Skip to Content

Abstract (source: ASUG/SAP)

This session will cover some of the external machine learning systems, and how SAP HANA can be used to integrate with them. Additionally, we will cover some of the other machine learning technologies in the SAP portfolio, and how SAP HANA integrates with them.  This webcast was hosted by BITI.

Figure 1: Source: SAP

Store models, train models, use it for scoring, part of the machine learning libraries in HANA

Figure 2: Source: SAP
Webcast focus is on 3rd party algorithms integration

Figure 3: Source: SAP
What is new in SP02 for predictive
Tensorflow is new
Also see this recording:

Figure 4: Source: SAP

Access HANA data from R studio
Write stored procedures that contain R

Figure 5: Source: SAP

Connect via ODBC

Or in HANA itself embed R

Figure 6: Source: SAP

Language type is RLANG and body of code is R, executed on R server

Results come back as a table for you to use

Figure 7: Source: SAP

Key features – R is integrated into the HANA landscape and how you can use it

Figure 8: Source: SAP

Figure 8 shows some best practices

Summary of R Capability (Source: SAP)
• SAP HANA allows use of HANA data from client side, or execute R logic from within the server
• Can be combined with SQL code and other HANA analytical functions (e.g. PAL)
• Can have multiple R Serve for better performance, availability


Key topic: SAP HANA integration with TensorFlow

Figure 9: Source: SAP

Figure 9 shows that SAP recommends starting with PAL, for performance reasons

Options for areas not covered by PAL – R or TensorFlow

Figure 10: Source: SAP

Figure 10 explains what TensorFlow is

Figure 11: Source: SAP

Figure 11 covers the architecture, integrated with Application Function Library, integrated with PAL
Easier to integrate SQL with non-SQL

Steps for using this with HANA (Source: SAP)
Two steps
Train the Model (done outside of SAP HANA)
▪ Create and train a model using TensorFlow
▪ Deploy the model to a TensorFlow Serving server (on a separate machine)
Using the Model (within SAP HANA)
▪ Register the model with the EMLAFL to create a custom procedure (similar to the PAL method)
▪ Access the model via the procedure for scoring

Figure 12: Source: SAP

You can mix and match as shown in Figure 12

Figure 13: Source: SAP

Different deployment choices are shown in Figure 13

Figure 14: Source: SAP

Verify components are installed

Check privileges are granted

Figure 15: Source: SAP

Figure 15 covers the steps on the HANA side

Figure 16: Source: SAP

Make sure HANA is aware that you have a remote source that contains Tensorflow models

Figure 17: Source: SAP

Each saved model has a signature
Map to HANA columns


Figure 18: Source: SAP

On Tensorflow, load multiple models


Summary (source: SAP)

SAP HANA EMLAFL framework enables TensorFlow models to be exposed as stored procedures
within SAP HANA
TensorFlow models can be combined with other SAP HANA models
Currently scoring is supported
TensorFlow models can be deployed on premise or in the cloud and consumed within SAP HANA


References from SAP:

SAP HANA Documentation



Coming this week – another BITI Webcast:

Alan Rickayzen, SAP Mentor Alumnus, gives a webcast on A Whirlwind Tour of Workflow in S/4 HANA Cloud & On-premise (Session 1) – see you there?

To report this post you need to login first.

Be the first to leave a comment

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

Leave a Reply