Ask HN: What is the right way to self-learn Machine learning?

[prando]

I have an electrical engineering background and the recent developments of ML, Deep Learning & AI is very interesting and I hoped to self-learn this. I signed up for Udacity's Machine Learning Nano Degree program and found it to be at a much higher level than what I had hoped. They usually drop an algorithm and talk about using it to solve the problem, rather than getting into the rudiments of it. Could you please suggest some methods - books, videos and general techniques to master the theory as well as the practical aspects of ML?

[minimaxir]

Learn the fundamentals of statistics first; then worry about the fundamentals of machine learning.

[prando → minimaxir]

I'd greatly appreciate any good recommendations for text/ videos on statistics.

[uptownfunk → prando]

A good understanding of Stats IMO depends heavily on a solid basis in Probability (e.g. Pitman, Feller, Ross) which itself relies on a solid basis in Mathematical Analysis (e.g. Rudin). The hard work on the fundamentals now will really prepare you to do some serious damage later in ML. And if you really want to be prepared I would also read some elementary theory in Measure theory as well.

[sharmi → uptownfunk]

Seriously, this is different from coursework. You need some really strong motivation to cover all this through self study and stay on track. I believe the user should start simpler and build from there, both above and below.

Unfortunately, I don't have a well defined plan to recommend to the OP but any input from someone who has managed to learn the theory on their own can shed some light on how it is done. Especially on how to stay on track with a full time job on hand.

[uptownfunk]

Some good resources... probably many more.

deep learning book http://www.deeplearningbook.org/ for theory. Cs231 http://cs231n.github.io/ http://yerevann.com/a-guide-to-deep-learning/

[indigentmartian]

Andrew Ng's Coursera course simply titled "Machine Learning" is good - it addresses the mathematics of fundamental algorithms and concepts while giving practical examples and applications: https://www.coursera.org/learn/machine-learning

Regarding books, there are many very high quality textbooks available (legitimately) for free online:

Introduction to Statistical Learning (James et al., 2014) http://www-bcf.usc.edu/~gareth/ISL/

the above book shares some authors with the denser and more in-depth/advanced

The Elements of Statistical Learning (Hastie et al., 2009) http://statweb.stanford.edu/~tibs/ElemStatLearn/

Information Theory: Inference & Learning Algorithms (MacKay, 2003) http://www.inference.phy.cam.ac.uk/itila/p0.html

Bayesian Reasoning & Machine Learning (Barber, 2012) http://web4.cs.ucl.ac.uk/staff/D.Barber/pmwiki/pmwiki.php?n=...

Deep Learning (Goodfellowet al., 2016) http://www.deeplearningbook.org/

Reinforcement Learning: An Introduction (Sutton & Barto, 1998) http://webdocs.cs.ualberta.ca/~sutton/book/ebook/the-book.ht...

^^ the above books are used on many graduate courses in machine learning and are varied in their approach and readability, but go deep into the fundamentals and theory of machine learning. Most contain primers on the relevant maths, too, so you can either use these to brush up on what you already know or as a starting point look for more relevant maths materials.

If you want more practical books/courses, more machine-learning focussed data science books can be helpful. For trying out what you've learned, Kaggle is great for providing data sets and problems.

[pizza]

Kevin Murphy - Machine Learning: A Probabilistic Perspective