Introduction to the Pipelines Interfaces

The ways you can interact with the Kubeflow Pipelines system

This page introduces the interfaces that you can use to build and runmachine learning (ML) workflows with Kubeflow Pipelines.

User interface (UI)

You can access the Kubeflow Pipelines UI by clicking Pipeline Dashboard onthe Kubeflow UI. The Kubeflow Pipelines UI looks like this: Pipelines UI

From the Kubeflow Pipelines UI you can perform the following tasks:

  • Run one or more of the preloaded samples to try out pipelines quickly.
  • Upload a pipeline as a compressed file. The pipeline can be one that youhave built (see how to build apipeline) or onethat someone has shared with you.
  • Create an experiment to group one or more of your pipeline runs.See the definition of anexperiment.
  • Create and start a run within the experiment. A run is a single executionof a pipeline. See the definition of arun.
  • Explore the configuration, graph, and output of your pipeline run.
  • Compare the results of one or more runs within an experiment.
  • Schedule runs by creating a recurring run.

See the quickstart guide for moreinformation about accessing the Kubeflow Pipelines UI and running the samples.

When building a pipeline component, you can write out information for displayin the UI. See the guides to exportingmetrics and visualizing results inthe UI.

Python SDK

The Kubeflow Pipelines SDK provides a set of Python packages that you can use tospecify and run your ML workflows.

See the introduction to the Kubeflow PipelinesSDK for an overview of the ways you canuse the SDK to build pipeline components and pipelines.

REST API

The Kubeflow Pipelines API is useful for continuous integration/deploymentsystems, for example, where you want to incorporate your pipeline executionsinto shell scripts or other systems.For example, you may want to trigger a pipeline run when new data comes in.

See the Kubeflow Pipelines API referencedocumentation.

Feedback

Was this page helpful?

Glad to hear it! Please tell us how we can improve.

Sorry to hear that. Please tell us how we can improve.

Last modified 28.06.2019: Added an overview of Pipelines interfaces (#829) (eab9c56a)