(I. Introduction
In the process of deriving a closer look PCA {$ ^ [1] $}, the start of a very smooth, but later a stuck extremum formula, and this is why the problem of taking the maximum $ n '$ th a key feature of the value of reason, a card that day, it is too weak chicken. . . Later find all kinds of information on the results of watermelon book to write more briefly, saying only that the process of using the Lagrange multiplier regarded as results, but looking watermelon book data, we found there is a "pumpkin book" $ github $ on $ ^ {[2]} $ of the project, which is the great God written in great detail a complete derivation, the core ideas mainly from a question on the $ StackExchange $ site $ ^ {[3]} $, its fundamental source is a classic formula of convex optimization tutorials $ ^ {[4]} $ of $ 5.9 $ Festival, the real process of winding (funny. Did not talk much, and quickly began to carry derivation!
(B) process
- lemma:
- Question: set $ X \ in n × m, W \ in n × p $, seeking $ \ mathdrop {\ argmax} _ {c} $