421
SQL Server 2012 Upgrade Technical Guide
Table 2: Data Mining Features in SQL Server 2012 Editions
Feature
EE
BI
SE
WE
SSE
SSEA
Standard data mining algorithms
Yes
Yes
Yes
No
No
No
Data mining tools: wizards, editors,
and query builders
Yes
Yes
Yes
No
No
No
Parallelism for model processing
Yes
Yes
No
No
No
No
Text-Mining Term Extraction
transformation (SSIS)
Yes
No
No
No
No
No
Text-Mining Term Lookup
transformation (SSIS)
Yes
No
No
No
No
No
Data Mining Query transformation
(SSIS)
Yes
No
No
No
No
No
Data Mining processing destination
(SSIS)
Yes
No
No
No
No
No
Algorithm plug-in API
Yes
Yes
No
No
No
No
Advanced configuration and tuning
options for data mining algorithms
Yes
Yes
No
No
No
No
Unlimited concurrent data mining
queries
Yes
Yes
No
No
No
No
Unlimited attributes
for association
rules
Yes
Yes
No
No
No
No
Multiple prediction targets for Naïve
Bayes, Neural Network, and Logistic
Regression
Yes
Yes
No
No
No
No
Cross-validation
Yes
Yes
No
No
No
No
Models on filtered subsets of mining
structure data
Yes
Yes
No
No
No
No
Time series:
custom blending
between ARTXP and ARIMA models
Yes
Yes
No
No
No
No
Time series: prediction with new data Yes
Yes
No
No
No
No
Time series: cross-series prediction
Yes
Yes
No
No
No
No
Sequence prediction
Yes
Yes
No
No
No
No
Please note that if you are planning to downgrade an edition when migrating from SQL
Server 2008 R2 to SQL Server 2012—for example, downgrading from the Enterprise
Edition to the Business Intelligence Edition—you are going to lose some data mining
functionality. Edition downgrading is not a supported in-place upgrade path, so you
would have to migrate your mining models using other means, as we describe later in
this chapter. Because SQL Server 2012 brings quite a few data mining enhancements
compared to SQL Server 2005, rebuilding your 2005 forecasting data mining models is
probably the best strategy. Additional validation of the 2005 predictive model is
422
SQL Server 2012 Upgrade Technical Guide
recommendable as well. Upgrading 2008 and 2008 R2 data mining models should be a
straightforward process, as there are nearly no changes in version 2012. However,
please check the discontinued and deprecated features, and the breaking and behavior
changes in SQL Server 2012, in order to prevent unpleasant surprises in your data
mining applications.
Preparing to Upgrade
After you select the SQL Server 2012 edition that suits your needs, you need to
investigate which features are deprecated in SQL Server 2008 R2. These features will
not affect your upgrade, but you will need to update your models to stop using them
before your next upgrade. You also need to know what functionality cannot be
upgraded because is it discontinued or because it has changed in SQL Server 2012. And
you also need to be aware of some behavior changes between data mining in SQL
Server 2005, 2008, 2008 R2, and 2012; otherwise, you could get unexpected results.
Let’s look at each of these categories of changes. This section also notes potential
issues with data mining models. For a complete reference of SSAS changes in SQL
Server 2012, see
SQL Server Database Engine Backward Compatibility
(http://msdn.microsoft.com/en-us/library/ms143532(SQL.110).aspx) in SQL Server 2012
Books Online.
Deprecated Features
SSAS 2000 supports the XML markup language called Predictive Model Markup
Language (PMML) 1.0. PMML is a standard language to describe data mining models.
However, the language is incomplete from the standards point of view, although some
specific extensions have been added. In contrast, SSAS 2012 supports standard PMML
2.1 and deprecates SSAS 2000 PMML extensions, meaning that you should not use
them. Note that this is probably not a big issue because you use PMML directly only if
you export your SSAS 2000 mining models to PMML. You can create a mining model in
SSAS 2012 from PMML and store it in an SSAS 2012 database. If you export it from
SSAS 2012, standard PMML will be generated. Some of the most important SQL Server
2000 extensions to PMML 1.0 include:
Support for nested tables.
The Discretized, Ordered, and Cyclical model variables besides the simple
Categorical and Continuous model variables.
Support for Key columns in nested tables.