Skip to Content

At SAPPHIRE NOW 2016, we took the next step in an exciting transformational journey by announcing SAP BusinessObjects Predictive Analytics 3.0 to the world.   On behalf of the entire team, I am happy to announce that SAP BusinessObjects Predictive Analytics 3.0 is now GENERALLY AVAILABLE and on the SAP Service Marketplace now! 😀

 

This major update looks deceptively evolutionary but don’t let it’s Fiori-style good looks fool you – version 3.0 packs quite a punch once you learn more about it.  This is article is just a peek for what’s in store, so don’t forget to check out the official What’s New document.

 

 

First, a Word About Model Management

 

When many people think of predictive analytics, they tend to imagine the process of creating a predictive model that is both accurate (being able to predict values) and robust (continuing to be accurate as new data comes in). An initial model is created using the data we have today, but tomorrow when we get new data coming in, how do we know that the model is still efficient (accurate)?  This is typically a manual step that requires someone to periodically re-test the model to calculate its accuracy and robustness. If it deviates too far from an acceptable tolerance, a new model must be trained before it can be applied to new target data.

 

These manual validation and re-training steps are critical components of the predictive model’s lifecycle, but ones that can be mind-numbingly tedious.  However without this validation and potential re-training, the model’s accuracy suffers and you could be basing your decisions on flawed insights.

/wp-content/uploads/2016/06/automation_man_972671.jpg

Checking the efficiency of a single model every day isn’t that big of a deal, but what if you had 100 models?  How about 1000? Now what if 1% of those models require adjustment or retraining? That’s valuable time an analyst is re-solving old problems and not solving new ones.

 

 

This is where automated model management comes in. It makes far more sense for an automated system to check all those models for deviations on its own and alert the user when an exception occurs, and even to offer to retrain and reapply the new model automatically.  That sounds truly awesome – so why don’t all predictive solutions do this?  Well, it’s a little more complicated than that.

 

Detecting a model deviation is only one step – one that still requires a human to do something about it. In order to have automated model management, the system must be able to autonomously orchestrate everything from data acquisition/preparation, characteristic creation, algorithm selection, model creation, and validation all the way to model application.   Without those things, all you have is a glorified validation script that sends email to tell you there is more manual work to do!

 

 

A New Predictive Factory

 

SAP BusinessObjects Predictive Analytics 2.x (and many years before!) has always provided best-in-class automated model management capabilities, and version 3.0 continues that tradition.  The first thing you will notice is that the next generation Model Manager is now called Predictive Factory.  But this is not a simple renaming – the new Predictive Factory is built upon a more scalable and capable platform that shares some programmatic DNA with the SAP BusinessObjects BI suite.   This enterprise-grade, battle-tested platform enables Predictive Factory to continue the best-in-class predictive model management functionality of its predecessor while serving as the scalable core of our next generation solution.

 

/wp-content/uploads/2016/06/business_process_automation_972684.png

The Predictive Factory in SAP BusinessObjects Predictive Analytics 3.0 also introduces a fresh, new user interface based on the award winning Fiori User Experience. Designed with simplicity and productivity in mind, new and existing users will have a much easier time performing model management functions, aided by context sensitive help and proactive notifications in the interface.  But this is only the start of the journey – we have big plans for Predictive Factory going forward.

 

Version 3.0 also brings Segmented Time Series modelling to the Predictive Factory.  This powerful capability enables users to create a single model in a project, and with a few clicks, create thousands of segmented models based on an attribute in the dataset.  Take for example, an analyst that wants to forecast the sales each of the company’s 1,000 stores.  She can now create a forecasting model for one store and using the Predictive Factory, create the 1,000 sub-models – one specific to each store.  The analyst still interacts with a single “segmented model” but can easily drill into each individual segment to see the sub-model’s forecast.  This was possible in version 2.x’s Model Manager by writing scripts, but in version 3.0 it can be done in just a few clicks!

 

 

Extended Support for SAP HANA

 

SAP BusinessObjects Predictive Analytics is already recognized as the de facto standard predictive environment for SAP HANA, but version 3.0 adds a host of highly requested features that make it an even better choice:

 

  • Data Manager optimization for SAP HANA on sub-data manipulation
  • Increased support for SAP HANA views in Modeler and Predictive Factory
  • Generation of CCL script to apply a predictive model in SAP HANA Smart Data Streaming (SDS)
  • Updated support for SAP HANA SPS11

 

The Automated Predictive Library (APL) combines the power of our automated machine learning techniques with the in-memory processing and speed of SAP HANA.  The APL in version 3.0 now also supports Social Network Analysis algorithms that previously required processing outside of SAP HANA. We are also proud to announce that the APL is now also supported on SAP HANA on IBM Power Systems.

