[R] Intercept in lasso models

[Martin Keller-Ressel]

Hello,

Im using the function l1ce (L1-constrained estimation) from the lasso2 
package.
when I try the example from the help pages

data(Prostate)
l1c.P <- l1ce(lpsa ~ ., Prostate, bound=.5)
coef(l1c.P)
  (Intercept)       lcavol      lweight          age         lbph          
svi
0.7284810757 0.4936540169 0.2681863403 0.0000000000 0.0092825881 
0.4550584943
          lcp      gleason        pgg45
0.0000000000 0.0000000000 0.0001812107

everything looks fine to me. But when I use

l1c.P <- l1ce(lpsa ~ ., Prostate, bound=(1:20)/20)
coef(l1c.P)[10,]
   (Intercept)        lcavol       lweight           age          lbph
-4.9496587740  0.4936540169  0.2681863403  0.0000000000  0.0092825881
           svi           lcp       gleason         pgg45
  0.4550584943  0.0000000000  0.0000000000  0.0001812107
the Intercept is different !! Also

coef(l1c.P)[20,]
   (Intercept)        lcavol       lweight           age          lbph
-23.678040866   0.587022881   0.454460641  -0.019637208   0.107054351
           svi           lcp       gleason         pgg45
   0.766155885  -0.105473570   0.045135964   0.004525324
should be the same as the unconstrained linear model:

coef(lm(lpsa ~ ., data = Prostate))
  (Intercept)       lcavol      lweight          age         lbph          
svi
  0.669399027  0.587022881  0.454460641 -0.019637208  0.107054351  
0.766155885
          lcp      gleason        pgg45
-0.105473570  0.045135964  0.004525324

which is true except for the Intercept...

What is going on here? Can anybody enlighten me?

Thanks,

Martin Keller-Ressel







--

______________________________________________
[EMAIL PROTECTED] mailing list
https://www.stat.math.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html