6_kubeflow_pipelines.md

Kubeflow Pipelines

Table of Contents:

• Components of Kubeflow Pipelines
• Executing a Sample Pipeline
• Delete Resources
  • Remove Kubeflow
  • Remove the Kubernetes Cluster

Kubeflow Pipelines is a simple platform for building and deploying containerized machine learning workflows on Kubernetes. Kubeflow pipelines make it easy to implement production grade machine learning pipelines without bothering on the low-level details of managing a Kubernetes cluster.

Kubeflow Pipelines is a core component of Kubeflow and is also deployed when Kubeflow is deployed.

Components of Kubeflow Pipelines

A Pipeline describes a Machine Learning workflow, where each component of the pipeline is a self-contained set of codes that are packaged as Docker images. Each pipeline can be uploaded individually and shared on the Kubeflow Pipelines User Interface (UI). A pipeline takes inputs (parameters) required to run the pipeline and the inputs and outputs of each component.

The Kubeflow Pipelines platform consists of:

• A user interface (UI) for managing and tracking experiments, jobs, and runs.
• An engine for scheduling multi-step ML workflows.
• An SDK for defining and manipulating pipelines and components.
• Notebooks for interacting with the system using the SDK. (Taken from: Overview of Kubeflow Pipelines)

Executing a Sample Pipeline

1. Click on the name [Sample] Basic - Condition.

2. Click Start an experiment.

3. Give the Experiment a Name.

4. Give the Run Name.

5. Click on the Run Name to start the Run.

Delete Resources

See the end of Deploying an End-to-End Machine Learning Solution on Kubeflow Pipelines to delete billable GCP resources.