弱问一下：Proc mixed 中final hessian 为啥要positive definite?

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

如果要最大化likelihood的话，final hessian不是应该negative definite么？

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏0 回帖

关键词：proc mixed definite positive Hessian finite proc Mixed definite final Hessian

真有用，谢谢楼主的分享了。

汗，啥意思

真的很困惑，哪位高手给解答一下啊

emptyrain 发表于 2011-3-22 10:11
如果要最大化likelihood的话，final hessian不是应该negative definite么？

In that procedure, it does minimization instead of maximization. The loglikehood is timed by -2 rather than 2. Some procedures may time no 2/-2 at all.

data t1;
  do i=1 to 30;
     x=rannor(678);
  y=1+1*x+rannor(678);
  output;
  end;
run;
proc nlmixed data=t1;
parms a b=0;
e=y-(a+b*x);
loglik=log(pdf('normal',e));
model y~general(loglik);
run;
proc model data=t1;
parms a b=0;
y=(a+b*x);
fit y/fiml;
run;
quit;

谢谢，非常感谢！

Nothing to do with +/- sign here. Minimization and maximization will be captured within the computational algorithm, not how you definted them. And Hessian matrix is related to your final variance-covariance matrix of the parameters, so have to be positive definite conceputally.