Manopt example not working

[AH3]

Hi there,

I am completely new to Manopt and I tried the first example on my computer. And it just failed unexpectedly. Here is my code and error message. I have no idea why this happened.

Please help me there.

Thanks

[Nicolas Boumal]

Hello,

Could you please share the full script you are executing, and also let us know which version of Matlab you are running? (type version at the command prompt).

Thanks!

Nicolas

[AH3]

Hi Nicolas

I am currently using version 9.4.0.813654 (R2018a). And I tried several times, it now works. But when I try to run the example with 'caching', another problem comes out. 

Here is my script, 

importmanopt

 

n = 1000;

A = randn(n);

A = .5*(A+A.');

 

% Create the problem structure.

manifold = spherefactory(n);

problem.M = manifold;

 

problem.cost = @mycost;       % Cost function

problem.egrad = @myegrad;     % Euclidean gradient of the cost function

 

% Numerically check gradient consistency (optional).

checkgradient(problem);

 

% Solve.

[x, xcost, info, options] = trustregions(problem);

 

 

function [f, store] = mycost(A, x, store)

 

    if ~isfield(store, 'Ax')

        store.Ax = A*x;       % The store memory is associated to a specific x

    end

    Ax = store.Ax;

    

    f = -x'*Ax;               % No need to cache f: cost values are cached 'under the hood'

    

end

 

function [g, store] = myegrad(A, x, store)

 

    % This could be placed in a separate function

    % to avoid code duplication.

    if ~isfield(store, 'Ax')

        store.Ax = A*x;

    end

    Ax = store.Ax;

 

    % Euclidean gradient; this is also cached 'under the hood'.

    g = -2*Ax;

    

end

When I run this, an error message appeared as 'Undefined function or variable 'A'.

Then I added 'A' to the two functions prototype, error message became, 'Undefined operator '*' for input arguments of type 'StoreDB'.'

Thanks for your reply! Much appreciated!

[Nicolas Boumal]

Got it: just use these lines:

problem.cost = @(x, store) mycost(A, x, store);       % Cost function

problem.egrad = @(x, store) myegrad(A, x, store);     % Euclidean gradient of the cost function

The function handle you place in problem.cost and problem.egrad should (in this case, using the caching system in this way) take 2 inputs. The matrix A cannot be provided by Manopt; it must be provided by you. With the code above, you provide A explicitly, and leave it to Manopt to provide x and store, as it should.

Best,

Nicolas

[AH3]

Thanks Nicolas, 

Just another quick question, when I try to change the default value for solver, I do not know the correct syntax. I have tried 

steepestdescent(problem,[],options.tolgradnorm = 1e-4)

It will not work!

[Nicolas Boumal]

Hello,

Try this:

options = struct(); % this line is optional, but it's good practice.

options.tolgradnorm = 1e-4;

steepestdescent(problem, [], options);

Best,

Nicolas