Linear Algebra Review and Reference (2012) [pdf]

cs229.stanford.edu

47 points·dstein64··4 comments

I have a gripe with references in general. Most references are just reference formulae and symbol manipulation. In general, they are the form of - We call this as Title X. In this X, you have ABC; you then do MNO* operations to get YZ. There's hardly context of when to use them or what they really mean. "A Mathematician's Lament" by Paul Lockhart hits hard here.

https://www.maa.org/external_archive/devlin/LockhartsLament....

References are not made to learn from, they are made to refer back to. If you look at the URL you will see that this reference in particular is on the page for CS229, which lists knowledge of basic linear algebra as a prerequisite. I really don't think this was made with the intention of anyone using it to learn linear algebra.

Yes, references are to "refer back to" - the key part here is that what are they meant to "refer back to"? A set of formulae, symbol manipulation or how to apply functions? References should be notes that help you recollect what you'd learn. In this case, the references recollect a set of manipulation of symbols which happen to be Linear Algebra.

Do note that I am talking about "Reference Manuals" in general and not this particular CS 229 manual.

The Lagrangian on the last page associated with computing the maximal eigenpair should probably have the term lamba (x^T x - 1), not lamba x^T x. The latter gets the gradient with respect to x right, but not the gradient with respect to lambda, and it would provide the wrong dual problem.

Reply on news.ycombinator.com