What's New Table of Contents |
What's New in SAS Enterprise Miner 4.3
Overview
Two parallel versions of Enterprise Miner, either 4.3 or 5.1, are available in SAS 9.1. Enterprise Miner 4.3 is a continuation of the SAS client/SAS server system. Enterprise Miner 5.1 uses the production SAS 9.1 server and is a Web-deployable client application, which was developed using Java technology.
Enterprise Miner 4.3 includes the following new features and enhancements since Enterprise Miner 4.2:
- The Link Analysis, Time Series, and Memory-Based Reasoning nodes are production.
- Interactive training in the Tree node is supported through the Enterprise Miner Tree Desktop Application.
- Enterprise Miner Tree Desktop Application uses the new procedure ARBORETUM. A new view, Local Tree, is available.
- The Interactive Grouping node supports variables that have the model role of Group.
- You can register a report in the Enterprise Miner Model Repository and view the report using the Web application Enterprise Miner Model Viewer.
A comparison of Enterprise Miner 4.3 and 5.1 is summarized at the end of this document.
Details
Link Analysis Node
The Link Analysis node is now production and includes the following new features and enhancements:
- Weighted and unweighted centrality measures are computed by default.
- Any node data variable with the name TEXT, TXT, or LABEL is assigned as the variable label.
- The default number of nodes that are read from the Nodes table is increased from 32 to 100.
- The Results Browser now includes a Graphs tab. The Graphs tab displays multiple plots, such as link and node distribution histograms, plots of link Chi-square values, and plots of node centrality measures.
- The Output includes a brief summary of the nodes and links data sets.
- In the Detailed Results window, you can use any column to sort the Nodes and Links tables.
- The network display controls are now contained in the Display Control window, which enables you to modify the nodes, links, link graph layout, and cluster options from a centralized window.
- You can create a custom display of your links based on the link value. You can set the color and width of a link based on a cutoff value that you specify in the Link Value window.
- You can define your own clusters or modify existing clusters in the Interactive Clustering window.
- The MDS Options window contains a Grouping Bias option that you can use to specify how to space nodes within groups.
Tree Node
In Enterprise Miner 4.2 and earlier releases, interactive training is performed in the local SAS session. Beginning with Enterprise Miner 4.3, interactive training is supported through the Enterprise Miner Tree Desktop Application.
Enterprise Miner Tree Desktop Application
The Enterprise Miner Tree Desktop Application in SAS mode is production in Enterprise Miner 4.3. New features and enhancements include the following:
- The Enterprise Miner Tree Desktop Application now uses the new
procedure, ARBORETUM. Eventually, the ARBORETUM procedure will replace the
SPLIT procedure. In general, PROC ARBORETUM runs faster than PROC SPLIT on
larger data sets. Also, PROC ARBORETUM provides more informative
messages and enables you to
- prune specific nodes and explicitly define splitting rules
- output data sets that describe the nodes and splitting rules
- generate PMML (Predictive Model Markup Language) output
- have more control over the performance.
- The Local Tree view enables you to view a portion of the whole tree, depending on the primary selected node and the depth that you define.
Interactive Grouping Node
The groupings of a variable in your input data might have been defined by other methods outside of the Interactive Grouping node. If you set the model role to Group and the status to Use, the Interactive Grouping node places each level of the variable in a group, and calculates the weight of evidence and other statistics without applying the node settings.
Model Repository
The Enterprise Miner Model Repository has been enhanced to include the following:
- Setting the model registration options and saving models to the model repository now occurs through the Reporter node. In addition, you can set the model repository that you want to use by submitting macro variables in the Program Editor.
- You can save models directly to the model repository by using the SAS Metadata Server. You do not have to define the Open Metadata Architecture Java server to communicate with the model repository.
- You can search for, browse, and delete models from a model repository by using the Enterprise Miner Model Viewer.
Advantages of Using Enterprise Miner 5.1
The following list summarizes the advantages of using Enterprise Miner 5.1 instead of Enterprise Miner 4.3:
- Model evaluation and reporting tools provide more information.
- Graphics and data visualization tools are more tightly integrated into the product.
- Data processing and model training executes entirely on the server.
- Multiple process flow diagrams can be run concurrently.
- A single process flow diagram can execute in multiple streams on multi-CPU systems.
- Training processes can be stopped.
- The client can be run under any system, not just a SAS System.
- The server definitions and meta services are administered by the SAS Management Console.
