MLOps Engineering Labs Recap // Part 2 // MLOps Coffee Sessions #31

5 Просмотры
This is a deep dive into the most recent MLOps Engineering Labs from the point of view of Team 3.

// Diagram Link:

--------------- ✌️Connect With Us ✌️ -------------
Join our slack community:
Follow us on Twitter: @mlopscommunity
Sign up for the next meetup:

Connect with Demetrios on LinkedIn:
Connect with Laszlo on LinkedIn
Connect with Artem on LinkedIn:
Connect with Paulo on LinkedIn:
Connect with Dimi on LinkedIn:

[00:00] Engineering Labs Recap Take 2 with Team 3!
[01:12] Laszlo Sranger Background
[02:05] Artum Background
[04:45] Dimi Background
[06:31] Paulo Background
[08:51] Can you give us an idea of what you came out with?
[09:12] "We came up with a decent product. We used the data sets that are usually used for benchmarks. That is YELP polarity review data sets. We tried to classify the review negative or positive." Paulo
[10:32] "About the backend side, the Streamlight is the facade system. We have the direction of what we want to achieve." Artum
[13:52] Do you feel that is a bad practice?
[14:11] "For the demo, it works perfectly. It's very visual and interactive. But for the real used case that triggers the pipelines is the best idea. We work with code, we don't work with models." Artum
[15:12] Can you walk us through the 50 lines of code on Streamlet?
[15:16] "It wasn't hard to do. From our workflow point of view, I knew Artem will deploy the model and provide me with API which is a decoupled architecture so the separation of the front end and the back end is easier." Dimi
[16:54] Managerial side of things
[19:00] "We don't want to work on what we feel uncomfortable with even if our intentions are good. This level of engineering and infrastructure cannot be solved in psychic learning." Laszlo
[20:36] What were your biggest things that you wanted to take out of this?
[20:42] "I have played with MLFlow around and I knew what is good for and what's not. I want to see which path I'm going to take from an architecture wise perspective." Dimi
[22:21] What else do you want to throw on top of this?
[22:31] "MLFlow is trying to be a kitchen sink from end to end solution. So let's try to monitor the incoming data that comes from a user." Dimi
[24:00] We could make this a bit more robust. It would be interesting to watch
God of War
Комментариев нет.