How to learn about a new domain as a data scientist

parachute.png

“The best thing about being a statistician is that you get to play in everyone’s backyard.”

John Tukey

Professional life as a data scientist often means changing domains

As a data scientist, your skill set is powerful because it is very generally applicable. ML models and statistics give us ways to use data to understand the world and make decisions a little better, regardless of the domain

Despite this generality, YOU NEED TO UNDERSTAND THE DOMAIN TO DO A GOOD JOB. I’ve developed a method that I use to do this

Over the last ten years, I’ve had a lot of jobs, and a lot of teams. I’ve been lucky! I’ve been able to work with so many interesting people, solving lots of interesting problems . even if you don’t yet have my ~level of jaded frustration~ wealth of experience, maybe you:

• got a new job. congratulations!
• changed to a new team
• are a consultant

this post is meant to help you X. it is idiosyncratic, and reflects what has worked for me (it’s what I used most recently to onboard at google)

it has three major components, performed roughly in order:

• talk
• actually access the data and do case studies, then see if your case intuition pans out
• look at metrics

Talk: Get to know the team

The goal: Meet everyone. understand what they do, and how they concieve of success for themselves/for the group, how you can help.

Talk talk talk talk talk!! you can’t overdo this

Each of these includes themes that I try and piece together, which are included in bold. Each theme includes some specific questions you might ask. I try to cover all the themes, but I don’t necessarily ask each question (and I may ask others besides).

There are three big things:

• The Org
• The Data
• The Projects

I’ve written this for organizations which use the PM-Analyst-Eng structure. your org may have only a subset, or other ones besides, or something else

DS/Data team

Titles like: DS, Analyst, AE, Product Analyst

Org

• How is the org structured? Are DS centralized or embedded?
• How do you work with stakeholders/biz owners?
• How do you work with tech teams (non-biz)
• What are the goals and north star?

Data

• How is the data generated, and how do you access it? What are the data sources? Do analysts access a DW, or prod data, or…?
• What metrics are implemented?

Projects

• How do you plan and track projects? Roadmap? Sprints?
• How do you track deployed solutions? repo? model/system observability?
• How do you understand key levers? exp, ci, etc
• Do you run experiments? How? how many, how often, how long do they take? how do you think about power? how are decisions made? who owns the exp engine?

stakeholders/partners/pms

Titles like: Product Manager, Marketer, Designer

sentence

• how do you keep an eye on the north star? do you take actions that move it directly, or a proxy metric? do you have goals for it?
• what levers do we have? what key decisions do we need to make?
• Main metrics? key subgroups? trends?
• biggest oppties, past+future?
• what are the key entities? key interactions? main dashes?
• roadmap? where is it, where does it come from, how are priorities set?

Engineering/ops

Titles like: Software Engineer, Release Engineer, Eng manager

sentence

• what is the prod source?
• how are key levers controlled?
• how do experiments work?
• release cycle length?
• your key decisions, esp recurring?
• your metric goals? automation oppties?
• how do you support DS? ML Eng?

Wrap up exercise

You should have some examples of: entities interactions value etc

Build intuition from the bottom up: case studies

The goal

1

2

3

Wrap up exercise

Build intuition from the top down: metric studies

The goal

1

2

3

Wrap up exercise

Other sources

https://bpb-us-w2.wpmucdn.com/sites.umassd.edu/dist/8/754/files/2019/01/Deconstructing-Statistical-Questions.pdf

chris chatfield