MicroStrategy ONE

Frequently asked questions

Q: What is PMML?

A: PMML stands for Predictive Model Markup Language. It is an XML standard to represent data mining models that has been developed by the Data Mining Group, an independent consortium consisting of over two dozen companies including MicroStrategy. PMML thoroughly describes how to apply a predictive model. PMML allows the use of a number of different model types as well as support for data transformation and descriptive statistics. PMML is generated by nearly all data mining applications and workbenches, including SAS®, SPSS®, Oracle®, IBM®, Teradata®, KXEN® and others.

Q: What third-party data mining tools does MicroStrategy support?

A: In general, MicroStrategy supports any tool that generates XML and that conforms to the PMML Standard (version 2.0 through 4.0). The following vendors have announced support for PMML (for more details, see http://www.dmg.org/products.html):

  • Angoss® Knowledge Studio

  • IBM ® InfoSphere Warehouse (formerly Intelligent Miner)

  • IBM® SPSS Modeler (formerly Clementine).

  • IBM® SPSS Statistics

  • KXEN® Analytical Framework

  • KNIME®

  • Open Data Group® Augustus

  • Pervasive® DataRush

  • RapidMiner®

  • R/Rattle PMML Package

  • Salford Systems®

  • SAS® Enterprise Miner

  • TIBCO® Spotfire Miner

  • Zementis® ADAPA Predictive Analytics Engine

Q: What are the support details for SAS?

A: Most SAS Statisticians use the products Base SAS (for data integration, formatting, programming, and reporting), and SAS/Stat (for data analysis and statistical techniques). For SAS to generate PMML, SAS Enterprise Miner version 9 or later must be used. Not all SAS models can be exported into PMML; consult the SAS Enterprise Miner documentation for specific details.

Q: What are the support details for IBM DB2?

A: IBM has incorporated the DB2 Intelligent Miner product functionality as part of its InfoSphere Warehouse Enterprise Edition. MicroStrategy can create and apply predictive models with DB2 by calling Intelligent Miner features using SQL.

Q: What are the support details for Teradata?

A: Teradata incorporates various data mining and advanced analysis features, including scoring accelerators and Analytical Datasets. The results of analyses are typically created and saved in structures like tables that MicroStrategy can query.

Q: What are the support details for SPSS?

A: SPSS was acquired by IBM in 2009. Most IBM SPSS products generate PMML. MicroStrategy has been working with SPSS for several years to ensure interoperability of SPSS PMML in MicroStrategy.

Q: What versions of PMML does MicroStrategy support?

A: MicroStrategy supports PMML versions 2.0, 2.1, 3.0, 3.1, 3.2, 4.0, 4.0.1, and 4.1.

Q: What models can be imported from third-party data mining applications, and which model types can MicroStrategy create?

A: The following table shows the PMML model types currently supported by MicroStrategy:

Model Type

Create with MicroStrategy

Import into MicroStrategy

Regression
(Logistic and Linear)

Yes

Yes

Decision Tree

Yes

Yes

Clustering

Yes

Yes

Mining Model

Yes

Yes

Association Rules

Yes

Yes

Time Series

Yes

Yes

General Regression

No

Yes

Neural Network

No

Yes

Text

No

Yes

RuleSet

No

Yes

Support Vector Machine

No

Yes

Sequence

No

No

Naïve-Bayes

No

No

 

Q: Is documentation available?

A: Yes. The Data Mining Services chapter in the Advanced Reporting Help contains information on data mining services, including examples.

Q: Is there online help documentation available?

A: Yes. Refer to Using Data Mining Services. This online help can be opened by clicking the Help button or pressing F1 while using an interface that is used for Data Mining Services. For example, these interfaces include the Training Metric Wizard, the Import Data Mining Model dialog box, and the Predictive Model Viewer.

Q: Are these predictive models executed in the Analytical Engine or in the SQL Engine?

A: It depends. A key differentiator is the answer to "Where is the scoring done?"

  • When you deploy the model in MicroStrategy (import PMML into MicroStrategy), the Analytical Engine does the scoring. MicroStrategy has created our own data mining functions, such as Neural Network and TreeModel, which are customized by the PMML.

  • When you deploy the model in the database, the native data mining features of that database can be used to do the scoring. Currently, MicroStrategy supports IBM DB2 Intelligent Miner by adding pre- and post-SQL statements to enable the SQL Engine to execute database data mining commands. For more information, see the Data Mining chapter in the Advanced Reporting Help.

Q: When executed in the Analytical Engine, is the processing done on the client machine or on the server machine?

A: If you use a direct project source (two-tier), the calculation is done on the client machine. If you use a server project source (three-tier), the calculation is done on the server machine.

