MFMLOps

A framework for modeling ML-based software development processes.

The framework provides a graphical tool for process modelers to design their own AI/ML development processes, which includes assets for representing activities, artifacts, resources and roles. Those entities have been identified and designed from a literature review on best and actual practices.

We have built our framework on top of the Eclipse Modeling Framework (EMF).

Additional features have been included, such as:

• A BPMN exporter that converts the process model into a BPMN 2.0 compliant file that can be run by any compatible business process automation platform.
• An HTML documentation generator that creates a single navigable document including all the information extracted from the different entities within a process model.

Installation

Requirements

• Eclipse SDK version 2023-03 (4.27)
• EMF - Eclipse Modeling Framework SDK version 2.33
• Sirius Properties Views - Runtime Support version 7.0.7
• Sirius Properties Views - Specifier Support version 7.0.7
• Sirius Specifier Environment version 7.0.7
• Acceleo version 3.7.12

Installation of the Graphical Modeler

The MFMLOps graphical modeler can be installed from its Update Site, using the following URL: https://som-research.github.io/MFMLOps/updates/

For step by step installation instructions and an initial model setup, please go to the Graphical Modeler Installation Guide.

Repository Structure

The following tree shows the list of the repository's sections:

├── edu.uoc.som.dsl4ai.TDSP                     // The source code of the TDSP modeled process using the graphical modeler
├── edu.uoc.som.dsl4ai.converter                // The source code of both the BPMN exporter and the HTML documentation generator, implemented in Acceleo
├── edu.uoc.som.dsl4ai.design                   // The designer code of the graphical modeler, including the views' definitions
├── edu.uoc.som.dsl4ai.edit                     // Code generated from the meta-model in which the designer is based on
├── edu.uoc.som.dsl4ai.editor                   // Code generated from the meta-model in which the designer is based on
├── edu.uoc.som.dsl4ai.feature                  // Feature for the Eclipse update sites
├── edu.uoc.som.dsl4ai                          // Meta-model of the graphical modeler (aka DSL4AI)
└── examples
      ├── TDSP process
      |     ├── TDSP.dsl4ai                     // Concrete syntax of the TDSP method based on the DSL4AI meta-model
      |     └── representations.aird            // XMI file with the visualization of the TDSP method activities and other elements
      ├── generatedTDSPBPMNProcessModel.bpmn    // The generated BPMN 2.0 file from the TDSP modeled process
      └── generatedTDSPHTMLDocumentation.html   // The generated HTML documentation navigable file from the TDSP modeled process  

Examples

The examples folder contains outputs based on the Microsoft Team Data Science Process. Namely:

• a TDSP process model built with the MFMLOps graphical modeler;
• a BPMN 2.0 file for enacting the process on any compatible BPM suite; and
• an HTML documentation file to browse the contents and descriptions of the process elements.

Publications

• Morales, Sergio, Robert Clarisó, and Jordi Cabot. "Towards a DSL for AI Engineering Process Modeling." In Product-Focused Software Process Improvement: 23rd International Conference, PROFES 2022, Jyväskylä, Finland, November 21–23, 2022, Proceedings, pp. 53-60. Cham: Springer International Publishing, 2022. https://doi.org/10.1007/978-3-031-21388-5_4

• Domain Specification Language for AI Engineering (PDF)

License

The source code for the site is licensed under the Eclipse Public License - v 2.0, which you can find in the LICENSE file.