Machine Learning Cockpit – What’s New in 2302
SAP Delivered Scenario
Maybe you like an idea of using machine learning for making predictions in SAP Business ByDesign And you heard of Machine Learning Cockpit, released in 2211 but you are not sure how to start?
In SAP Business ByDesign release 2302 we provide out-of-the-box one scenario that you can inspect and try it out directly. Objective of the scenario is to predict if Opportunity will end up Won or Lost.
You can find the scenario in Scenario work center view after you select view filter All SAP Delivered Scenarios. Name of the scenario is CRM_OPP_SUCCESS.
For detail instructions how to use the scenario please refer to Online Help > Use Cases > Opportunity – Success Prediction (SAP Delivered Scenario).
Machine Learning Cockpit for SAP Business ByDesign allows to you design you own predictive solutions end-to-end without ever leaving SAP Business ByDesign UI.
In case, you are searching for inspiration of what exactly could be predicted and how to set it all up make sure to check out new repository of ML use cases in Online Help > Use Cases.
As of release 2302 we provide detailed, step-by-step instructions how to implement following solutions:
- Opportunity – Success Prediction (SAP Delivered Scenario)
- Sales Quote Item – Success Prediction
- Purchase Order Item – Prediction of Incomplete Delivery
- Purchase Request Item – Supplier Proposal
- Customer – ABC Classification Proposal
- Outbound Delivery – Cancelation Prediction
Feel free to try them out.
We are working on more use cases, so stay tuned.
Feature functionality is never complete if the feature is hard to use. Feature functionality is never complete if it is hard to understand what-is-what in UI, and what exactly I am expected to do in any specific step. In Machine Learning Cockpit team we never take UX lighthearted.
In 2302 version we have made two improvements which I would like to explain bit bit more in details
Scenario Field Selection
Selection of fields in Scenario requires that you need to decide and manually select those fields that will be used in scenario Models. Specifically, you need to flag a Target Field.
Before 2302 – field selection was error-prone process because field names and field selection checkboxes were on different parts of the screen. It was easy to make a mistake and select wrong field, or to select regular field as Target Field. It was also hard to tell what field was selected as Target Field, if any.
Starting 2302 – the Target Field is selectable from new selection-box and its value is always visible. Field selection checkboxes are positioned next to field names, bringing clarity what fields are selected and what not.
Model Training Results
Model Training Results provides quantitative indicators of Model quality
Before 2302 – there were four different Model Evaluation Indicators displayed and it was not clear which one was the most important one. Secondly, the Field Contributions list was unsorted, showing 5 rows only. To see and understand how important individual fields were for model training you needed to expand the list and sort it manually.
Starting 2302 – we show Model Accuracy only, as dominant indicator of model quality. Experts, who demand more detailed indicators will find them organized in a table – Model Evaluation Indicators per Class. The Field Contributions lists was enlarged and sorted by Contribution ranking by default.
These were not the only UX improvements in 2302. Several other UX issues were fixed, which are maybe more subtle to notice. However, these two are the most visible ones and I believe that you will like them.
Feel free to share your experience and impression either, here, in comments. Or drop me an email at: Dalibor Knis . Thank you.
thanks a lot! I have created the Outbound Delivery – Cancelation Prediction with a run, but it doesn't write anything into my custom fields. The protocol of the run shows only "
Any advice would be appreciated!
thank you for your feedback.
this message is returned from application layer when Prediction Run tries to copy prediction values to the document. Typically, closed document cannot be updated, e.g.
The Filter Condition in Prediction Run should include to prediction dataset only editable documents.
Have all packages returned the error for every documents within a package? Or are there also successfully updated documents?
Another option how to get overview of prediction outcomes is to check data source Prediction Result (MLSINFB) and filter data for your Scenario only. The selecting Prediction Outcome Name as characteristic will give you statistics of predicted values. The value "#" represent the case when document could not be updated.
All packages have returned the error. The filter is set to only Outbound Deliveries that are not released, so they are not locked:
The report is ok, but I would really like to make it work so that predictions are directly showing in the Outbound Deliveries...
in case ALL documents returned error then you should not be able to update any these documents manually neither.
However, if you are able to update them then please report and incident. I have never seen such situation... Thank you.
BTW: checking the MLSINFB would give you confirmation whether all documents returned error or only some of them. Also please take into account there could be multiple reasons why documents are not updatable.
Thank you, I will report an incident.