Use case tutorials

The following tutorials describe some more specific use cases that you may find helpful.

•  Tracking experiments

  ---

  • Track and organize model-training runs – in this tutorial, we train a classifier model and explore the central experiment-tracking features in Neptune.
  • Track distributed training jobs – learn how to track metadata from single-node, multi-node, or multi-GPU jobs.
  • Log in a sequential pipeline – we demonstrate how to access the same run from multiple scripts and organize the metadata per pipeline step.
  • Track and visualizing cross-validation results – see how to use Neptune namespaces to organize cross-validation metadata.
  • Track HPO jobs – use Neptune to track metadata using either a single run or separate runs for each trial.
  • Monitor model training live – how to use Neptune to monitor metrics during training.

•  Reproducibility

  ---

  • Reproduce a run – you can reproduce any run by retrieving its metadata and logging it into a newly created run.
  • Restart a run from a checkpoint – save checkpoints and, when necessary, resume the run from them.
  • Re-run a failed training – model training runs don't always go as planned. Learn how to fetch the parameters and metadata of a failed Neptune run and use them for a new run.

•  Data versioning

  ---

  These tutorials walk you through artifact version tracking and comparison.

  • Version datasets in runs
  • Compare datasets between runs
  • Share dataset versions on project-level

---

The related Colab notebooks and all necessary scripts are located in our GitHub examples repository.

Visit examples repo