I am a 2nd year PhD student in physics. Tenure-track positions are highly competitive and I do not love research enough to pursue it as a life career. Since I like programming and playing with data, I want to have a job as a data scientist after finishing my degree. I read some success stories of people who got degrees in Physics but works as data scientists but the people are from top universities like UC Berkely, Stanford, etc... So my question is how doable it is for someone who only gets Physics degree from the low-rank university to find a job as a data scientist. What is the plan for the next years when I am still in my PhD program? What should I learn? How should I have real projects and internships to work on? Will working unpaid in a research lab about data analysis in my current university help?
Yes, you absolutely can go from a Physics PhD to a data science career.
The three major routes I've seen have been:
- Apply to a program like the Insight Data Science Fellows (there are many like this), where they take students with strong quantitative backgrounds and build up some of their more industry-relevant skills, then place them in jobs. These can be quite competitive, and my impression is that students who get placed in these fellowships have already done significant work on "side projects" in data science - i.e. you create your own research topic, and find out something interesting. [Also, since they are competitive, I suspect students from high-profile universities have an advantage.]
- Find an internship at a local company; use this to bootstrap your way into industry (or just go work there if you like it!). Again, usually before you get an internship, you usually need to show some interest, working on a more closely data-science-relevant side project, or providing a solution in a Kaggle competition.
- Personal connections. Keep an eye on graduating students now, and see what they do! Many companies need coders with strong quantitative skills, and might offer referral bonuses - someone who graduated a few years before I did reached out to me at one point because of this.
Since you are just starting out, you also have the important option to make your PhD project more closely aligned with interesting data science ideas. It is possible to do both physics and data science - for instance, if I look at the list of sessions at the 2017 APS March Meeting, I see three or four with "machine learning" in the title alone. Of course, this depends on an advisor who is willing to do this and able to teach you relevant things!
However, it is still important to remember that a Physics PhD is a long time commitment, and you have to choose an advisor and a project you will be happy with in the mean time - not just what is going to be popular in industry. (After all, in 3-4 years, the market for data scientists may not be nearly so good.)
I did exactly this (Physics PhD to data science). I didn't do any 'specific projects' but did some self teaching.
If you want to do help yourself you could learn:
- Brush up on Linear algebra.
- Good knowledge of one high-level programming language (Python, R, etc)
- Awareness of Machine Learning algorithms.
I was already competent in Python and did some basic Machine learning (i.e., regression and basic image classification). I also read the first half of the book:
'Hands-On Machine Learning with Scikit-Learn and TensorFlow' by A. Geron. (I have no affliation to this book or author).
The most important thing is to concentrate on getting your PhD! A good PhD will get you a job in this field rather than the basic understanding you could gain yourself in your free time. I did all my learning while working at a different job for a few months after my PhD.
Following this, I then approached some Data Science jobs and was honest: I have a strong numerate background, but have very little knowledge about data science but want to learn. Several companies were very happy for me to 'train up' because of the potential someone with a PhD has! Particularly, as a physics PhD teaches you great research and problem solving skills.