Git For Data

Learn With A Friend

Welp this is one of those new initiatives I was talking about, community member and Coffee Session veteran Charlie You said he would take the reins and help organize a paper reading group for the community. The idea here is that we all can bounce ideas off each other after reading specific papers that pertain to MLOps in some way. We are also thinking that depending on the paper we can invite the author to come to the group meeting to give us more insight into the paper or book.

Obviously, this is the first iteration and we will be improving and evolving but so far it's shaping up to be something special with 50 people already signed up and frothing at the mouth to sink their teeth into some good ol' academia.

What's that I hear? You want to participate also?

No problemo, fill out the form below and we will put you on the list! Right now we are figuring out when the first meeting will be and what the first paper we will read is going to be. If you have any questions or comments feel free to reach out to myself or Charlie for more details.

Cliché Level = Cringeworthy

Luke Feeney and his cofounder Gavin Mendel-Gleason were able to do a brilliant job of walking the line between conversation and presentation last meetup. The topic? 'Git for Data'. How many times have we heard an MLOps tool that claims they do that? Welp, Luke managed to show us all the open-source options out there and breakdown each of them, putting the 12 that he coagulated onto this list into 4 buckets:

Versioning Layer: Not a database or server
Data Catalogs: And other tools of that type
Data Pipeline Versioning: Reproduce the entire pipeline
Version Controlled Database: Append only databases with version control features

+ After a look at each of the tools on the market right now, Gavin explained the motivations and engineering work that went into their open-sourced baby TermnibusDB. Can you guess what their tag line is?

+ Extra credit for someone who can tell me how many times we said Git or Get in the hour-long meetup. I'm guessing it was over a thousand times.

+Check out the video here and the podcast here.

To Hell And Back

It wasnt an easy task. The teams had a low chance of completion, only 50% made it through. For those that did though, glory and praise awaited them! They are able to call themselves champions... of this small little cohort that lasted 2 months.

Okay okay okay. If you are sick of hearing about the Engineering Labs already, I'm sorry. To me it's one of the coolest things that we can do as a community. So for this coffee session, I spoke with Team 1 about what the engineering labs cohort was like for them. We dug in deep not only to the problems they faced with MLflow and Pytorch but also their experience of being thrown on a team with little to no guidance. Basically, we threw them off the deep end when they barely knew how to swim!

Anyway, the whole team (John, Michel, Varuna, and Alexey) were able to walk us through everything from the design decisions they made to how they iterated on the original plan. Next week we will be talking to team 3 to hear how their experience was and what they chose to create.

Personally, I love seeing people from Europe the Americas, and Asia all coming together and making magic happen and that's why I can't be overly excited about this initiative. If you want to take part in the next cohort join the #engeneering-labs channel in slack to stay up to date!

Responsibility

The Theme: Depending on who you talk to you may hear that training and deploying ML models have become relatively fast and cheap. With the rise of ML use cases, more companies and practitioners face the challenge of building “Responsible AI.” So what does that even mean anyway?

One of the barriers these companies encounter is increasing transparency across the entire AI lifecycle to not only better understand predictions, but also to find problem drivers. Explainability, governance, ethics, responsible, these are all buzzwords under the gigantic MLOps buzzword umbrella. But does it actually account for something? If you have been round the community enough you know my stance, YES! We can never have enough of these kinds of conversations no matter where you sit in the food chain!

Meetup: We will discuss the meaning of explainability and responsibility within ML. And since we are talking to Krishna Grade who has been round the block a few times I'll get him to share examples from real-world scenarios especially those times when things weren't all gravy.

Bio: Krishna is the co-founder and CEO of Fiddler, an Explainable AI Monitoring company that helps address problems regarding bias, fairness and transparency in AI. Prior to founding Fiddler, Gade led the team that built Facebook’s explainability feature ‘Why am I seeing this?’. He’s an entrepreneur with a technical background with experience creating scalable platforms and expertise in converting data into intelligence. Having held senior engineering leadership roles at Facebook, Pinterest, Twitter and Microsoft, he’s seen the effects that bias has on AI and machine learning decision making processes, and with Fiddler, his goal is to enable enterprises across the globe solve this problem.

+ For some prerequisite reading to get you in the mood check out this article that one of Krishna's colleagues Henry wrote for the MLOps Community Medium page on Model Performance Management Done Right.

+ See you at 5pm GMT / 9am PST tomorrow, Wednesday by clicking the link below.