Q: How does MicroStrategy perform predictive analysis using PMML from third-party tools?

A: To add a predictive model to a MicroStrategy project, you use MicroStrategy Developer to import the PMML that represents the model. During the import process, the PMML is examined and the appropriate function is selected based on the model type defined in the PMML, along with inputs required by the model. The combination of the function, the PMML, and the input results in a MicroStrategy predictive metric. This predictive metric is part of the metadata and can be used like any other metric.

Q: What type of data can be used as inputs to predictive models?

A: All inputs to predictive models must be metrics. Metrics provide the aggregation and dimensionality necessary to deploy predictive metrics reliably. Attributes and facts lack these features. However, it is possible to create an attribute-based metric, which is often necessary to make an accurate prediction. See the Advanced Reporting Help to learn how create an attribute-based metric.

Q: How are inputs to the model identified?

A: The PMML contains the names of each input used in the model. The Import feature in MicroStrategy Developer looks in the metadata for metrics that match that name. One of the following situations is possible:

  • There is one and only one match: That matching metric is used as the input.

  • There is more than one match: You are prompted to select the metric to use.

  • There is no match: You are prompted to select an existing metric to use from the metadata.

  • There is no match and an appropriate metric does not exist in the metadata: You must abort the import and create an appropriate metric, and then perform the import again.

Q: Does MicroStrategy have to be the source of the data used to develop the predictive model?

A: No. But to deploy the model once it is developed, the data used as inputs to the model must be available in the MicroStrategy project as metrics.

Q: What if the data is transformed?

A: MicroStrategy supports all the transformations defined in the PMML standard. However, it is common for data to be transformed prior to being used in the model, such as taking the logarithm of an input. If the model expects the input to be transformed before it is used as an input to the model, then you can create a metric that performs the transform. For example, if the logarithm of the metric Revenue is used as an input, you can create a log(Revenue) metric.

Q: What privileges are required for Data Mining Services?

A: Developer Designer privileges are required to use the Import Predictive Metric feature or to create training metrics. No specific privilege is required to use predictive metrics or training metrics.

Q: What is a training metric and what is a training report?

A: A training metric is a MicroStrategy metric that creates a PMML model when it is used on a report and that report is executed. The report that contains the training metric is called a training report.

Q: What is the difference between a training metric and a predictive metric?

A: A training metric is used to analyze an input dataset and produce a predictive model. A predictive metric uses an existing predictive model to produce predictions for input provided to the model.

Q: What kinds of neural networks does MicroStrategy support?

A: MicroStrategy supports all neural network features defined by the PMML standard, which includes Radial Basis Function Neural Networks and Multilayer Perceptrons, with various activation functions such as threshold, logistic, tanh, identity, exponential, reciprocal, square, Gauss, sine, Elliott, and arctan.

Q: Can the Predictive Model Viewer be used to edit my predictive metric or its PMML?

A: No. The Predictive Model Viewer only displays information about MicroStrategy Predictive Metrics. Any actions you take using the viewer are not written back to MicroStrategy metadata and the predictive metric itself will remain unchanged.

Q: Why am I unable to view my model in the Model tab?

A: Depending on the size of your model and available memory on the system running MicroStrategy Developer, certain large models may not be viewable. Adding more memory or closing other applications can help this situation. While the visualization may not be possible, other features should be available.

Q: Why do I get ##UNDEF## as results in simulation?

A: When the model returns an undefined score, ##UNDEF## is the result. Inspecting the inputs for that record can sometimes help determine what caused that result.

Q: When I open the Predictive Model Viewer, it says that simulation is not possible with this model. Why?

A: When the Predictive Model Viewer is opened, it attempts to load the model into the Scoring Engine. If there is a problem, a message is displayed and simulation will not be possible.

Q: When I try to simulate my model, I get an error. Why?

A: There are a number of potential reasons for this. It could be that the model cannot handle the values being passed in. Or, there could be a problem with the data type of the inputs where the model is expecting one type and receiving another.

Q: Does MicroStrategy support JSR-73?

A: JSR-73, also referred to as JDM for Java Data Mining, is a Java standard for representing data mining models in Java code. This is useful if you are moving models across components from the same vendor, since you can trust the source. MicroStrategy does not support JSR-73 since that would require the integration of third-party code from another vendor and introducing it into the MicroStrategy environment. This could potentially expose MicroStrategy users to malicious or faulty code and affect their MicroStrategy installation and stability. Therefore, support of JSR-73 is seen as a security risk. Instead, MicroStrategy uses PMML, which is a complete XML description of the model and contains no executable code. MicroStrategy can save the PMML in the metadata and use it like any other metric. MicroStrategy then interprets the PMML and generates results.