The Hard Problems™️ of Data Observability w/ Kevin Hu of Metaplane
The Analytics Engineering Podcast
00:00
I'm Just One Step Out of This Part of the World
We don't have a very large mapping from tables to the sematic annotation of those tables. The way we went about that is honestly scraping every single table off of the web and using the column name as the ground truth label, so to speak. I think that is an excellent example where you kind of boot strap a lot of these computer vision models. We have canonical data sets like image net but the same thing isn't true for arbitrary, tabular data. So i think my first response to that conundrum is that we just don't havethe training data.
Transcript
Play full episode
