Dr. Julia Silge | RStudio Voices | RStudio
Julia Silge recently sat down with Michael Demsko Jr for an interview, the first in a new Voices of RStudio PBC series. In this excerpt, Julia discusses where she sees the most value created in the data science lifecycle--and it's not advanced machine learning models. Read the full interview at https://blog.rstudio.com/tags/rstudio-voices/
image: thumbnail.jpg
Transcript#
This transcript was generated automatically and may contain errors.
Monica Rogati, who has moved around a little bit, but she's like an important data science leader and writer, has like the data science hierarchy of needs, or like the AI, like hierarchy of needs. Like at the bottom are things like, like logging and like collecting data, and then you like move up and like things like at the middle are like analytics. And things like machine learning are like way at the top, like way at the top of this pyramid.
And in terms of like value that people get, you get a lot of value from analytics, from like counting things. You get a lot of value from like your first models you train, you know, like simple models, linear models, you know. And it's like, you have to be pretty mature. You have to be pretty up to your hierarchy of needs to be able to the point where you're like, I need a fancy machine learning model that's like learning nonlinear relationships or taking advantage of complicated math.
I wouldn't say I'm motivated to work on machine learning because it's the thing that like offers the most value, or that it's the thing that like solves the most problems. Because I think like the thing that solves the most problems is like making data accessible to people in their workplaces, and that they have the tools to be able to like do basic analytics or train that first model. Like you get the biggest wins at that, at those lower levels.
the thing that solves the most problems is like making data accessible to people in their workplaces, and that they have the tools to be able to like do basic analytics or train that first model. Like you get the biggest wins at that, at those lower levels.
I do think the problems like like machine learning tooling is interesting because it is complex, and people enter into it with different levels of experience, and being able to build tools to make that process more streamlined is like is an interesting challenge.
