Comment by RobinL | Hacker News Reader

Comment by RobinL | Hacker News Reader

Parent Full thread

I've been working as a data scientist for a few years now, primarily self taught using the wealth of high quality videos and other materials online.

I recently worked as a teaching assistant on an expensive and well-regarded data science bootcamp course. I did this primarily because I was curious what classroom-based training in data science was like.

I found it much less useful than online materials. It seemed like the primary benefit was just that once you've stumped up several thousand pounds, you're motivated to actually show up and concentrate. And you also get to meet some people with a common interest.

But the quality of teaching fell well below the standard online (not surprisingly, when online you have access to lectures by superstars like Hadley Wickham, Peter Norvig and Andrew Ng, with lectures that have been carefully recorded).

And on top of that, it's not a great use of time. A half hour to commute to get to the lecture. Then the lectures being in 'real time'. Online, I find myself running video lectures at 1.5 or 2x speed for large portions of the material - the 'filler' - and then having to watch, re-watch the hard concepts several times at normal speed and pause just to think.

I found that at the end of the classroom course, I had developed an awareness of the existence of concepts and techniques, but not really an understanding of them.

Can you talk about finding a job in data science despite being self taught?

+1'd. Also, whether working in the field has satisfied you as you had hoped.

I've been a government analyst for around 10 years, starting my career as an economist (after graduating with an economics degree). This meant I was working with numbers a lot, and I found that my strengths were mainly in statistics and working with complex datasets rather than e.g. macroeconomic analysis.

This meant I had started to build up examples of projects which were getting close to 'data science' (the term, I think, is vague). Custom interactive data vis in d3.js, forecasting and regression modelling etc.

I also started quite a large web scraping project in my own time, which led to me needing to develop a much stronger programming skillset, and gave me some more tangible examples of having done 'data sciency' stuff.

When data science took off in UK gvt a few years ago, that meant I was well positioned to make the switch. No-one had much experience at that point, so it wasn't too hard to get a first role. I think generally, there's such strong demand for data scientists that there's a mismatch between the job spec (which look scary), and the candidates (who aren't the data science unicorns the job ads ask for). So with a few projects under your belt, it worth having a crack at applying for stuff.

In terms of being satisfied: It helps working for government because we're trying to use data to make better decisions that help people. There's definitely too many buzz words and fluff, and it's difficult to help non-data scientists understand what's hard (all the data engineering and modelling leading up to delivery) and what's not (a flashy data vis) - so it's hard to convince non-data scientists to allocate resource to the best areas. The flip side is that it's really clear there is a great opportunity to change how we do analysis for the better. So overall it's pretty good.

I've taught professionally for some time. I agree that the quality of many in-person courses is worse than the best online classes. Many popular books are also junk.

It's very hard to judge the quality of teaching, because nearly all students of a non-catastrophic class will give good ratings. Few students will take the same material from more than one teacher so they can make comparisons.

Reply on news.ycombinator.com