Introduction¶

https://github.com/allisterk2703/mlops-project

Team Members¶

Allister KOHN, allister.kohn@student-cs.fr

Elizaveta VASILEVA, elizaveta.vasileva@student-cs.fr

Project description¶

The project proposed to us as part of our MLOps course aims to design an MLOps workflow to industrialize the lifecycle of Machine Learning models. A classic lifecycle includes data ingestion, model training, evaluation, and deployment. The goal is to identify and implement key features to facilitate the management, scalability, and reproducibility of models in production. The project is divided into two parts:

Write a specification detailing the chosen workflows and features
Implement the corresponding code, to be hosted on GitHub.

The objective is to structure a modular, clear, and reusable system, adopting a pragmatic and robust approach.

To this end, the GitHub repository dsba-platform is a relevant starting point, as it offers a modular architecture that integrates the essential steps of an MLOps workflow. It also includes a CLI, an API (with FastAPI), and a Dockerfile, facilitating containerization, integration, and deployment to the cloud. Furthermore, its structure easily allows for adding advanced features that ensure a smooth and reproducible workflow.

The code in this repository is suited to binary classification tasks, like the famous Titanic dataset, which predicts passenger survival based on their characteristics. Consequently, we have chosen to focus on a tool intended exclusively for binary classification tasks in order to remain focused on a clear and well-controlled scope.