I've read the answers in What are "coefficients of linear discriminants" in LDA?, but I still don't understand what coefficients of linear discriminants on output of R means.

What is it? (How) Is it related to the decision boundary?

nb: my knowledge about LDA can be summed up in this slide.

library(ISLR, MASS)
train <- (Smarket$Year < 2005)
lda.fit <- lda(Direction ~ Lag1 + Lag2, data = Smarket, subset = train)    
lda(Direction ~ Lag1 + Lag2, data = Smarket, subset = train)

Prior probabilities of groups:
    Down       Up 
0.491984 0.508016 

Group means:
            Lag1        Lag2
Down  0.04279022  0.03389409
Up   -0.03954635 -0.03132544

Coefficients of linear discriminants:
Lag1 -0.6420190
Lag2 -0.5135293
  • $\begingroup$ stats.stackexchange.com/q/92109/3277 is similar question $\endgroup$
    – ttnphns
    Apr 11, 2014 at 7:59
  • $\begingroup$ @ttnphns Your answer:'"Coefficients" are the regressional weights to compute the LDs by the Xs.' Linear discriminant function based on the slide I gave above is: $\delta_k (x) = x^T \Sigma ^{-1} \mu_k - \frac{1}{2} \mu^T_k \Sigma ^{-1} \mu_k + \log(\pi_k)$ Do you mean the coefficients is $\Sigma ^{-1} \mu_k$ in this case? $\endgroup$
    – hans-t
    Apr 11, 2014 at 9:52
  • $\begingroup$ I didn't examine the presentation under your link. And I'm not agile to decipher formulas. I meant there (by the coefficients) what I've outlined in a description of LDA algorithm here. You might want to take iris data and do LDA, and compare results with my output. $\endgroup$
    – ttnphns
    Apr 11, 2014 at 10:14
  • 1
    $\begingroup$ I believe that pdf doc you base yourself on is sure correct (and is quite mathematical). But I suspect it is for 2-class case only. The algorithm outlined by me (extraction, classification) is general k-class LDA algorithm. $\endgroup$
    – ttnphns
    Apr 11, 2014 at 10:55


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.