Kubeflow takes ML to Kubernetes

The newly announced project from Google engineers, called Kubeflow, aims to leverage machine learning to address the hurdles of launching convoluted workloads on Kubernetes, which is an open source platform that serves as the backbone of container orchestration management. With the arrival of Kubeflow, Kubernetes will be able to use machine learning (ML) stacks anywhere.

Specifically, Kubeflow includes the JupyterHub platform, which enables data science and research groups to create and manage Jupyter notebook servers. Additionally, Kubeflow includes a TensorFlow Customer Resource, which can support either CPUs or GPUs, and be tailored to manage a specific container cluster size.

In a company blog post, Philip Winder, an engineer and consultant at Container Solutions, wrote, “Like DevOps has merged operations and development, DataDevOps will consume data science.”

The company also shared that it believes working with multiple environments, from development to production, will become the norm for most Kubeflow users. Consequently, Kubernetes is making use of the Ksonnet project, which is intended to make it easier to transfer workloads across multiple environments.

Kubernetes is currently working to cultivate a community around the project. Among the companies collaborating on the project are CaiCloud, Red Hat and OpenShift, Canonical, Weaveworks, Container Solutions, etc. The project was much needed to make it easier to set up and productionise machine learning workloads on Kubernetes.