A Kubeflow notebook is a web-accessible Jupyter, VS Code, or RStudio environment running on the same Kubeflow cluster as ML pipelines. It supports interactive data exploration while keeping experiments close to datasets, specialized resources, and deployment-like environments. section 7.1
Use Kubeflow notebooks to:
- Inspect training data and model results interactively, including loading a model for ad hoc inference. section 7.1
- Run experiments in an environment that resembles deployment by sharing base images with pipelines. section 7.1
- Strengthen versioning and CI/CD by building notebook images through the same CI pipeline as pipeline and deployment containers. section 7.1
- Access specialized resources such as GPUs while working near the datasets required for processing. section 7.4
Interactive object-detection analysis
Launch a Kubeflow notebook, inspect an object-detection training dataset, load the model, and perform ad hoc inference before turning repeatable preparation work into Kubeflow pipeline steps. section 7.0 section 7.1
