Product Information
Announcement: Invoice Object Recommendation becomes part of Data Attribute Recommendation – Technical Details (Part 2)
Introduction
As announced in our previous blog post, we decided to merge the Invoice Object Recommendation service into the Data Attribute Recommendation service as a business blueprint. Let’s have a deep dive into the technical changes starting from December 2021 after the merger of Invoice Object Recommendation and Data Attribute Recommendation.
Why is it happening? Benefits from the Merger
With the announced merger of the Invoice Object Recommendation service into Data Attribute Recommendation, the functionality of Invoice Object Recommendation continues as a business blueprint in Data Attribute Recommendation, that would serve your business case as it used to be. Customers can now choose the Invoice Object Recommendation business blueprint model for training the data they have already uploaded.
The Invoice Object Recommendation business blueprint – as of December 2021 – leverages the Data Attribute Recommendation capabilities which sequentially will benefit SAP customers like one single solution for classification tasks or full model lifecycle. If you would like to read more on the benefits of this merge, please read the first part of this blog.
What is Changing
Here is a deeper look at what customers can expect when using the Invoice Object Recommendation business blueprint. The below table compares the main aspects in the context of handling data:
Invoice Object Recommendation Service | Invoice Object Recommendation Business Blueprint | ||
Functionality | Features |
Classify G/L Accounts (HKONT), Cost Objects (KOSTL), and CO-PA Dimensions (COPA)
|
Classify G/L Accounts (HKONT), Cost Objects (KOSTL), and CO-PA Dimensions (COPA), additionally adding more features and labels possible |
Applications & Endpoints | Single application with the same base URL | Multiple applications (three) with multiple base URLs (data management, model management, inference) | |
Data Upload | Data model management | Input data is already defined by the service | Create dataset and define dataset schema to fit requirements of the business blueprints, in addition to adding more labels |
Type | Upload data set: .CSV | Dataset schema: .JSON upload data set: .CSV |
|
Pre-processing | Handled by SAP’s side (data filtering, modelling, processing, feeding in the training pipeline) | Handled by SAP’s side (data filtering, modelling, processing, feeding in the training pipeline) | |
Conditions, batch upload |
Only limited to 20MB per upload (batch upload) File support: .CSV |
Up to 5GB (no batch upload) File support: .CSV and zipped .CSV |
|
Dataset lock/Dataset deletion | The data set is locked after all batch uploads. Not able to delete the datasets nor upload batches till lock is removed/training is completed | No data lock is required. Not able to delete the dataset unless you delete the model and the training job | |
Training | Dataset selection | N/A | Enter the Dataset’s ID for the model to be trained |
Template selection | N/A | Choose Invoice Object Recommendation Business Blueprint | |
Trigger training | N/A. Trigger the training by providing the job ID | After choosing the template and the dataset Dataset ID and providing a model name,you can trigger the training | |
Metrics | Training accuracy and test accuracy | Test accuracy, F1 score, precision and recall | |
Inference | Format | Input and Output file format: .CSV | Input and Output file format: .JSON |
Parameters | N/A | Can be defined |
It is also good to visualize the differences in the process flow through the below diagram:
Please use our dedicated Q&A section in SAP Answers to ask questions about both Data Attribute Recommendation and Inovice Object Recommendation going forward
Read our product documentation