Model Deployment and Prediction Processing 

Use

You must deploy the models from the data mining servers to the prediction server. The prediction tasks are processed in the prediction server.

Deployment of Models: You must deploy a data mining model for real-time prediction from a source data mining server to the prediction provider. Model deployment includes the deployment of model metadata from the source server to the model store of the Intelligence Connector. The entire model is exported from the source server to the model store in PMML2.0 format. For further details on PMML, see Predictive Model Markup Language[SAP1] .

Usually, the source data mining server is the BW system where the data mining model has been trained and can be exported to PMML2.0 format. However, the CRM Intelligence Connector is not restricted to BW but also allows for model deployment from third-party data mining servers who support specific standards such as the IBM’s Intelligent Miner.
After the model has been deployed in the CRM Intelligence Connector, it can be used for real-time prediction.  The prediction is carried out in the data mining server to which it is connected.

Processing of Prediction. The local mining server of the Intelligence Connector supports prediction for all classification and clustering models in PMML2.0 format. The Data Mining Workbench in SAP BW supports prediction for all classification, clustering, and regression models. In case a data mining model cannot be exported in the PMML2.0 format to the CRM Intelligence Connector, the processing of the actual prediction can be done in the source data mining server.

Since the regression models cannot yet exported from SAP BW to the CRM Intelligence Connector in PMML format, the prediction can be processed in the BW system. If a data mining server supports the import of a PMML2.0 model for prediction, the CRM Intelligence Connector can use this data mining server for prediction processing for every PMML2.0 model.