Ai+ Training

While developing MLOps in an organization, one often encounters the “chicken or the egg” paradox:  On one hand, a certain abstraction, workflow, or tool is hypothesized to be a good fit. On the other hand,  there are often considerable integration efforts even before feasibility testing can commence. 

In other words, you cannot know if the tool works unless you “buy in”.

The common solution is to “build” internal implementations which optimize for integration ease.  Most of these “interim” solutions take a life on their own and will only be replaced once (if)  they outlive their usefulness or another abstraction is adopted.  

In this webinar, we will simulate this low-integration-cost process by building a feature-rich Feature Store for offline use using the ClearML Open Architecture Stack. 

We will use this surprisingly simple process to understand the following:

• How to iteratively add MLOps onto functioning code

• How DataOps and MLOps can be used during R&D

• How to involve Data Scientists in the feasibility testing stage

• How ClearML core components provide a robust infrastructure to build upon

Finally, we will compare our own DIY Feature Store with off-the-shelf solutions and understand the gap between “in-house” and “store-bought”.

PS: This webinar expands upon the ClearSHOW episodes S02E04-S02E08

Prior knowledge needed: 

• “What is a feature store” (2 min video)

• Basic pandas (we will be using a few lines from this)

• Jupyter notebooks

Environment requirements (in case you want to follow along):  

• ClearML  installed (Free or self-hosted)

• Jupyter server or colab notebook. 

• clearml-agent running (for automation)