node now creates optimized score code by default. The Score node can also output the nonoptimized score code for comparison.
      
    For example, the Imputation node can add SAS code that creates many new variables, but a subsequent model selection step may keep only a few of the new terms. The optimized code eliminates unused terms that were created by the Imputation node.
      
    The optimized code will have a major positive impact on scoring and deployment processes. Fewer variables will need to be saved in the score input data sets in operational systems, which can save enterprises large amounts of resources and labor.
      
• Utility — The following changes have been made to the Utility node tools in Enterprise Miner 6.1:
  • Metadata node — The Metadata node modifies the variable information, or metadata, that is passed on to subsequent tools in a process flow diagram. In SAS Enterprise Miner 6.1, you can select a single source of data and metadata for each variable table role. For example, if you have a process flow diagram with three branches, you can use the Metadata node to select a training table for one branch, a validation table for another branch, and a test table for the third branch.
      
    The Metadata node improvements also let you modify the metadata for individual variables in each table role. This function is useful when creating jobs that process many tables. Metadata node users can also merge metadata from multiple sources. Merging metadata from multiple sources is useful when aggregating the results from multiple variable selection strategies.
      
    For example, consider the task of combining the results of terms that were selected by a stepwise selection algorithm and a decision tree algorithm. You can retain terms that were selected by a single model, terms that were selected by a majority of models, or terms that were selected by all models. This capability provides users with a large degree of control over model creation strategies.
      
  • Reporter node — The Reporter node generates PDF and RTF documents for archiving and reporting. In SAS Enterprise Miner 6.1, the Reporter node provides new SAS ODS (Output Delivery System) functions. The new functions create document graphs, process flow diagrams, and analytical plots that match the graphics that are displayed in the SAS Enterprise Miner user interface.
      
    The SAS Enterprise Miner 6.1 Reporter node also provides new Decision Tree results plots for use in PDF and RTF documents. In Reporter node output, the properties list for each node tool indicates the property settings that have been changed from their default values. The Reporter results window now contains a standard external file viewer that you can use to view the PDF or RTF document that was produced.
      
• Credit Scoring —  The following changes have been made to the add-on Credit Scoring node tools of Enterprise Miner 6.1:
    
  • Interactive Grouping node — The Interactive Grouping node creates and manages the grouping of raw values into modeling terms. In SAS Enterprise Miner 6.1, the Interactive Grouping node provides improved support for special code mappings; treats interval variables that have limited numbers of values as interval variables rather than categorical variables; and adds new properties that control the binning method and the number of fine detail bins.
      
  • Scorecard node — The Scorecard node builds predictive models from scorecard functions. In SAS Enterprise Miner 6.1, the Scorecard node contains several new configurable properties. The new Model Ordering property specifies the order of terms that were entered into the regression equation model selection search. New Stay, Stop, and Force properties have been added to enhance the model selection search.

Extension Tool Programming

In Enterprise Miner 6.1, the Extension Tool Programming interface has been updated and significantly enhanced. For more information about the SAS Enterprise Miner 6.1 Extension Tool Programming Guide, see the product documentation page for SAS Enterprise Miner at http://support.sas.com.

Copyright © 2009 by SAS Institute Inc., Cary, NC, USA. All rights reserved.