A Review of Signal Data Science

I took part in the second signal data science cohort earlier this year, and since I found out about Signal through a slatestarcodex post a few months back (it was also covered here on less wrong), I thought it would be good to return the favor and write a review of the program.

The tl;dr version:

Going to Signal was a really good decision. I had been doing teaching work and some web development consulting previous to the program to make ends meet, and now I have a job offer as a senior machine learning researcher1. The time I spent at signal was definitely necessary for me to get this job offer, and another very attractive data science job offer that is my “second choice” job. I haven’t paid anything to signal, but I will have to pay them a fraction of my salary for the next year, capped at 10% and a maximum payment of $25k.

The longer version:

Obviously a ~12 week curriculum is not going to be a magic pill that turns a nontechnical, averagely intelligent person into a super-genius with job offers from Google and Facebook. In order to benefit from Signal, you should already be somewhat above average in terms of intelligence and intellectual curiosity. If you have never programmed and/​or never studied mathematics beyond high school2 , you will probably not benefit from Signal in my opinion. Also, if you don’t already understand statistics and probability to a good degree, they will not have time to teach you. What they will do is teach you how to be really good with R, make you do some practical machine learning and learn some SQL, all of which are hugely important for passing data science job interviews. As a bonus, you may be lucky enough (as I was) to explore more advanced machine learning techniques with other program participants or alumni and build some experience for yourself as a machine learning hacker.

As stated above, you don’t pay anything up front, and cheap accommodation is available. If you are in a situation similar to mine, not paying up front is a huge bonus. The salary fraction is comparatively small, too, and it only lasts for one year. I almost feel like I am underpaying them.

This critical comment by fluttershy almost put me off, and I’m glad it didn’t. The program is not exactly “self-directed”—there is a daily schedule and a clear path to work through, though they are flexible about it. Admittedly there isn’t a constant feed of staff time for your every whim—ideally there would be 10-20 Jonahs, one per student; there’s no way to offer that kind of service at a reasonable price. Communication between staff and students seemed to be very good, and key aspects of the program were well organised. So don’t let perfect be the enemy of good: what you’re getting is an excellent focused training program to learn R and some basic machine learning, and that’s what you need to progress to the next stage of your career.

Our TA for the cohort, Andrew Ho, worked tirelessly to make sure our needs were met, both academically and in terms of running the house. Jonah was extremely helpful when you needed to debug something or clarify a misunderstanding. His lectures on selected topics were excellent. Robert’s Saturday sessions on interview technique were good, though I felt that over time they became less valuable as some people got more out of interview practice than others.

I am still in touch with some people I met on my cohort, even though I had to leave the country, I consider them pals and we keep in touch about how our job searches are going. People have offered to recommend me to companies as a result of Signal. As a networking push, going to Signal is certainly a good move.

Highly recommended for smart people who need a helping hand to launch a technical career in data science.


1: I haven’t signed the contract yet as my new boss is on holiday, but I fully intend to follow up when that process completes (or not). Watch this space.

2: or equivalent—if you can do mathematics such as matrix algebra, know what the normal distribution is, understand basic probability theory such as how to calculate the expected value of a dice roll, etc, you are probably fine.