mlops

Open-source tool for tracking & monitoring machine learning models.

Table of Contents

• Introduction
• Technologies
• Installation & usage
• License
• Contact
• References
• To-Do

Introduction

End-to-end machine learning projects require long-term lifecycles during which different models are evaluated, with various hyperparameters or data representations. Then, out of all the experiments, a final model must be selected for deployment in the production environment. There are some solutions available to manage the model creation process, such as mlflow or neptune.ai. However, none of them support the functionality of monitoring a deployed model in production.

As part of the mlops project, we aim to create a ready-to-use tool for professionals in the Machine Learning industry allowing them not only to manage experiments during model creation process (tracking module), but also monitoring a deployed model working on real-world production data (monitoring module)

Technologies

Application consist of two main components:

• Main application (client + server) written in React and FastAPI, which you can run using Docker.
• Python package for communication with the application.

Additionally, we use mongoDB database for storing tracking module data.

Installation & usage

To install the application locally, you need to have docker and docker-compose installed on your machine. Then, you can run the following command:

docker-compose up

After that you can access the application at http://localhost:3000.

To install the python package make sure you have Python >=3.9 installed on your machine. Then, you can install the package using pip:

pip install mlops-ai

You can find more information about the package in the PyPI documentation. Some example notebooks are also provided inside the library/tests/notebooks directory.

License

Distributed under the open-source Apache 2.0 License. See LICENSE for more information.

Contact

Project authors are (in alphabetical order):

• Jędrzej Rybczyński (back-end, library)
• Kacper Pękalski (back-end, library)
• Kajetan Szal(back-end, library)
• Paweł Łączkowski (front-end)

Feel free to contact us in case of any questions or suggestions.

References

This project was created as a final BE project of Computer Science course at Faculty of Mathematics and Computer Science of Adam Mickiewicz University.

To-Do

Application is still under development. Here is a list of features we plan to implement in the future:

• Add support for the whole monitoring module
• Add support for mail alerts
• AWS ECR integration
• Add support for multiple users (optionally)