Tackling the ML Reproducibility Curse with the Kedro-MLflow Plugin

Dive into the power of the Kedro-MLflow plugin — a seamless integration of Kedro’s robust pipeline management with MLflow’s comprehensive experiment tracking and model registry. In this video, we explore how combining these two frameworks can resolve the reproducibility challenges that often plague modern ML solutions.

Chapters:
00:00 – Introduction
01:29 – Use-Case Study: The Spaceflights Tutorial
05:49 – Solution Pipeline Overview
11:03 – Defining the Reproducibility Problem
14:21 – Exploring Kedro-viz
19:40 – Kedro Project Structure
23:37 – Deep Dive: ETL Application
27:17 – Deep Dive: ML Application
27:44 – Kedro-MLflow Plugin in Action
37:43 – Advanced Usage: Pipeline as a Model
39:43 – Testing the Inference Pipeline
48:42 – Kedro Context in Jupyter Notebook
56:21 – Conclusion

Links:
• Tutorial used in the video: https://github.com/OlegBEZb/kedro_mlflow_tutorial
• Kedro-MLflow plugin: https://github.com/Galileo-Galilei/kedro-mlflow
• MLflow: https://mlflow.org/docs/latest/index.html
• Kedro: https://kedro.org/
• First-party plugins maintained by the Kedro team: https://github.com/kedro-org/kedro-plugins
• Original Kedro-MLflow Tutorial which I used a lot: https://github.com/Galileo-Galilei/kedro-mlflow-tutorial
• Kedro spaceflights tutorial: https://docs.kedro.org/en/stable/tutorial/spaceflights_tutorial.html

Connect with Me:
• Telegram: https://t.me/DeepStuffChannel

Видео Tackling the ML Reproducibility Curse with the Kedro-MLflow Plugin канала Oleg Litvinov