<style> .reveal { font-size: 28px; } .container{ display: flex; } .col{ flex: 1; } .reveal section img { background:none; border:none; box-shadow:none; } </style> # MLOps in 2021 Alexander Guschin ML Engineer at Iterative.ai --- <!--  --> Bio: - ML Engineer at Iterative.ai - Teaching at Coursera, Data Mining In Action - Earlier - Top-5 on Kaggle - Tech lead at Mechanica.ai - Head of DS team in Yandex.Taxi --- <!-- - Q: Why MLOps? - A: Going to production and staying there up-to-date is hard. - 87% of data science projects never make it into production, [Gartner](https://venturebeat.com/2019/07/19/why-do-87-of-data-science-projects-never-make-it-into-production/) --> <!-- ---  https://www.gartner.com/en/articles/it-budgets-are-growing-here-s-where-the-money-s-going --> <!-- --- --> What's about MLOps?  [MLOps tools by year of launch and color-coded by category by Chip Huyen](https://huyenchip.com/2020/12/30/mlops-v2.html) ---  [MLOps Platform timeline by Thoughtworks](https://www.thoughtworks.com/content/dam/thoughtworks/documents/whitepaper/tw_whitepaper_guide_to_evaluating_mlops_platforms_2021.pdf) ---  [MLOps software with poles of specialization in lifecycle stages and all-in-one platforms gravitating towards poles (or middle if no particular pole), by Thoughtworks](https://www.thoughtworks.com/content/dam/thoughtworks/documents/whitepaper/tw_whitepaper_guide_to_evaluating_mlops_platforms_2021.pdf) --- [2021 MLOps trends from VC investor](https://mattturck.com/data2021/): - :bank: Feature stores (multiple startups raising rounds) - :ocean: Rise of ModelOps - :panda_face: Chinese AI stack grows [including MLOps](https://huyenchip.com/2020/12/30/mlops-v2.html) --- Some interesting spotlights: - [Using AntiPatterns to avoid MLOps Mistakes](https://arxiv.org/abs/2107.00079) paper - [Building Continuous Integration Services for Machine Learning](http://pages.cs.wisc.edu/~wentaowu/papers/kdd20-ci-for-ml.pdf) paper - [Adoption and Effects of Software Engineering Best Practices in Machine Learning](https://arxiv.org/pdf/2007.14130.pdf) paper - [Applying ML by Eugene Yan](https://applyingml.com) - [...and bunch of other links](https://mlops.community/mlops-2021-year-in-review/) <!--  --> --- ## :christmas_tree: [Iterative 2022 Predictions: MLOps tools in focus](https://vmblog.com/archive/2021/12/01/iterative-2022-predictions-mlops-tools-in-focus.aspx#.YbxVxC8RrD2) - :chart_with_upwards_trend: More tools to manage unstructured data in ML projects - :bank: MLOps solutions focus on collaboration and enterprise-grade features - :hammer_and_wrench: Specialization over full-stack MLOps --- - :male-technologist: Iterative is working towards MLOps trends: - DVC - Studio - CML - :tada: More products are coming in 2022: - [MLEM](https://mlem.ai) - VSCode extension for DVC - and even more :eyes: :confetti_ball: <!-- Iterative.ai enables data science teams to build models faster and collaborate better with data-centric machine learning tools based on Git best practices. --> <!-- Use Iterative --> ---  --- MLOps... Hmm, sounds interesting! How to start? Well, there is something... ... --- MLOps... Hmm, sounds interesting! How to start? Well, there is something... :tophat: The Hat Game!