Five Steps for Deploying Machine Learning Models Into Production

Deploying machine learning models into production is by far, the #1 challenge our clients experience in becoming AI driven enterprise.
In this webinar we share with you some of our best practices that will greatly accelerate your team's ability to deploy models consistently and at a high velocity.
Specifically, you will learn: 3 options for deploying ML models ; How to do a rapid, iterative (e.g. Agile) release cycle in; the interface between data science, data engineering, and IT; Metrics from a successful project; and how to avoid 5 most common causes of deployment failures.
⏰ Time Stamps ⏰
00:41 - Introduction: The biggest challenge is taking machine learning models you've trained and putting them into production
02:00 - Three skill you will learn in this video: (1) Understanding why deploying machine learning models is hard; (2) Five steps for deploying machine learning models; and (3) How to avoid the three common causes of model deployment failures.
02:21 - Why is deploying machine learning models hard?
02:56 - Reason #1: Differences between how data scientists and date engineers produce software
05:16 - Reason #2: Infrastructure challenges for model deployment and data pipelines
07:03 - Reason #3: Your company isn't organized for machine learning
09:43 - Five steps for deploying machine learning models
09:49 - Step 1 - Build a minimally viable model
12:40 - Step 2 - Organize your jupyter notebooks for deployment
15:38 - Step 3 - Turn the "production notebook" into a Python application
17:08 - Step 4 - Deploy with simple infrastructure
18:15 - Step 5 - Create a process for iteratively improving
20:16 - Avoid three common cause of machine learning model deployment failure
20:27 - The machine learning model doesn't solve a real problem
21:21 - Too much time spent on data
23:00 - Too much time spent on modeling
23:55 - Conclusion

[@frankpierce7563]

Well done. Your comments at the end are 100% right. My experience is analytics team’s are paralyzed seeking a mythical perfect state and think success is solving a math problem, not solving a business a problem.

[@shashankhegdek8623]

This was very nice and detailed 👍

[@Alice8000]

This is great. Thanks a lot. Any advice for someone who wants to do the same as you? Helping companies implement

[@TURALOWEN]

Dont use R? Dont agree.

[@frankpierce7563]

Well done. Your comments at the end are 100% right. My experience is analytics team’s are paralyzed seeking a mythical perfect state and think success is solving a math problem, not solving a business a problem.