Comparison of Enterprise Miner 4.3 and 5.1
The following table is a listing of the primary differences between Enterprise Miner 4.3 and Enterprise Miner 5.1.
| Configuration | Enterprise Miner 4.3 | Enterprise Miner 5.1 |
| Server | SAS 9.1 MVA | SAS 9.1 MVA |
| Middleware | N/A |
Java
It manages multiple users and model training can be disconnected. |
| Client | SAS 9.1 Windows | Java 1.4.1 |
| Metadata Server | Optional, but needed for model registration. | Required |
| Project Storage | Client and Server | Server |
| SAS Environment | Enterprise Miner 4.3 | Enterprise Miner 5.1 |
| Program Editor, Log, Output windows | Yes | Yes |
| Viewer for SAS/GRAPH output | Yes | Yes |
| SAS/INSIGHT | Yes | No |
| SAS DMS-based solutions | Yes | No |
| Enterprise Miner Interfaces | Enterprise Miner 4.3 | Enterprise Miner 5.1 |
| SAS System GUI | Yes | No |
| Java client GUI | No | Yes |
| Batch project execution | No | Yes |
| Web application results viewer | Yes | Yes |
| Java API | No | Yes (experimental) |
| SAS Code node interface | Yes | Yes. It provides better support for macro variables, macros, code generation, and results definition. |
| SAS Code node based custom nodes | No | Yes. XML definitions for node properties. |
| DMTOOL custom nodes | Yes | No |
| Enterprise Miner Procedures Usage | Unsupported | Unsupported |
| Metadata | Enterprise Miner 4.3 | Enterprise Miner 5.1 |
| Table and Column analytical metadata | Yes | Yes |
| Batch interface for creating table and column metadata | No | Yes |
| Batch interface for creating target decision profiles | No | Yes |
| Sample-based metadata calculation statistics | Yes | No |
| Complete data-based metadata calculation statistics | No | Yes |
| Configurable data advisory rules | No | Yes. You can set thresholds for missing percentages. |
| Extensible column attributes | No | Yes. You can add additional column attributes to be included in reports. |
| Report column attribute | No | Yes. You can include variables that have the report attribute in most reports such as score rankings and score distributions. |
| Hidden variables | No | Yes. You can hide rejected variables from the variable usage user interfaces but retain them in the data for score applications. |
| Data Model for entire project | No | Yes. It facilitates batch and GUI execution of common projects. |
| GUI Functionality | Enterprise Miner 4.3 | Enterprise Miner 5.1 |
| XML diagram exchange | No | Yes |
| Diagram copy/paste between projects | No | Yes |
| Open multiple diagrams | No | Yes |
| Open multiple results windows | No | Yes |
| Open multiple child windows in node results | No | Yes |
| Common property sheet | No | Yes |
| Individual property dialogs | Yes | No |
| Group processing | Yes (stratified, bagging, and boosting) | No |
| Job Execution | Enterprise Miner 4.3 | Enterprise Miner 5.1 |
| Stop running diagram | No | Yes |
| Run multiple diagrams | No | Yes |
| Disconnect while diagram running | No | Yes. Middleware configuration is required. |
| Continue work while diagrams running | No | Yes |
| Share projects with multiple users | Yes | Yes |
| Batch mode execution | No | Yes |
| Scheduling | No | No |
| Multi-threaded procedures | Yes | Yes |
| Multi-tasking projects | No | Yes |
| Reporting | Enterprise Miner 4.3 | Enterprise Miner 5.1 |
| HTML Reports | Yes | No |
| SAS Publish Packages (SPK) | No | Yes |
| Web application for viewing stored models | Yes | Yes |
| Interactive Analysis | Enterprise Miner 4.3 | Enterprise Miner 5.1 |
| Tree growing and pruning | Yes | Yes |
| Neural network model | Yes | No |
| Association rule: WHERE clause | Yes | No |
| Transformation bin allocation | Yes | No |
| Filter outliers selection | Yes | No |
| Link analysis | Yes | No |
| Interactive grouping | Yes | No |
| Decision threshold charts | Yes | No |
| Decision Processing | Enterprise Miner 4.3 | Enterprise Miner 5.1 |
| Class target profile matrix | Yes | Yes |
| Class target loss matrix | Yes | Yes |
| Cost values and cost variables | Yes | Yes |
| Class target variable number of decisions | Yes | No |
| Interval target decisions | Yes | No |
| Model Assessment | Enterprise Miner 4.3 | Enterprise Miner 5.1 |
| Class probability score rankings: gain, lift, etc. | Yes | Yes |
| Class probability score distributions | No | Yes |
| Classification tables in node output listings | Sometimes | Always |
| Decision tables in node output listings | No | Yes |
| Type I and Type II error table in node output listings | No | Yes |
| Interval target score rankings | No | Yes |
| Interval target score distributions | No | Yes |
| Interval target prediction vs. actual target | Yes. It is sample based and available in Model Manager. | Yes. It is user-generated plots of exported data. |
| Score rankings printed in output listings | No | Yes |
| Score distributions printed in output listings | No | Yes |
| Post-model decision matrix what-if investigations | Yes | No |
| Decision threshold charts | Yes | No |
| Komogorov-Smirnov (KS), Receiver Operating Characteristic (ROC) index, GINI statistics | No | Yes |
