Let us take a brief diversion from the blog series and continue a bit more in-depth into the concepts of how machine learning is embedded into SAP S/4HANA business processes and the applications. We will also briefly understand on how the side-by-side models are adapted as well leveraging the ISLM framework in detail.
In the earlier blog, we discussed the concepts of how machine learning is embedded leveraging the PAi framework and which is now enhanced using the ISLM framework. The blog series part 6, part 7 and part 8 discuss the various embedded as well as the side car approaches in building machine learning with SAP S/4HANA. In this blog, I would further explain the naunces of how ISLM is leveraged in these approaches.
Let us dive a bit into the ISLM framework concepts, the services and how this is leveraged for building intelligence into SAP S/4HANA. We will continue further by explaining the machine learning based business scenarios from the SAP S/4HANA centric perspective. Typically the ISLM framework is designed to handle the various operational requirements of building machine learning into SAP S/4HANA processes and also act as a self-service tool for the customers. The framework is available for both the On-Premise as well as the Cloud versions of SAP S/4HANA to reduce the total cost of ownership with regard to adapting machine learning into the business processes.
Some of the considerations include that the ISLM component is stacked into SAP S/4HANA to handle the machine learning functionality in a better way. This helps to build the machine learning scenarios in an SAP S/4HANA centric way. Additionally abstracts the differences between the embedded as well as the side-by-side models so that a regular business user or an end user can build the intelligent scenarios.
The ISLM framework is integrated into the ABAP layer (SAP Basis) so that the intelligent scenarios from above layers in SAP S/4HANA stack can utilize the framework completely. The embedded scenarios from the PAi framework are supported. Additionally, the enhanced framework of PAi (the ISLM) can handle side-by-side use cases as well. The ISLM framework also has a data model representation of machine learning scenarios (both embedded and side-by-side based on SAP Data Intelligence). It also provides a possibility to implement the prerequisite checks for machine learning scenarios delivered from different SAP application areas.
With ISLM framework you could maintain a good approach since it provides a consistent experience to the end user or the intelligent scenario owner while building the embedded or the side-by-side machine learning scenarios while doing the model development. A quick overview of the support provided by ISLM involving the machine learning operations include:
- Triggering the training of a scenario with technical parameters.
- Allow to compare different training results with metrics and quality information.
- Deployment of trained models for application consumption.
- Activation and deactivation of various deployment versions.
- Stable inference consumption to the business applications.
In a typical intelligent scenario, only one machine learning approach is represented in ISLM which helps to build the scenario – any change in the scenario triggers the creation of a new intelligent scenario.
While the ISLM framework resides in the ABAP layer in SAP S/4HANA, it can trigger the entities in SAP Data Intelligence for the creation of these intelligent scenarios for the side-by-side approach. ofcourse, the connectivity between the SAP Cloud Platform and SAP S/4HANA is pre-configured with a specific SAP DI tenant. This creates a 1-1 connection between the SAP DI tenant and the SAP S/4HANA system for a particular intelligent scenario which is maintained in an ISLM specific configuration table. Typically, training and execution of the intelligent scenario happens in the same key/value pair configuration which maintains the system load and system performance. Eventhough an intelligent scenario of type “side-by-side” is mapped to one ML scenario with a unique ID on the SAP DI side, this kind of intelligent scenario may have multiple versions on the SAP DI side but on ISLM side there is only one release version of the same for operations visibility.
While discussing the side-by-side approach, we should also note that the SAP DI scenario has only one training pipeline and one inference pipeline with the mapping between them derived by default. The connectivity mechanism between SAP DI and SAP S/4HANA is abstracted to the ISLM framework since the SAP DI does not have an ISLM specific operator in the the pipelines to interact with the ISLM framework in the ABAP stack. The existing Fiori applications of the PAi framework are already ported into the new ISLM framework which is available to all the application areas in the SAP S/4HANA stack and establishes the usage statistics with better monitoring of the trained models and offer a good supportability for side-by-side as well as the embedded ML scenarios.
The following figure explains the architecture with regard to the embedded and the side-by-side models in the context of using the ISLM framework.
The above figure explains the personas on the embedded and the side-by-side based scenarios leveraging the different approaches and how they are built to enhance SAP S/4HANA with machine learning. Let us unpack the above figure and dive into more architecture details.