 

Expert Analytics in version 3.0 now brings the ability to export model chains containing different types of algorithms to SAP HANA after training as stored procedures and views. This makes it very easy to invoke a stored procedure to apply an entire model chain on new datasets and have any application consume these predictive insights without requiring special integration steps.

 

Make sure you read the latest blog by Jayant Roy, our product manager on Expert Analytics: Empower business users with predictive insights from Expert Analytics

 

Updated – Marc Daniau (one of our product managers) has posted a new blog: Make Real-Time Predictions using SAP Predictive Analytics 3.0 and SAP HANA Smart Data Streaming

 

Additional Capabilities in Native Spark Modeling

 

/wp-content/uploads/2016/06/spark_logo_trademark_972672.pngNative Spark Modeling (NSM) was introduced in version 2.5 and enabled our automated classification algorithms to run directly on Hadoop using the Spark engine.   This powerful capability opened the door to automated predictive analytics on Hadoop data without requiring data extraction and enabled our algorithms to parallelize over a Spark cluster.  Another added benefit is that business users can now simply select a Hadoop system as a data source and perform predictive analytics without requiring knowledge of Spark, Scala, or other data science-specific technology.

 

In version 3.0, we have extended Native Spark Modeling to support automated regression models as well as advanced user options such as specifying a custom cutting strategy and user defined structures and values.

 

 

 

The Next Step in the Next Generation

 

/wp-content/uploads/2016/06/assisted_technology_icon_972696.pngSAP BusinessObjects Predictive Analytics 3.0 is a major milestone for us, but it is only the first in a very ambitious plan to re-invent “predictive analytics for the masses”.  Our best-in-class automated model management has gotten even better and we are now working on re-imagining the entire predictive workflow to not only be easier, but more scalable, and yes, even more accurate and robust as well.

 

You can get the latest scoop in this vide: Predictive Analytics Reimagined for the Digital Enterprise – YouTube.   Make sure you check out PA 3.0’s SAP BusinessObjects Predictive Analytics 3.0 Useful Links as well.

 

Keep an eye out on the Predictive SCN for more details as well as articles on new features in PA 3.0 written directly by the product managers themselves. I would also recommend that you sign up for our SAP BusinessObjects Predictive Analytics Newsletter to make sure you are in the loop.

 

If you haven’t checked out SAP BusinessObjects Predictive Analytics yet, there has never been a better time (click here for our free 30-day trial of the modeler).

To report this post you need to login first.

6 Comments

You must be Logged on to comment or reply to a post.

  1. Sainath Kumar

    Hi Ashish a)Data Manager optimization for SAP HANA on sub-data manipulation b)Increased support for SAP HANA views in Modeler and Predictive Factory c)Generation of CCL script to apply a predictive model in SAP HANA Smart Data Streaming (SDS) Does this mean that from my desktop I can access the HANA views in HDB and run the algorithms based on the views ? When I am training the model will the model generation utilize the HDB or use the desktop resources to build the model ? Do we need to additionally load data into SAP PA 3.0 and then process or utilize the data existing in the views for training

    (0) 
    1. Antoine CHABERT

      Hi, it would be great if you could post a new discussion and not ask the questions in the comment, thanks in advance. We can elaborate then in detail on the different questions. In a nutshell, yes you can use the HANA views as data sources for SAP PA 3.0 and you can also delegate the data processing to SAP HANA for major type of predictive model creation and application.

      (1) 
      1. Sainath Kumar

        Thank you Antoine. I will post a new discussion next time around.

        Once I enable delegation does the model training also happen in HANA ? or it still utilizes SAP PA server/local machine memory for the processes ?

        (0) 
  2. Alejandro Serrano

    Hi Ashish,

    I think you may be able to help me. In my case I have a table in BW. I want to access to this table, pre-process it and apply a AFL predictive function (Single Exponential Smoothing). Then, I want to store the results and access them from BW. After some research, I managed to find 5 different ways to approach this problem:

    1. Using SQLScript from SAP HANA Studio
    2. Using Application Function Modeller (AFM) from SAP HANA Studio
    3. Using SAP Predictive Analytics (SAP PAA)
    4. Use BW report (ABAP) to trigger a SQL PROCEDURE
    5. Using Predictive Algorithms native in BW

    In all cases I got stack at some point. Please, have a look to https://answers.sap.com/questions/148420/five-approaches-to-execute-a-predictive-afl-functi.html.

    Thanks and regards!

     

    (0) 

Leave a Reply