This is the third and final blog in a series of three about enriching SAP BW with SAP Predictive Analytics:
Steps of a Project
In the earlier articles of this series we have looked at the value that SAP Predictive Analytics is adding to SAP BW and how the two components can be technically integrated. In this article we look at the typical steps of a project.
Especially in a predictive project it is extremely helpful to follow a methodology to ensure that a clear business need is discovered and appropriately supported through predictions. The Cross Industry Standard Process for Data Mining (CRISP-DM) is very popular and we use it extensively at SAP. In this article we will not cover the details of this methodology, but you could read up more about it on this Wikipedia article for instance.
As it is usually the case, the technical steps break down into two phases:
- Initial setup and configuration
- Project (or use case) specific work
Step 1: Initial Installation
The following components have to be installed. Please check with your SAP Account Executive if you are not sure, whether these components are available to you:
SAP BW / SAP HANA:
- SAP BW has to be running on SAP HANA. This is a requirement, since the content will be accessed through SAP HANA Calculation Views. See the Product Availability Matrix of SAP Predictive Analytics for details on the supported environments. You can find the link to the current version on the Predictive Analytics Community.
- You should install the Automated Predictive Library (APL) on SAP HANA. The APL provides the framework that creates the predictive models automatically inside SAP HANA. Having this installed, allows models to be trained in-memory without having to extract data out of SAP HANA. Strictly speaking this is not a technical must, but I cannot think of a reason why you wouldn’t want to.
Windows Laptop of the person creating new predictive models:
- Install SAP Predictive Analytics (Desktop), which provides the graphical interface to structure your data and to create predictive models. You will spend most of the time in this environment.
- Set up an ODBC Source to SAP HANA, which is the interface used by SAP Predictive Analytics. You can obtain the ODBC drivers by installing the SAP HANA Client.
Server, to maintain models automatically:
- SAP Predictive Analytics Server and SAP Predictive Factory, which can maintain models independently. Predictive models can be retrained when needed and new forecasts can be produced on a specified schedule. For ongoing use of predictive models these components are crucial as they provide the full automation, independent of any manual steps. Should you just want to experiment with creating a model now and then, these components are not necessarily required.
- If you are using the SAP Predictive Analytics Server / SAP Predictive Factory you also need to set up an ODBC source to SAP HANA.
Step 2: Initial Configuration
It is technically possible to write the predictions into the existing data schema of SAP BW. Should you want to keep the predictions in a separate area, you can create a dedicated database schema. Please verify whether this is allowed under your licenses.
You will probably have some content in SAP BW of which you are certain that it will be used in a predictive project. Expose these objects as SAP HANA Calculation Views.
If you have installed the SAP Predictive Analytics Server / SAP Predictive Factory, you need to configure these. The ODBC source needs to be imported to the Factory for example. You can also set up users and access rights within the Factory.
After this step, you have an infrastructure in place to start creating your models!
Step 3: Create Predictive Model
Every time you want to create a new predictive model, you will come back to this step.
- If needed, expose any further required SAP BW content as SAP HANA Calculation View.
- If needed, create a semantic layer on top of the Calculation Views in the Data Manager of SAP Predictive Analytics.
- Once the data is structured as required, create a predictive model in SAP Predictive Analytics. If you have the APL installed, the model will be trained directly in SAP HANA, even if the data was semantically restructured in the Data Manager.
- From SAP Predictive Analytics you can also apply the model and write forecasts to a SAP HANA table. The target table can be created automatically from SAP Predictive Analytics. Again, creating these forecasts and writing them into a table happens without having to extract the data.
If you just want to experiment with predictive models, you might be done at this point!
Step 4: Deploy Predictive Model
If you want to put the model into productive use, you have to ensure that it is regularly retrained and that predictions are created when needed. This is taken care of by the SAP Predictive Analytics Server and SAP Predictive Analytics Factory.
- Import the metadata from the Windows Desktop into the Factory. Make sure to move both the metadata of the data structure from Data Manager, as well as the metadata of the predictive model.
- Set up a schedule in the Factory to retrain the model and to write predictions into the SAP HANA table.
- Now any application with access to this table can benefit from the predictive insight. SAP BW for instance can access that data and join the predictions with existing SAP BW content. SAP BW has been enriched with predictive insight!
You have closed the loop! SAP BW is not only showing the historic past, but also provide predictions about the future. A wide range of business process can be improved with that insight, whether through an automated decision process or by delivering additional insight to the end user.
Well, I hope you enjoyed these blogs. Thank you for reading! You can find more material in the SAP Predictive Analytics Community.