Hi!
Ok, probably my last mail was too long. Questions are:
* Did someone try lars on data with much more components than data
points? Did you observe bad overfitting, too? Any other observations?
* Why might the lars (least angle regression from lars package)
method produce beta values that shoot up into the sky (until they
overflow), and why might the Cp values start at a _negative_ value in
step 1, growing linearly? That's not normal behaviour. What might be
wrong there?
For details see my last mail. :)
Regards,
Wiebke
