Hi everyone,
I wanted to start a discussion on best practices for managing experiments and model versions while working on machine learning and deep learning projects. As projects grow, it becomes increasingly difficult to track which datasets, hyperparameters, code changes, and model checkpoints produced the best results.
What tools or workflows do you use to stay organized? Do you rely on experiment tracking platforms like MLflow, Weights & Biases, or simple approaches such as structured folder naming and Git tags? How do you handle versioning for datasets and features alongside models?
I’m currently running most of my experiments on an AI PC, which makes it easy to run multiple training jobs locally. However, this also increases the risk of losing track of experiments if things aren’t documented properly. I’m curious how others balance local experimentation with reproducibility and collaboration.
Any advice or real-world examples especially from course projects or production work would be greatly appreciated. Looking forward to learning from the community!