| Validation data assessment | Yes | Yes |
| Train and Test data assessment | Optional. You must enable it through Model Manager, | Yes |
| ROC chart | Yes | Yes |
| Scoring | Enterprise Miner 4.3 | Enterprise Miner 5.1 |
| SAS score code | Yes | Yes |
| C and Java score code | Yes | Yes |
| Separation of residual and non-residual score code | No | Yes |
| PMML generation | No | Yes |
| Nodes | Enterprise Miner 4.3 | Enterprise Miner 5.1 |
| Input Data | Yes | Yes |
| Sampling | Yes | Yes |
| Partitioning | Yes | Yes |
| Time Series | Yes | Yes |
| Variable Selection | Yes | Yes |
| SOM | Yes | No |
| Link Analysis | Yes | No |
| Insight | Yes | No. Graphs can be generated from the Results window of any node. |
| Distribution Explorer | Yes | No. Graphs can be generated from the Results window of any node. |
| Multiplot | Yes | Yes |
| StatExplore | No | Yes. This node computes univariate and bivariate distribution statistics for interval and class variables. Target and segment variables are used as by variables and/or correlation terms. |
| Merge | No | Yes. This node merges training, test, and validation data sets by row number or by ID variable. It is useful for combining predictions from multiple models or for matching ID in multiple tables. |
| Association | Yes | Yes, but the Results window does not support filtering rules and scatter plot for items. This node supports network display of rules. It generates a transposed data set that has one row per customer and variables for rules. The transposed data set can be used to cluster or predict customer behavior by rules. |
| Path Analysis | No | Yes. This node uses the new PATH procedure that includes a referrer variable for Web log analysis. |
| Transform | Yes | Yes, but this node does not support user-specified equations. It supports the creation of dummy and interactive terms. |
| Interactive Grouping | Yes. This node supports user-driven grouping of variable levels and bins based on GINI, Information Gain, and Weight of Evidence (WOE)scores. | No |
| Drop | No | Yes. This node drops variables from temporary tables for processing efficiency. |
| Filter | Yes | Yes, but this node does not support graphical selection of filter ranges. |
| Impute | Yes, this node is the Replacement node. | Yes |
| Principle Components | Yes. It is in the Princomp/Dmneural node. | Yes. This node does not support the selection of number of components in the Results window. |
| Regression (linear and logistic) | Yes | Yes |
| Dmine Regression | No. Dmine regression is available as an optional output from the Variable Selection node. | Yes. This node uses the DMINE procedure to produce models that directly include the Analysis of Variance (AOV), group, and interaction effects for interval and binary targets. |
| Decision Tree | Yes. Use the Tree Desktop Application for interactive training. | Yes. Use the Tree Desktop Application for interactive training. |
| Neural Network | Yes | Yes. This node does not support interactive training or advanced user network configuration. |
| Rule Induction | No | Yes. This node uses an algorithm for building models by recursively identifying target events. It is useful for modeling rare events. This functionality was formerly included in DMTOOL. |
| Autoneural | No | Yes. This node uses an algorithm for automated MLP network building. It selects the type and number of activation functions from four different architectures. This functionality was formerly included in DMTOOL. |
| Dmneural | Yes. This node is part of the Princomp/Dmneural node. | Yes. |
| Two Stage Model | Yes | Yes. You can specify the options for the first and second stage models, and the neural network models that have two targets. |
| Memory-Based Reasoning | Yes. This node is not recommended for score deployment, because it requires training table availability. | No |
| Ensemble | Yes | Yes. The node supports simple averaging and voting methods. It does not support bagging and boosting models. |
| Model Comparison | Yes. It is the Assessment node. | Yes. This node computes ROC and KS and automatically selects a model based on your selection. |
| Group Processing | Yes | No |
| Subdiagram | Yes | No |
| Control Point | Yes | Yes |
| SAS Code | Yes | Yes. This node provides extended support through better organized macros and macro variables. It supports building model and model assessment functions, and the creation of report tables and plots. |
| Score | Yes | Yes. If score data is defined, the node always scores data to create output view and table. |
| Score Converter | Yes | No. C and Java code are included in the SPK results package. PMML code for decision trees is available on a request basis. |
| User Defined Model | Yes | No |
| Reporter | Yes | No. Reports in the SPK format can be generated from any node. |
| Data Set Attributes | Yes | Yes. The Metadata node in Enterprise Miner 5.1 replaces the Data Set Attributes node in Enterprise Miner 4.3. |
| Data Mining Database | Yes | N/A |