Edit on GitHub

Machine Learning Experiment Tracking

Making progress on data science projects requires a large number of experiments — attempts at tuning parameters, trying different data, improving code, collecting better metrics, etc. Keeping track of all these changes is essential, as we may want to inspect them when comparing outcomes. Recovering these conditions later will be necessary to reproduce results or resume a line of work.

natural experimentation Difficulty tracking experiments manually (usually with notebooks or speadsheets)

DVC provides a layer of experiment management features out-of-the-box (no need for special servers or websites). Running DVC Experiments in your workspace captures relevant changesets automatically (input data, source code, hyperparameters, artifacts, etc.). This is possible because your entire ML pipeline is codified with DVC.

Other tools tend to focus on experiment navigation by saving metrics and artifacts that result from your pipelines, along with fragile links to code revisions. DVC's approach guarantees reproducibility by working on top of Git instead, and not as a parallel system.

  white:**Experiment**                yellow:**loss**      yellow:**acc**      blue:**epochs**   blue:**model.conv_u**
  workspace                 0.23508   0.9151   10       24
  main                            -        -   10       16
  ├── e2647ef [exp-ee8a4]   0.23146   0.9145   10       64
  ├── 15c9451 [exp-a9be6]   0.25231   0.9102   10       32
  └── 9c32227 [exp-17dd9]   0.23687   0.9167   10       256
  7317bc6                         -        -   100      16
  ├── 299dc5e [monograms]   0.24838   0.9176   100      128
  └── 8a9cb15 [bigrams]     0.24459   0.9134   100      64
_DVC Experiments are organized along project versions (Git commits, branches, tags, etc.)_

Major benefits of tracking experiment with DVC:

  • Quickly iterate on experiment ideas, with automatic bookkeeping of data dependencies, code, parameters, artifacts, ML models, and their metrics.
  • Use a controlled execution mechanism; Run one or queue many experiments (and run them in parallel if needed).
  • Create deep learning checkpoints from your code, and track live metrics.
  • Review and compare results based on params or metrics; Restore them from cache or reproduce them from scratch.
  • Adopt or stay on a Git workflow (distributed collaboration) and services such as GitHub.
  • Enable Experiment Versioning practices.

📖 Ready to dive in? Get started with experiments!

Among other differentiators (below), DVC Experiments are unique in the space in that they provides a simple yet flexible, local-first experience. Your code is unchanged and you control where data is saved and shared. DVC also improves storage efficiency and saves you time via caching, preventing repetitive data transfers or having to retrain models on-the-fly.

DVC ExperimentsExisting tools
UITerminal + webWeb (usually SaaS)
LoggingGit-basedCustom formats
StorageData versioningLogging artifacts and metrics
Executiondvc exp runCode API (usually Python)

💡 Note that other experiment tracking tools can be complementary to DVC, for example for detailed experiment environment logging with specialized visualizations.

Finally, DVC is completely language agnostic, offering the same functionality whether you're using Jupyter notebooks or Scala, CSV data frames or HDFS.

To complete your stack, we also offer DVC Studio, an advanced web interface to collaborate on DVC projects online, including full experiment management support. You can also integrate with CI/CD using CML, and run your experiments there straight from Studio!

🐛 Found an issue? Let us know! Or fix it:

Edit on GitHub

Have a question? Join our chat, we will help you:

Discord Chat