Just ran thru official tutorial and I'm pretty impressed. As I understood the idea is to: - use git branch for each experiment (change of hyperparameters etc.)

Just ran thru official tutorial and I'm pretty impressed. As I understood the idea is to: - use git branch for each experiment (change of hyperparameters etc.) - define pipeline stages (preprocessing, train/test split, model training, model validation) - after this steps you can change any part of pipeline (say data preprocessing or model parameters) and run `dvc repro` to reproduce all stages for which dependancies changed and track metrics for all branches, which os pretty cool and reduce experiment logs in wiki