The core of DVC is a command line tool. These pages contain the specifications,
self-contained descriptions, and comprehensive usage examples for
dvc -h to list them.
To run DVC in a specific directory, use
dvc --cd <path> ... before the command
and its options/arguments (does not change directories in your terminal).
- Initialize a DVC project in a Git repo with
dvc exp init.
- Copy data files or dataset directories for modeling into the project and use
dvc addto tell DVC to cache and track them.
- Create a simple
dvc.yamlfile to codify a data processing pipeline. It uses your own source code and specifies further data outputs for DVC to control.
- Execute or restore any version of your pipeline using
dvc repro, or experiment on it with
- Sharing the repository will not include locally cached data. Use
remote storage with
dvc pullto share data artifacts.