MLOps Guide on JayeX - Cloud Native CI/CD Built On Kubernetes

Introduction

Fri, 13 Mar 2020 15:03:05 +0000

Understanding MLOps

When we talk about MLOps, what we are referring to is best defined as the extension of the DevOps methodology to include Machine Learning and Data Science assets as first class citizens within the best known methods of DevOps. The aim is to demystify Machine Learning and re-integrate the delivery of technology solutions as a seamless process of product development and deployment.

The Continuous Delivery Foundation hosts a collaborative Roadmap for MLOps and JayeX is committed to implementing features aligned to that model.

Using Machine Learning Quickstarts

Fri, 13 Mar 2020 15:03:05 +0000

The JayeX MLOps Quickstarts Library provides template projects to make it quick and easy to set up everything you need to get started with a building a new ML-based asset.

Each quickstart project comprises two repositories, one which contains your training script and a second which takes the final model you have trained and wraps it as a RESTful service for deployment into your overall solution.

Getting started

You can create an instance of a project using the command:

Finding a Machine Learning Quickstart

Fri, 13 Mar 2020 15:03:05 +0000

This directory is intended to help you find your way around the JayeX MLOps Quickstarts Library and get you up and running rapidly with a template project based around the class of Machine Learning approach you wish to work with and the language and framework you prefer.

The directory is divided by target programming language (Python only at this stage, but with additional quickstarts to follow in other languages) and then by ML framework.

Working with GPUs

Fri, 13 Mar 2020 15:03:05 +0000

To use CUDA to accelerate your ML training and services, you first need to set up your Kubernetes cluster to add some physical GPU resources to your nodes.

To do this, typically you will need to request an allocation of GPU resources from your Cloud provider and then configure an additional Node Pool to provision a set of Nodes such that each Node has access to at least one physical GPU card.

Managing Data

Fri, 13 Mar 2020 15:03:05 +0000

In most scenarios, you will be expecting to remotely access data hosted elsewhere on your network and can manage this in code as part of your training scripts and service implementations. There are however a couple of situations in which JayeX can help you to manage certain types of data.

Handling data in Buckets

If you are working with data in the form of arbitrary files, you can transfer these to your training environment via a Storage Bucket in your Cloud project.