Benefits of PA

[Former Member]

Hi,

I am a data scientist and have recently started to work with SAP solutions for data analysis. I have already learned HANA studio and PAL. Now I am learning Predictive Analytics. As far as I have understood, the main benefit of PA is to make data analysis easier and faster. If I know HANA studio and PAL well, I don't need PA. First, I can quickly create views in HANA studio to extract data. Second, I can easily call different data mining algorithms implemented in PAL using SQL commands. Furthermore, PAL is more flexible and I am able to do something that might not already been implemented in PA.  In my understanding, PA is more suitable for those who don't know data mining. I think it is wiser if instead of learning PA, I spend time to work with HANA studio and PAL. I would appreciate if you tell me your opinion.

Best regards,

Rasoul

[achab · Accepted solution]

Hello Rasoul,

First of all I encourage you to read the following articles which provides clarity about SAP Predictive Analytics and its current positioning:

http://scn.sap.com/community/predictive-analytics/blog/2015/06/28/predictive-smackdown-automated-alg...

http://scn.sap.com/community/predictive-analytics/blog/2015/07/02/predictive-on-sap-hana-alphabet-so...

http://scn.sap.com/community/predictive-analytics/blog/2015/10/01/dear-data-scientists-its-not-all-a...

The following whitepaper from Erik Marcade, VP of Advanced Analytics at SAP will certainly be of interest:

http://a248.g.akamai.net/n/248/420835/91770f24e9570d450e5107ed43e880a3d476b2cdbc65b5cc33afbc6c8debce...

My opinion as a citizen data-scientist is the following:

I would rather use the Automated Analytics user interface of SAP Predictive Analytics - in conjunction with SAP HANA/APL capabilities and benefit from the power of automation to tackle my problems from data preparation to modeling and ultimately massive model management.

What SAP Predictive Analytics brings me on top of SAP HANA is Data Manager that allows to create analytical data sets that are managed with time (you can easily do back testing on the version of the data as it was 3 months ago for example… with no coding).

Furthermore, it has been recognized that the fight is not only on the algorithms side but also on the derived features: APL allows to create composite variables to see if variable interactions brings more than each of the variables (one particular case is position from latitude and longitude), and APL algorithms resists with data sets with more than 20,000 columns which is a useful feature when you push the derived features to their limits.

Not even mentioning processing of missing values, outliers and the like.

Even seasoned data scientists use APL in conjunction with their own data mining skills and PAL to have a better idea on what can be achieved in terms of automated performance, and they are sometimes surprised….

Ultimately, if I need specific capabilities that are not provided by default as part of Automated Analytics, I would use Expert Analytics which provides me a graphical user interface to leverage APL, PAL and R algorithms and explore the full palette of data mining possibilities.

I hope this answer will be helpful to you.

Thanks & regards,

Antoine