why do I get zero likelihood!

Question

narges darvish 2013-4-5

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/70021-why-do-i-get-zero-likelihood

I am trying to use fmincon for optimizing my likelihood, but while I get the solutions the log likelihhod is zero and two of the parameters are equal to the specified initial values.

2 个评论
显示无隐藏无

Walter Roberson 2013-4-5

Which exit status do you get?

narges darvish 2013-4-8

Hi Walter, the exit is equal to one.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Tom Lane 2013-4-9

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/70021-why-do-i-get-zero-likelihood#answer_81565

在 MATLAB Online 中打开

Are you using the Statistics Toolbox mle function in combination with the Optimization Toolbox fmincon function? Can you try computing the pdf of your data using your initial parameter values, to make sure the pdf is positive for each data point?

One trick may be to make the distribution more diffuse as opposed to more concentrated. In the following example, the first mle invocation is successful using sigma=exp(3). The second fails using sigma=exp(0), because the data points are too far out in the tails of the normal distribution with mean 20 and standard deviation exp(0)=1.

x = randn(100,1);
mle(x,'pdf',@(c,mu,logs)normpdf(x,mu,exp(logs)),'start',[20 3],'optimfun','fmincon')
mle(x,'pdf',@(c,mu,logs)normpdf(x,mu,exp(logs)),'start',[20 0],'optimfun','fmincon')

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

narges darvish 2013-4-10

Thanks Tom, I am only using the Optimization toolbox. I think there was a problem with variables definition. I changed them to global and now it works, but I get zero exit!

请先登录，再进行评论。