Completed Andrew ng's ML course and ISLR, what's next?
1. I completed Andrew Ng's Machine learning course on Coursera. It was maths heavy (with cost functions etc) and I am not sure if I remember all the equations though, is it necessary to memorize all those? 2. I have also completed ISLR. I am in a non-tech role and not an engineer. Can anyone let me know what should I take up next to become an ML engineer? Should I learn data structures or take up a deep learning course or some other ML course?
Leetcode
If you're super new to machine learning then you really need to do some projects. I could say grab a public dataset and build an interesting predictive models but you'll get stuck in defining interesting. Join kaggle and do few of the old competitions like housing and Titanic and digits.https://www.kaggle.com/competitions?sortBy=grouped&group=general&page=1&pageSize=20&category=gettingStarted
In no way is this meant to condescend, but if you think Ng's class is math heavy, you may want to study some linear algebra, probability, and calculus. That course is foundational, and is really just dipping your feet into the math and theory. Yes, study deep learning. Yes, other ML classes. Udacity has good courses that provide a more high-level, how-to-use ML type of instruction. Look at kernels on kaggle. Read research - the papers from groups like deepmind are very interesting and easily digestible. Try to think about how you would replicate the research, and even try to replicate it. Andrew Ng's class on Coursera is not even the tip of the iceberg. I know he says a lot about "you now know more than most people in industry" but it's simply not true. Maybe it was true when he recorded. You've got quite a ways to go to get caught up with the current state of ML.
Thanks a lot, this is insightful. I'll check out Udacity, kaggle, and deepmind for some more ML and DL courses. Are there any websites or blogs or GitHub accounts that I can use to stay up to date in the industry? I am a frog and I really don't know what the depth of the sea is.
Check out the machine learning subreddit. There are posts there every day with the state of the science research and tools. I'm sure that it is a small fraction of everything out there, but it's a good place to start.
I will be honest with you - if you have a non-tech background and aren't an engineer, it is going to be quite difficult to become an ML engineer. There is a reason most ML positions are filled by PhD/MS in ML. It is also the flavor of the month when it comes to CS specializations so the competition is only getting crazier and crazier with the current explosion of CS new grads. The only path I see right now is for you to continue studying on your own and do ML projects. Eventually, when you become knowledgeable with a good set of projects, you can apply to hundreds of companies and work your way up. It will be quite tough to land the first ML position though...
Thanks for letting me know. I understand it's gonna be tough but I just want to make sure I am on the right track and in line with what companies look for. I am up for doing a CS specialization if that means more returns in the longer run.
If you don't have at least a MSCS related to ML it'll.be very tough. Like others said ML industry required quite a but of schooling to get.inteeviews / opportunities (though it isn't necessarily needed it just seems to be the bar). Good luck!
Also start OMSCS from Georgia Tech or OMSA . It cost around 7 k and for A it is 11 K . Start Kaggle . Build your portfolio on GitHub . Start applying or try to get some part time gigs .
How much of an effort is it? I heard the drop off rate is very high because it gets super hectic along with a full time job.
It is a tremendous amount of effort. I am currently taking the ML class, which is 100% refresher for me. It is still taking 15 hrs+ per week. Reinforcement learning was 20-30 hrs/week. Not all classes are this demanding, but many are, including most of the interesting ones.
For most top companies, ml eng has a same coding bar as sde. Please spend some time leetcoding. Good luck.
The answer is still Leetcode and a bit of ML. I have given ML rounds at ANG (of FANG), Apple, Uber, Lyft etc. Except at Apple, Netflix, Lyft (where the position was technically an applied scientist), the rest of the ML rounds were very easy for me (I hold a PhD in ML). Leetcode rounds is where the real deal is. I assume for you, ML rounds will be of medium difficulty. You should study the theory of five - six basic ML algorithms. That’s all 90% people will ask, and some other concepts like overfitting, data imbalance etc.
Thanks, this is helpful. Do you think the basics which I covered in Andrew Ng's course and ISLR would suffice to get through ml rounds? Also, for coding rounds were you asked data structures or data manipulation coding (like numpy pandas and take home challenges)?
Read some math based tutorials (even better - books) for logistic/linear regression, kNN, random forest, decision tree, bias and variance, data imbalance, overfitting, maximum likelihood. Congratulations, you have passed 80% of ML interviews. I did one take home. Rest were phone rounds which are a mixture of Leetcode/ML/resume.
I feel like the average ml engineer these days gets away with libraries and doesn’t really use any math at all!
Because XGBoost tends to be the standardized (initial) answer for all the problems 😂
Lol Walmart, this is so true 😂
Kaggle