# Array indices must be positive integers or logical values

29 views (last 30 days)
Jonas Damsbo on 8 Jan 2020
Commented: the cyclist on 9 Jan 2020
I have this MatLab code where I want to run a metropolis algorithm.
In my s rho is a matrix 12x12 and L is a matrix 12x12. m0 is a vector 1x12.
When I run my code I got the error 'Array indices must be positive integers or logical values' in hfuncval(1) where I want to compute s in the starting point.
How can come throught that?
Thank
s = -0.5.*(rho + L);
% (4) Metropolis algorithm
K = 10000; %Number of samples
pts = zeros(length(m0),K); %Array with sample output points
hfuncval = zeros(K); %Array with function value outputs
pts(:,1) = m0; %Set starting points
hfuncval(1) = s(pts(:,1)); %Compute function in starting point
step = 0.5; %Set step length
%Start sampling
for k = 1:K
[ptpert] = pts(:,k-1) + (2*rand(2,1)).*step; % propose perturbed point
hfuncpert = hfunc(ptpert); % Compute function in perturbed point
u = rand; % Generate random number in [0,1]
if u < hfuncpert/hfuncval(k-1) % accepting new points
% if u < exp(log(hfuncpert)-log(hfuncval(k-1)))
pts(:,k) = ptpert;
hfuncval(k) = hfuncpert;
else % Rejecting new points
pts(:,k) = pts(:,k-1);
hfuncval(k) = hfuncval(k-1);
end
end
The theory of the algorithm is:

Rik on 9 Jan 2020
You're using the values in the first column of pts as indices into s. So m0 contains one or more values you can't use as indices. Once you fix that you will notice this line is trying to store 12 values to a single position. Note that s is not a function, it is an array.
Jonas Damsbo on 9 Jan 2020
Okay, now I have doing this.
s = @(rho,L) -0.5.*(rho + L);
rho = (m-m0)'.*Cm.*(m-m0);
L = (y-g)'.*C_d.*(y-g);
sigma = s(rho,L);
% (4) Metropolis algorithm
K = 10000; %Number of samples
pts = zeros(length(m0),K); %Array with sample output points
hfuncval = zeros(1,K); %Array with function value outputs
pts(:,1) = m0; %Set starting points
hfuncval(1) = sigma(1,1); %Compute function in starting point
step = 0.5; %Set step length
%Start sampling
for k = 2:K
ptpert = pts(:,k-1) + (2*rand(12,1)-1)*step; % propose perturbed point
hfuncpert = sigma(ptpert); % Compute function in perturbed point
u = rand; % Generate random number in [0,1]
if u < hfuncpert/hfuncval(k-1) % accepting new points
% if u < exp(log(hfuncpert)-log(hfuncval(k-1)))
pts(:,k) = ptpert;
hfuncval(k) = hfuncpert;
else % Rejecting new points
pts(:,k) = pts(:,k-1);
hfuncval(k) = hfuncval(k-1);
end
end
But I have again the error 'Array indices must be positive integers or logical values.'
Now in this line:
hfuncpert = sigma(ptpert);
Rik on 9 Jan 2020
Look at the values of ptpert at that point. Use the debugging tools to step through your code line by line.

the cyclist on 8 Jan 2020
In this line:
hfuncval(k) = hfuncval(k-1);
in the first iteration of the for loop, k == 1, so you are attempting to access the "zeroth" element of the array hfuncval, which does not exist.

Jonas Damsbo on 8 Jan 2020
So the theoy of the algorithm is:
Jonas Damsbo on 9 Jan 2020
Okay so, I have changed a little bit in my code:
s = @(rho,L) -0.5.*(rho + L);
rho = (m-m0)'.*Cm.*(m-m0);
L = (y-g)'.*C_d.*(y-g);
sigma = s(rho,L);
% (4) Metropolis algorithm
K = 10000; %Number of samples
pts = zeros(length(m0),K); %Array with sample output points
hfuncval = zeros(1,K); %Array with function value outputs
pts(:,1) = m0; %Set starting points
hfuncval(1) = sigma(1,1); %Compute function in starting point
step = 0.5; %Set step length
%Start sampling
for k = 2:K
ptpert = pts(:,k-1) + (2*rand(12,1)-1)*step; % propose perturbed point
hfuncpert = sigma(ptpert); % Compute function in perturbed point
u = rand; % Generate random number in [0,1]
if u < hfuncpert/hfuncval(k-1) % accepting new points
% if u < exp(log(hfuncpert)-log(hfuncval(k-1)))
pts(:,k) = ptpert;
hfuncval(k) = hfuncpert;
else % Rejecting new points
pts(:,k) = pts(:,k-1);
hfuncval(k) = hfuncval(k-1);
end
end
But I have again the error 'Array indices must be positive integers or logical values.'
Now in this line:
hfuncpert = sigma(ptpert);
the cyclist on 9 Jan 2020