🏆 2021 MLOps Awards

Reading club had some great insights last week on Challenges in Deploying Machine Learning: a Survey of Case Studies.

Key takeaways:

It's imperative to catch data errors along the ML pipeline as they affect its overall quality. To do so, catch them with data validation routines in the ML pipeline to increase the observability on your ML system (e.g. using assertions, great expectations, etc)
A possible data drift in NLP can be new or seasonal text named-entities appearing in your text input. It's relevant to monitor the input tokens of your pipeline to be aware of new named-entities and re-train models accordingly

See you today at 9am PST/ 5pm BST for the weekly meetup! You can find the link to join here.

These 12 factors come from DevOps culture, more specifically from the people at Heroku in the end of the last decade. They are recognized as what makes apps easy to scale, or, conversely not easy to scale.

In this session Ivan takes us through how these 12 factors can be applied when building ML apps and which parts of them can be more tricky than others.

Towards the end of the session he gives us a run through of the open source project Kedro and demos its capabilities. He brings it all back round to show how Kedro fits into this 12 factor paradigm. 9.8 for sticking the landing.

Check out the episode below and tell me you dont love Kedro Viz.

James Lamb is a person I would consider the perfect blend to be an MLE. He has the data science depth to talk stats and random forests but after talking with him for an hour it was clear his heart lays in the software eng world.

My favorite part of the conversation was when we touched on how the scope and goals of his current team (4-5 data scientists) differs from the last job where he was enabling the work of 70 data scientists. What does this key difference entail? He goes deep into the bottlenecks and how he has solved them differently in each situation. It reminds me of the paper "You are not Google" and I think about how many of us working small teams could benefit from James' way of thinking.

Check out the convo below. This episode is also found wherever you get your podcasts, which acording to the analytics is mainly apple podcasts... so leave us a review!

As AI becomes more prominent across different industries, organizations are increasingly scrutinized for unfair ML algorithms and lacking clear explanations behind AI-driven decisions.

How do you avoid such risks and build trustworthy AI solutions? How do you guard against potential AI mishaps and build performant MLOps practices? In other words, how do you build responsible AI?

Check out the free O'Reilly eBook Model Performance Management with Explainable AI to learn about MPM, how each stage of ML can be improved with explainable AI, and how to build responsible AI.

One of the most watched videos we have done thus far was building a complete ML platform from scratch. Since it seems that people like it we are back for more of the from scratch series!

We're going to implement Machine learning and DevOps to train the Fashion MNIST model and achieve the best-desired accuracy. We're not going to manually retrain the model by manually adjusting hyperparameters or kernels. We will use DevOps to automate the process using tools like Github, Docker, and Jenkins. Keep yourself motivated and excited.🎉

Rohit will be our guest of honor to lead us through this adventure. Get prepped by reading his amazing blog post about MLOps automatization

Sub to our public calendar and click the button below to join the meetup. Put that in your pipeline and stream it!

All the MLOps podcasts the last year were special.I want to call out a few of them as I reflect on the year.

I wrote a blog about my most memorable list over the past year. It may be useful for you in case you were looking for something to do with all the free time you’ll have over the holidays.

2021 MLOps Podcast/Meetup Awards
TL;DR And the winners are
Best MLOps On Prem Convo - Chris Albon
Best Fundamentals Convo- Svet Penkov which led to #testing-ml
Most Underrated Convo - Alex Patry
Best Non Financial Advice - Sarah Catanzaro
Most Legendary Guest - D. Scully
Best Case Study - Leonard Aukea
Biggest Surprise - Ben Wilson
Easiest Interview of the Year - Erik Bernhardsson & Mike Del Balso
Most Entertaining - Jacopo Tagliabue