What’s New in Predictive Analysis Version 1.0.11? · HANA stored procedure. ... new export to...
Transcript of What’s New in Predictive Analysis Version 1.0.11? · HANA stored procedure. ... new export to...
What’s New in Predictive Analysis Version 1.0.11? In early June 2013, SAP Released version 1.0.11 of Predictive Analysis. This release is the first to
introduce some great new features that will make it easier than ever to implement predictive models.
Nearly all the updates in this version affect the HANA-Online mode only, though the new “Add a
Component” feature is available in both online and offline modes.
PAL Algorithm Export One of the best features of Predictive Analysis is the new feature to export a saved model within PA as a
HANA stored procedure. This appears to work for PAL algorithms, excluding clustering models. In fact,
as of this release it appears that the clustering model save option has been removed altogether. This
new export to stored procedure functionality is incredibly useful because new data can be scored
without having to open and manually run a predictive workflow in Predictive Analysis—this HANA stored
procedure can be used to score any new observations or even integrated into applications or exposed as
a web service to enable real-time scoring.
In order to export a scoring procedure back to SAP HANA, you must run the analysis and then right-click
on the algorithm transform in the predictive and select “Save as Model”. The saved model then appears
on the “Saved Models” tab in the Predict pane. Right clicking on a model in the “Saved Models” tab and
clicking on “Export Model” will bring up the following options for PAL algorithm models.
Selecting “SAP HANA stored procedure” will bring up the dialog below, which allows the user to select
the procedure name that will be stored and available through SAP HANA Studio:
In HANA Studio, we can open the procedure we just created and review the code:
We can then call the algorithm within HANA Studio as follows:
And we now have a table of our predicted values from the model:
Algorithm Additions
HANA Online Algorithms
There are several new algorithms now available in HANA Online mode. These include:
HANA Normalization: a pre-processing algorithm that scales data from -1 to 1 or 0 to 1 as
defined by the user
HANA Binning: a pre-processing algorithm that discretizes continuous data, and is useful for
converting continuous data into ranges (ex. age integer into age ranges) prior to modeling
HANA Logistic Regression: a generalized linear modeling algorithm that predicts the probability
of an event occurring (always returns values between 0 and 1); however the implementation of
this algorithm in Predictive Analysis allows only numeric inputs and no categorical variables.
There were no algorithm additions in HANA Online mode.
Adding a Custom Component The “Add a New Component” button was added at the top of the Predict pane, next to the
Import/Export models buttons:
Clicking on the “Add New Component” allows the selection of “R Component” only. It then steps
through a wizard-like interface to create the component, first setting the tab the new object should
appear under and the name of the new component:
Then allowing us to input the R code (note: this new feature only accepts R functions, not R script code)
and enter in the input/output variables:
And finally allows us to configure the GUI interface (determining which variables are available for
selection and how many):
After clicking finish, we have our new algorithm available under the Custom R Component category of
the Algorithms tab:
It looks like there is no additional functionality beyond editing and deleting this component once it is
created; eventually I expect you will be able to export/import components and share them with others.
Visual Intelligence/Lumira Updates
It looks like there were also a number of updates for the visualization component of the tool; perhaps
the most notable are enhanced support for additional external data sources (Greenplum,
Salesforce.com, etc…) via Freehand SQL and support for calculated measures from SAP HANA.
Hillary Bliss, Business Intelligence Consultant
Decision First Technologies
twitter @HillaryBlissDFT
Hillary Bliss is a Business Intelligence Consultant specializing in data warehouse design, ETL
development, statistical analysis, and predictive modeling. Hillary works with clients and vendors to
integrate business analysis and predictive modeling solutions into data warehouses based on their data
and business needs. With Decision First Technologies, Hillary uses Data Services, Web Intelligence,
Predictive Analysis, and HANA. Hillary has Masters in Statistics and an MBA from Georgia Tech.