Skip to main content

FedML-Databricks for Databricks platform

Last updated on by s-krishnamoorthy

FedML-Databricks provides end-to-end integraton for training models in Databricks machine learning platform, using live business data from SAP systems and eliminates the need for duplicating the data. With only few lines of code, fedml-databricks enables Data discovery, Pyspark support and deployment support to SAP BTP Kyma all while enabling instant access to source business data from SAP systems.

Architecture​

image of solution diagram
Copy to clipboard

Solution Diagram Resources
You can download the Solution Diagram as a .drawio file for offline use. Alternatively, you may view and edit the Solution Diagram directly on draw.io.
Please note that any changes made online will need to be saved locally if you wish to keep them.
DownloadEdit
Resources

Flow​

FedML, the Python Library is imported directly into Databricks workspace's notebook instances. FedML connects to SAP Datasphere via secure Python/SQLDBC connectivity and helps federate the critical business data needed for training models in Databricks platform.

Models trained in Databricks ML platform can also be optionally deployed in SAP BTP Kyma for inferencing via FedML-databrick's seamless deployment integration.

When to use​

  1. When a customer already has Databricks as part of their cloud platform strategy, and have invested in using Databricks ML as their data science platform for machine learning projects.
  2. Majority of training (non-SAP) data resides in the Databricks delta lake tables , with critical SAP data from various SAP applications (with semantics intact) is still needed for training.
  3. Trained models have potential to be deployed in SAP BTP Kyma for quick inferencing that involve SAP data.

Resources​