Ask HN: Resources to teach myself Statistics?
All are welcome, what seems to be in short supply is stuff at either end, the fundamentals and what a master student might encounter.
To briefly explain why I'm asking this, and why the answers may be of value to others in this community:
I'm trying to break into AI, to understand a large amount of the theory you effectively have to be well versed in statistics. I come from an engineering background where it's really more understanding abstract concepts at speed and applying them correctly with frameworks. While I understand a lot can be achieved by going through projects in TensorFlow and doing some Googling, I feel like I don't understand the internals. Thanks for taking the time to read this.
12 comments
[ 4.1 ms ] story [ 41.0 ms ] threadhttps://www.youtube.com/playlist?list=PL5102DFDC6790F3D0
For Probability Theory:
https://www.youtube.com/playlist?list=PLUl4u3cNGP60hI9ATjSFg...
https://www.youtube.com/playlist?list=PL2SOU6wwxB0uwwH80KTQ6...
His content is a mix of two modes: in one mode, he's actually teaching a college class and just recording the class. In those videos, you get to hear the questions from the students and some more banter back and forth with the class. The other mode is him teaching "to air" so to speak, where it's just him in a room with a camera and a whiteboard. Diffeq is done that way, because he didn't have a Diffeq class scheduled to record. As best as I can remember, this Precalc, Calc I, Calc II, and Calc III series, and that Stats series, are done with a live class. Not sure if that will matter to you or not: I've found his style engaging and enjoyable in either case.
If you want Linear Algebra, people swear by Gilbert Strang's videos on Youtube.
https://www.youtube.com/playlist?list=PLE7DDD91010BC51F8
(There are also videos of Gilbert Strang teaching Calculus)
There's also a series of videos out there from a class called "Coding The Matrix" which was taught at uummm... I don't remember... Brown, maybe? Anyway, the class was based on teaching Linear Algebra from a programming perspective.
https://codingthematrix.com/
Math is wide for AI and moves into multiple disciplines such as calculus, linear algebra.
Once you have done the above courses you can dive accordingly http://sgsa.berkeley.edu/current_students/books/
EDIT: Also very basic. Should have added that.
This is one of the clearest and most respected statistics books ever written.
I personally owe the start of my machine learning career to this book!
You will find so many people around the world (online and IRL) who consider this the bible for statistical learning.
It is so readable, and yet filled with gems and insights from one of the world's most preeminent statisticians.
Free PDF: http://ow.ly/v5Uw50obzpO
Datasets & code: http://ow.ly/a8UZ50obzpN
R Code: http://ow.ly/EEdf50kXUBS
Videos: also available at various sources.
Since no one has really said much about Bayes yet, I think it worth mentioning just how useful it is in DS and ML. A Bayesian approach makes a very good baseline and often one that is hard to beat.
If you're not particular fluent with Probability and Statistics now, let me suggest you add in Khan Academy (make sure to pick the CLEP version) and JBstatistics. Khan has the advantage of quizzes (so you're not just kidding yourself that you know the material). JBstatistics has the advantage of really good explanations. You'll probably want to watch Khan at x1.5 speed.