Mlops column introduction

from 2016 From the year onwards , Machine learning fire represented by deep learning , Until now, , Algorithm engineer 、 Data scientist 、 Machine learning Engineer 、 Data engineers are becoming more and more popular in the market . Maybe you are one of them , Maybe you want to join them . You need to maintain a right AI Keen perception of new directions in the field , Will not miss the development opportunity of the times again and again .

You know, AI What practitioners are concerned about ？

They have long been divorced from the pursuit of ultra-high accuracy in the ideal laboratory environment 、 Recall rate 、F1-score, But to seek high quality in the business environment 、 Low risk large-scale deployment , agile 、 Standardized delivery process , Realize the closed loop of the whole life cycle of machine learning . Such a set of management process is derived from software engineering project management DevOps, But the difficulty is much higher than DevOps, Its name is MLOps.

Why do you say MLOps The implementation difficulty of is much higher than DevOps Well ？

1.DevOps Management of revolves around code , and MLOps You need to code 、 data 、 Algorithm 、 Models and other process objects and products are managed ;
2.DevOps The change process of code optimization iteration of is relatively gentle , and MLOps Training data in machine learning 、 Reasoning data is dynamic , There is a potential problem of data drift , After the model goes online, it will face the risk of degradation .
3.DevOps Our team members are business demanders 、 Software Development Engineer 、 Software test engineer , and MLOps The team members involved are business demanders 、 Data Engineer 、 Data scientist 、 Machine learning Engineer 、 Model review team , There may also be IT Relevant software development engineers 、 Test Engineer 、 O & M Engineer , Different teams have different processes and tools , There are many obstacles in cross team communication and coordination .

MLOps Who is suitable for learning ？

Whether you are a machine learning related student in school 、 practitioners , I'm also a colleague in the middle and downstream business department of the enterprise , As long as your organization uses machine learning enabling business , Machine learning is not run Just a moment , Instead, we expect to generate business growth points and new values , You should understand the management system and cultural concept of the whole life cycle of machine learning . At work , The author found that , from 2021 At the end of the year to 2022 In the first half of , More and more AI The project manager of the team and even the manager of the organization began to pay attention to MLOps, Hope to pass MLOps Improve the R & D efficiency of machine learning projects , Use as little manpower as possible 、 financial 、 Manage the cost to realize the standardization of the model in the production environment 、 Standardization 、 Integrated smooth operation .

In the next three to five years , I predict , Yes MLOps Understanding and mastering methodology will become a bonus item for machine learning entrants or job transfer interviews . meanwhile MLOps It will definitely breed new job opportunities , Future period ！

ML、DevOps They are all professional fields , And the combination of the two MLOps It is a high-intensity specialty , Engineers need data analysis 、 Model development 、 test 、 Integrate 、 Deploy 、 Operation and maintenance knowledge . Then the purpose of this column , It's me for MLOps Your summary and thinking will continue to be shared . The first stage of this column will focus on the interpretation of this book , The list of these classic books is as follows ：
Summing up is not easy , Remember to connect three times with one button （ Pay attention to , Welcome to exchange and discuss with us ！）

1、Introducing MLOps

Through this book , Readers can understand MLOps Key concepts of , Help operate ML Model , And optimize the model over time .

2、Machine Learning Engineering

This book is one of the most complete books on the application of artificial intelligence , There are best practices and design patterns for building reliable machine learning solutions on a large scale .

3、MLOps: Operationalizing Data Science

Many analyses and ML The model has not entered the production stage , And the commentary of this book is introduced ML Four steps ： construct 、 management 、 Deploy ／ Integrate 、 monitor , Help readers import ML Model .

4、Building Machine Learning Powered Applications

This book is divided into four parts , How to plan ML application , How to set up ML Model , How to improve the model , And deployment and monitoring strategies , Help readers create ML Driver application .

5、Building Machine Learning Pipelines

Through this book , Readers can learn how to use TensorFlow Ecosystem , Realize the automation of machine learning pipeline .

6、Practical MLOps

This book takes readers to understand what is MLOps, And it's with DevOps The difference between , And explain how to operate ML Model , yes MLOps Introduction to tools and methods .

7、Beginning MLOps with MLFlow: Deploy Models in AWS SageMaker,Google Cloud,and Microsoft Azure

This book covers MLFlow And will be MLOps Methods of integrating into existing code , To track indicators 、 Parameters 、 Graphics and models .

8、What Is MLOps？

This book introduces data science –ML–AI Project life cycle , cover ML Model construction ,AI Monitoring of project life cycle .

9、Engineering MLOps

This book provides real cases , Share in-depth MLOps knowledge , To help readers write programs , Train strong and scalable ML Model , And construct ML The Conduit , To train and deploy the model .