I’ve been studying data science, math, and machine learning for about 1 year now, and have put about 500-1000 hours in (large range since I also spend a lot of time studying for my role as a resident physician and measure hours in the same tool). You don’t just need to learn the math and algorithms, you need to learn multiple entirely new skillsets; but, start with the math and algorithms 🙂
- If you can do basic python (numpy, pandas, loops, if/else, build a class with methods/attributes) then skip computer science and come back to it at a later time otherwise do it first.
- Start with Ng courses they are very good and cover everything you need. Expectation is to get an initial grasp of a lot of different things. This doesn’t make you an ML engineer, it gets you started. A lot of this stuff takes many repetitions and projects to understand well. Using Octave in the first course is kind of weird, but it’s not a big deal and the language does show matrices cleanly which is good for learning linear algebra.
- Math is a slow burn, linear algebra is a must, but the rest of it depends on your life goals. If you really want to know math, then do a proofs book (Chartrand) along w LA. Get a Chegg subscription so you have answers to all the questions in the chapters of whatever books you use.
Finding ways to apply what you learn and building adjunct skills is essential.
Slowly work on
- Effective pandas (Harrison)
- Learn SQL (DeBarros book + CodeSignal practice problems)
- Learn regular expressions (regex101.com questions are good)
- Read book on how to visualize data
- Learn matplotlib. Not a lot of great resources on this, I literally just remade all the graphs from the book “Better Data Visualization.” I’ll say, it was a STRUGGLE – but now I got it 🙂
- Sign up for AWS and Google Cloud Services and learn how their services work. There are some good course courses I’ve been looking at to get better at this myself.
- Listen to a bunch of ML/DS podcasts
Life goals really matter here. Without background you’re in for a long haul here. I’m about 1 year in, and have grown tremendously, but I still have so much to learn. I’m expecting that it’ll take about 3-5 years of constant work on this (probably about 2500 hours) to be competent. My definition of competent is: able to develop and deploy multiple different model types along with evaluation, production monitoring, and iteration.
Studying online courses for hours per day can be hard, it’s very active engaged learning. I’ve found 6 hours on days off and 2-4 hours on work days is a nice middle ground. I usually read 2 hours, work on math for 2 hours, work on ML courses for 2 hours. I’ve had a couple of nice work related data science projects that I fully commit time to when they come up. I always apply methods to my own datasets and build my own implementations alongside the coursework.
8 hour days were not working out well for me from a balance/guilt perspective. I’ve done this will being a resident physician working many 80 hour weeks, so you can definitely fit this in with the rest of your life. The caveat is, it really must be a priority. I think it’s actually a great idea to start slow and tickle away at it for a few months. Then, if you like it, you can ramp up.