On the left-hand side, you will notice how the data science developer builds the machine learning scenario with the various data science tools depending on the approach chosen, then the intelligent scenario owner registers the scenario and publishes into the SAP S/4HANA. Then finally, the ABAP developer builds the UI code that embeds into the SAP S/4HANA business app leveraging the Fiori technology. You can also see how both the embedded and the side-by-side artifacts are leveraged by the Intelligent Scenario Manager in SAP S/4HANA.
While on the left-hand side we dealt with how the scenario is developed, registered and embedded into the S/4 apps, on the right-hand side we will see how it is executed by the end user.
On the right-hand side, you will see how the business administrator or the analytics specialist can enable the machine learning scenario by training, deploying and activating the scenario. You will then see how the business user or the end user can leverage this functionality for performing the needed business actions.
In the bottom box, you will see the various data science tools & products that are leveraged in building the embedded or the side-by-side scenarios. In the side-by-side approach, in addition to the SAP DI and the SAP AI business services, you will also see how the predictive models built using the SAP Analytics Cloud – Smart services can be published back into SAP S/4HANA. We discussed this very briefly in our blog, extending machine learning with SAP Analytics Cloud.
These various approaches are built in the perspective of the embedded and the side-by-side scenarios which also provide in the context of how these personas are involved in building these different approaches.
The machine learning scenarios as we explained above can be developed in the following ways – talking in terms of the architecture drivers:
- Embedded scenario can be developed in the SAP Analytics Cloud with HANA APL and imported into ISLM in SAP S/4HANA.
- Embedded scenario can be developed in HANA Embedded Machine Learning Interface (leveraging HANA PAL) and imported into ISLM in SAP S/4HANA.
- Embedded scenario can be developed directly in ISLM using capabilities of SAP HANA libraries.
- Side-by-side scenario can be developed via SAP DI and register to delegate the ABAP class in ISLM.
Any of the machine learning scenarios developed irrespective of the type will be operated in ISLM.
The ISLM framework is delivered with 2 Fiori applications namely, the Intelligent Scenarios and the Intelligent Scenario Management. These apps will be accessible via the Fiori Launchpad based om “Intelligent Scenarios” business catalog.
Let us now briefly discuss about these 2 apps, the Intelligent Scenarios and the Intelligent Scenario Management and how they are used to create the intelligent ML scenario.
This is a typical registration and a creation app for the developers to register the intelligent scenarios in the ISLM framework with the basic details and an ABAP class representing the intelligent scenario. This app will be accessible to the customers to create a custom development of intelligent scenario. Additionally, for the embedded approach, the framework generates a class internally during this registration process. In the side-by-side use case (machine learning logic class), the developers need to create their own class. Eventually ISLM will offer an ABAP development tool in native integration with SAP GUI to create intelligent scenarios.
Intelligent Scenario Management:
Using this app, the business administrators (domain experts) can perform the operations like prerequisite check, training, deployment, activation and monitoring of the intelligent scenario for a specific business domain. This app is built with more focus on non-ML expert expectations to operate intelligent scenarios.
A few points to summarize:
- Since the ISLM framework is part of the ABAP platform, the system landscape design is similar to the ABAP platform deployments.
- Usage of HANA as a database is a must for the embedded use cases since it is dependent on the HANA libraries.
- For the side-by-side scenarios, external SAP DI tenant is required and all the PAi based applications are to be migrated to the ISLM framework.
Here are some quick links to the blogs in this series to give you a complete understanding of how Predictive Intelligence is infused into SAP S/4HANA.
- Resources and journey to machine learning with SAP S/4HANA
- Part 1 – Leveraging Predictive Intelligence with SAP S/4HANA
- Part 2 – Architecture and deep-dive of the different approaches around Predictive Intelligence
- Part 3 – Process flow leveraging Machine Learning and Predictive Analytics
- Part 4 – Scope and functionality in the context of an end-to-end process leveraging ML
- Part 5 – Activating machine learning functionality for SAP S/4HANA
- Part 6 – Building ML into the digital core of SAP S/4HANA – Embedded ML
- Part 7 – Enhancing the digital core with ML Services (Side-by-Side ML)
- Part 8 – Extending the digital core by leveraging ML with SAP Analytics Cloud
- Part 9 – ML Extensions to SAP S/4HANA processes
- Blog series – ISLM for machine learning with S/4 (this blog)
- Introducing the book – Implementing Machine Learning with SAP S/4HANA
Happy predicting the future!!