Guys boarding the #MachineLearning train by learning Python and a bunch of libraries and just plain old stacking and ensembling and stuff - is not going to get you anywhere.
A thread on Machine Learning.
A thread on Machine Learning.
The first and foremost important thing to do if you really want to do meaningful stuff that can even be remotely considered Machine Learning with an impact - you gotta learn the math that makes ML happen.
Must know these math topics:
1. Probability Statistics
2. Linear Algebra
3. Multivariate Calculus (and throw in some Stochastic Calculus too)
Get the basics right.
1. Probability Statistics
2. Linear Algebra
3. Multivariate Calculus (and throw in some Stochastic Calculus too)
Get the basics right.
Sure you can throw in gazillions of processor power, CPU, and quantum computing stuff after large data sets and use pre-built libraries and come up with solutions. That's what Machine Learning engineers do. But if you want to understand research papers meaningfully, math is must.
Especially in the world of quant finance, ML is getting all the hype. Two Sigma, MAN AHL have all tried advancing significant ML based investing/trading and all that - but ML is largely unnecessary for a retail trader. If anything, it's an overhead.
If you want to learn ML Engineering with proper Data Science best practices imbibed in you, go through fast.ai bootcamp - all their courses. Stanford AndrewNG course is pretty old and while it covers ML, not as relatable/accessible to layman (esp with octave lang)
Check out fast.ai - these guys get you read with the basics (math, data science and ML) in the most practical manner to do ML engg from get go, practicing with Kaggle platform problems.
Check out - ods.ai (Open data science bootcamp) - with mlcourse.ai and dlcourse.ai - getting you prepped in Machine Learning and Deep Learning. Again, the most practical bootcamps.
You follow the book recommendations they give in those two bootcamps for the math, and the other required knowledge preliminary to good ML engg practice. You won't need anything else.
If you want to understand current research, build state of the art systems, and hardware isn't a problem for you, you would definitely need to dive headfirst into the math first and master calculus, probability, linear algebra, stochastics, numerical analysis, real analysis, etc.
You want to learn Reinforcement Learning, start with David Silver's course - youtube.com/playlist?list=…
You want to learn ML and DL the way led by research - in labs like MILA/UToronto/McGill. Go through SummerSchool videos - leading experts teaching you.
videolectures.net/Top/Data_Scien…
Exhaustive list available in this website. Pick one, see it through.
videolectures.net/Top/Data_Scien…
Exhaustive list available in this website. Pick one, see it through.
Set up a @kaggle account and start solving problems. Go from the very basic, pick up skills, solve 10-20 problems, start contesting. Beyond a point, to actually win contests, you need to study research papers and implement stuff - which you will get to, step by step.
Alas, don't mix up with Machine Learning with keeping on learning. Atleast with respect to ML Engg and research, you learn by doing. Clean as much data as possible. Solve as many problems as possible. Go through as many datasets as possible.
With respect to Machine learning - once you pick it up, it's like Indian taxation system. You can't lag behind current research. Pick an area within the field, specialize and gain expertise, keep yourself updated, and have friends who specialize in other niches within ML.
