how to get elements of an array from a matrix in matlab

% Compute the optimal value function using the Value Iteration algorithm.
for runs=1:1000
    for m=1:N1
        for n=1:N2
            for p =1:length(d)
                % Take all possible actions.
                global action
                action = d(p);
                
                s_next = [x1(m); x2(n)]+ 0.01*tank(0,[x1(m); x2(n)]);
                
                % Compute the closest discretized state.
                [r,s] = closest(s_next);
                nextV(p) = V(r,s);
                %                 
            end
            
            %             [V_best,best_ind] = max(nextV);
            %             
            %             % Improve value function estimate using Bellman's equation.
            %             V(m,n)= Reward([x1(m); x2(n)])+gamma*V_best ;
        end
    end
end

Error:

Unable to perform assignment because the left and right

sides have a different number of elements.

Error in ANN_RL (line 50)

nextV(p) = V(r,s);

Answer 1

[Walter Roberson]

                [r,s] = closest(s_next);
                nextV(p) = V(r,s);

That code could fail if V() is a function that is not returning a scalar.

That code could also fail if V is an array, but closest() is returning a non-scalar. For example what does closest() do if there are two locations that are the same distance apart?

[DEBOTRINYA SUR]

Closest function:

% computes the closest discretized state to snext

function [r,s] = closest(s_next)

global x1;
global x2;
global delta1;
global delta2;

if s_next(1) > x1(end)+delta1
    r = length(x1);
else 
    r = find(abs(x1-s_next(1)) <= delta1,1);
end

if s_next(2) > x2(end)+delta2
    s = length(x2);
else 
    s = find(abs(x2-s_next(2)) <= delta2,1);
end

end

[Walter Roberson]

    r = find(abs(x1-s_next(1)) <= delta1,1);

it is not obvious to me at the moment that that find() will definitely find any locations. Before

    nextV(p) = V(r,s);

you should put in

   assert(isscalar(r), 'r is not scalar')
   assert(isscalar(s), 's is not scalar')

[Walter Roberson]

You have floating point roundoff problems. Your delta2 is half of the distance between your x2 values.

If your x2 values were exactly evenly spaced in floating point representation, then all s_next2(2) input values within the range should be within half nominal distance between samples. However, your x2 are not equally spaced in floating point representation

K>> fprintf('%.999g\n', unique(diff(x2)))

1.0101010101009961772433598525822162628173828125

1.01010101010100328267071745358407497406005859375

1.010101010101006835384396254085004329681396484375

1.0101010101010086117412356543354690074920654296875

1.01010101010100949991965535446070134639739990234375

1.010101010101009944008865204523317515850067138671875

1.0101010101010101660534701295546256005764007568359375

1.010101010101010388098075054585933685302734375

1.010101010101013940811753855086863040924072265625

1.01010101010101749352543265558779239654541015625

Those are hex 3ff0295fad40a540 to 3ff0295fad40a5a0 -- a variety of values, not even just single bit errors.

Because of this it is possible to find s_next2(2) values that are not within delta2 of any value in x2 .

I recommend you switch to

s = interp1(x2, 1:length(x2), s_next(2), 'nearest');

If you do

s = interp1(x2, 1:length(x2), s_next(2), 'nearest', 'extrap');

then you do not need to check whether s_next(2) is greater than the end of the array (or, for that matter, before the beginning of the array): extrap will cause the first or last index to be returned as appropriate.

So you could replace closest with

% computes the closest discretized state to snext
function [r,s] = closest(s_next)
    global x1;
    global x2;
    r = interp1(x1, 1:length(x1), s_next(1), 'nearest', 'extrap');
    s = interp1(x2, 1:length(x2), s_next(2), 'nearest', 'extrap');
end