We are happy to announce new DVC release 0.9! This is our biggest release yet.
DVC was significantly improved in the new version:
- Symlinks were replaced with hardlinks. So, your data files look more natural in a workspace.
- No predefined data/ and state/ directories. Data files can be in any place of your repository.
- Now DVC writes meta-information in special DVC-files (Dvcfile or .dvc) instead of creating files in state/ directory.
- Cache directory cache/ and config file were moved under .dvc/ directory.
- We had to refuse compatibility with 0.8 due to the improvements from above.
- To avoid uncertainty, $ dvc run does not auto-derive output files anymore. All dependencies and outputs have to be explicitly specified.
- A directory can be specified as dependency or output.
- Now DVC provides packages for major OS: Mac OS X, Linux, and Windows. Pip installation package is available: $ pip install -U dvc. For Mac OS X brew is also supported.
- Documentation was added to the website.
- DVC tutorial was updated.
To improve communications and collect more user feedback we added a discussion forum - https://discuss.dvc.org in addition to our traditional communication channel - issue tracker. Please feel free to submit any feedback or comment regarding DVC.
We crafted a fun video about reproducibility issues in machine learning and DVC: