SAP FS-PER Rel 3.0 SP12 - Machine Learning functions - Stored Procedure - APL - Functionality

[former_member700635]

Hello Experts,

Hope you guys are doing well!!
We are currently exploring machine learning(ML) functions in SAP PaPM which was released in different Support packages and were able to test all SAP PaPM ML functions based on different datasets.

1. Forecast
2. Clustering.
3. Regression.
4. Classification.
5. Recommendation.

But we would like to understand the backend algorithm being used for different Machine learning functions in SAP PaPM and this would help us to explore on different business use cases based on algorithm definition.

For example 1- when we analyze Forecast stored procedure, then we have 2 different technique's mentioned in help.sap.

• The exponential smoothing technique
• The linear regressions technique

I understand that we are using Procedure syntax for forecast as per below link.

Please confirm if my understanding is correct.

https://help.sap.com/viewer/cb31bd99d09747089754a0ba75067ed2/3.3/en-US/34ec8a14b4a442a29351bb26bb082...

For example 2 - For clustering below stored procedures will be created in backend which is defined based on the settings maintained in SAP PaPM ML Functions.

But we have different algorithm's for each machine learning function as shown below.

Appreciate your help if you can share the details at the earliest.
Thanks a lot.

Best Regards
Shivaraj

[former_member672550 · Accepted solution]

Hi Shivaraj,

SAP PaPM is leveraging APL library which is more relevant for business users, comparing to PAL library that targets more advance users like data scientists (I believe your second picture is referring to PAL library), but it is not embedded in PaPM ML logic. In terms of that, APL is more business user friendly as it also takes care of data preparation automatically, e.g. normalizations, outliers, missing values, and choose the best model, therefore it requires less effort and time to address business problems for non-experienced analysts.

In Administration guide of SAP PaPM, you can find more information required for using PaPM ML functions: the first APL library must be installed by customer, but its version has to be equal or higher than 1910, and also more information about each PaPM ML type, under Analytics section. More details about APL can be found here: SAP HANA Automated Predictive Library

Considering its core mechanism, PaPM ML offers possibility to automate many models in order to solve business problems leveraging time series, classification, clustering, regression or recommendation models. It runs several models behind one ML type on trained data, determines the best model trained which is used later for prediction.

Back to your question as you ask about backend algorithms behind each ML type, in details:

Forecast: actually it is default model of the AutomatedAnalyitics engine, which means it automatically decomposes and runs bunch of models to evaluate each additive element of equation: trend, cycles and fluctuation. If none can be used, it runs Exponential Smoothing algorithm as default one. If you check APL library documentation, you will see that this forecast function can be forced to run exponential smoothing or linear regression model, but that is not the case in PaPM ML logic;

Clustering, runs k-means algorithm;

For Classification (here PaPM user can explicitly set which model is going to be applied: binary, multinominal or default one – allowing algorithms behind to choose better one) and Regression respective model based on the gradient boosting algorithm is called in generated procedure and more information can be found here: APL Service Description

Hope this helps.

Best Regards,
Rada

[shivarajsanthojirao · Accepted solution]

Hi Rada Todorovic,

Thanks a lot for sharing your insights on machine learning functions in PaPM and this will help us to build ML functions with different use cases.

Regards

Shivaraj