data-version-control

What problem does it solve?

Data-intensive research often becomes impossible to reproduce when datasets, preprocessing logic, and experiment outputs change over time without a reliable version history.

Core Features & Use Cases

• Git-like dataset versioning for large files: track dataset states with DVC so you can retrieve exact inputs later.
• Reproducible pipelines with dvc.yaml: define stages (preprocess → train → evaluate) so only changed dependencies rerun.
• Experiment tracking and comparison: use DVC experiments to compare metrics, artifacts, and hyperparameter settings across runs.

Use case example: you train multiple ML models across different preprocessing settings and hyperparameters, then reproduce the exact best-performing dataset + pipeline + metrics on another machine or in CI.

Quick Start

Use the DVC pipeline to add a dataset to version control and run the full reproducible workflow with a single command: dvc repro.