Error using optimization and likelihood

Question

Liliana Sierra 2017-12-7

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/371780-error-using-optimization-and-likelihood

评论： Duncan Lilley 2017-12-13

Hi, I keep getting this errors:

    Error in selectivitykk>@(initial)neg_log_likelihood(initial,c,p2,R2,R3) (line 23)
    [params,fval,exitflag,output,grad,hessian] = fminunc
    (@(initial)neg_log_likelihood(initial,c,p2,R2,R3),initial,OPTIONS);
    Error in fminunc (line 292)
            f = feval(funfcn{3},x,varargin{:});
    Error in selectivitykk (line 23)
    [params,fval,exitflag,output,grad,hessian] = fminunc
    (@(initial)neg_log_likelihood(initial,c,p2,R2,R3),initial,OPTIONS);
    Caused by:
        Failure in initial objective function evaluation. FMINUNC cannot continue.

My code so far:

    function [R0,SIG2,K]=selectivitykk
    load data.mat  % LOAD DATA
    c=zeros(31,3);
    for k=14:44 % Length from 14 cm to 44 cm
        c(k-13,1)=k; % length
        c(k-13,2)=sum(LENG2==k); % Number of fish with length k caught in 2 inch
        c(k-13,3)=sum(LENG3==k); % Number of fish with length k caught in 3 inch
        end
    R2=c(:,1)/(2*2.54); % Length divided by mesh size in cm for 2 inch mesh
    R3=c(:,1)/(3*2.54); % Length divided by mesh size in cm for 3 inch mesh
    p2=EFFORT2/(EFFORT2+EFFORT3);
    %%OPTIMIZATION
    initial=[1,1,1]; % Initial guess of R0 and Sigma (can be negative)
    % Setting options for minimization routine
    OPTIONS = optimset(@fminunc);
    OPTIONS.MaxFunEvals=1000*8;
    OPTIONS.MaxIter=10000;
    OPTIONS.TolFun=1.0000e-21;
    OPTIONS.TolX=1.0000e-14;
    OPTIONS.Algorithm='quasi-newton';
    %*********************************************************************************
    % minimize likelihood
    [params,fval,exitflag,output,grad,hessian] = fminunc  (@(initial)neg_log_likelihood(initial,c,p2,R2,R3),initial,OPTIONS);
    loglik=-fval; % Log Likelihood
    R0=sqrt(params(1)^2); % making sure R0 is positive
    SIG2=params(2)^2; %Converting Sigma to Sigma Squared (STD to VARIANCE)
    K=params(3);%making sure K is positive
    a=(1/sqrt(2*pi)*exp(R2-R0/SIG2)/2);%lowecase phi mesh size 2
    b=1/2*(1+erf(3*(R2-R0/SIG2)/sqrt(2)));%capital case phi mesh size 2
    c=(1/sqrt(2*pi)*exp(R3-R0/SIG2)/2);%lowecase phi mesh size 3
    d=1/2*(1+erf(3*(R3-R0/SIG2)/sqrt(2)));
    end
    function loglik=neg_log_likelihood(params,~,~,R2,R3)
    % Negative log likelihood assuming normal distribution
    % Cirresponidng to Equation 15 in Fujimori and Tokai 2001
    R0=sqrt(params(1)^2); % making sure R0 is positive
    SIG2=params(2); %Converting Sigma to Sigma Squared (STD to VARIANCE)
    K=params(3);%making sure K is positive
    a=(1/sqrt(2*pi)*exp(R2-R0/SIG2)/2);%lowecase phi mesh size 2
    b=1/2*(1+erf(K*(R2-R0/SIG2)/sqrt(2)));%capital case phi mesh size 2
    c=(1/sqrt(2*pi)*exp(R3-R0/SIG2)/2);%lowecase phi mesh size 3
    d=1/2*(1+erf(K*(R3-R0/SIG2)/sqrt(2)));%capital phi mesh size 3
    ps2=(2.*a.*b)./SIG2;
    ps3=(2.*c.*d)./SIG2;
    loglik=-sum(c(:,2).*log(ps2./(ps2+ps3))+c(:,3).*log(ps3./(ps2+ps3)));%Negative of equation 15
    end

Thanks!

2 个评论
显示无隐藏无

Matt J 2017-12-7

在 MATLAB Online 中打开

Try executing this

neg_log_likelihood([1,1,1],c,p2,R2,R3)

and tell us the result.

Duncan Lilley 2017-12-13

This error can occur when the objective function produces a NaN output. Try executing the command Matt suggested to see what is produced by your "neg_log_likelihood" function.

To resolve this, edit the function so that it does not produce NaN values, or set constraints on the function's inputs to prevent the objective function from producing NaN outputs.

请先登录，再进行评论。

请先登录，再回答此问题。