Home > manopt > core > getCost.m

# getCost

## PURPOSE Computes the cost function at x.

## SYNOPSIS function cost = getCost(problem, x, storedb, key)

## DESCRIPTION ``` Computes the cost function at x.

function cost = getCost(problem, x)
function cost = getCost(problem, x, storedb)
function cost = getCost(problem, x, storedb, key)

Returns the value at x of the cost function described in the problem
structure.

storedb is a StoreDB object, key is the StoreDB key to point x.

## CROSS-REFERENCE INFORMATION This function calls: This function is called by:
• getCostGrad Computes the cost function and the gradient at x in one call if possible.
• getGradientFD Computes an approx. of the gradient w/ finite differences of the cost.
• approxgradientFD Gradient approx. fnctn handle based on finite differences of the cost.
• linesearch Standard line-search algorithm (step size selection) for descent methods.
• linesearch_adaptive Adaptive line search algorithm (step size selection) for descent methods.
• linesearch_decrease Backtracking line-search aiming merely for a decrease in cost value.
• linesearch_hint Armijo line-search based on the line-search hint in the problem structure.
• pso Particle swarm optimization (PSO) for derivative-free minimization.
• trustregions Riemannian trust-regions solver for optimization on manifolds.
• checkdiff Checks the consistency of the cost function and directional derivatives.
• checkhessian Checks the consistency of the cost function and the Hessian.
• plotprofile Plot the cost function along a geodesic or a retraction path.
• surfprofile Plot the cost function as a surface over a 2-dimensional subspace.

## SOURCE CODE ```0001 function cost = getCost(problem, x, storedb, key)
0002 % Computes the cost function at x.
0003 %
0004 % function cost = getCost(problem, x)
0005 % function cost = getCost(problem, x, storedb)
0006 % function cost = getCost(problem, x, storedb, key)
0007 %
0008 % Returns the value at x of the cost function described in the problem
0009 % structure.
0010 %
0011 % storedb is a StoreDB object, key is the StoreDB key to point x.
0012 %
0014
0015 % This file is part of Manopt: www.manopt.org.
0016 % Original author: Nicolas Boumal, Dec. 30, 2012.
0017 % Contributors:
0018 % Change log:
0019 %
0020 %   April 3, 2015 (NB):
0021 %       Works with the new StoreDB class system.
0022 %
0023 %   Aug. 2, 2018 (NB):
0024 %       The value of the cost function is now always cached.
0025 %
0026 %   Sep. 6, 2018 (NB):
0028 %       with the store as input as per the user's request), then the
0029 %       gradient is also cached.
0030
0031     % Allow omission of the key, and even of storedb.
0032     if ~exist('key', 'var')
0033         if ~exist('storedb', 'var')
0034             storedb = StoreDB();
0035         end
0036         key = storedb.getNewKey();
0037     end
0038
0039
0040     % Contrary to most similar functions, here, we get the store by
0041     % default. This is for the caching functionality described below.
0042     store = storedb.getWithShared(key);
0043     store_is_stale = false;
0044
0045     % If the cost function has been computed before at this point (and its
0046     % memory is still in storedb), then we just look up the value.
0047     if isfield(store, 'cost__')
0048         cost = store.cost__;
0049         return;
0050     end
0051
0052
0053     if isfield(problem, 'cost')
0054     %% Compute the cost function using cost.
0055
0056         % Check whether this function wants to deal with storedb or not.
0057         switch nargin(problem.cost)
0058             case 1
0059                 cost = problem.cost(x);
0060             case 2
0061                 [cost, store] = problem.cost(x, store);
0062             case 3
0063                 % Pass along the whole storedb (by reference), with key.
0064                 cost = problem.cost(x, storedb, key);
0065                 % The store structure in storedb might have been modified
0066                 % (since it is passed by reference), so before caching
0067                 % we'll have to update (see below).
0068                 store_is_stale = true;
0069             otherwise
0071                     'cost should accept 1, 2 or 3 inputs.');
0072                 throw(up);
0073         end
0074
0076     %% Compute the cost function using costgrad.
0077
0078         % Check whether this function wants to deal with storedb or not.
0080             case 1
0082             case 2
0084             case 3
0085                 % Pass along the whole storedb (by reference), with key.
0086                 cost = problem.costgrad(x, storedb, key);
0087                 store_is_stale = true;
0088             otherwise
0090                     'costgrad should accept 1, 2 or 3 inputs.');
0091                 throw(up);
0092         end
0093
0094     else
0095     %% Abandon computing the cost function.
0096
0097         up = MException('manopt:getCost:fail', ...
0098             ['The problem description is not explicit enough to ' ...
0099              'compute the cost.']);
0100         throw(up);
0101
0102     end
0103
0104     % If we are not sure that the store structure is up to date, update.
0105     if store_is_stale
0106         store = storedb.getWithShared(key);
0107     end
0108
0109     % Cache here.
0110     store.cost__ = cost;
0111
0112     % If we got the cost via costgrad and it took the store as input, then
0113     % the gradient has also been computed and we can cache it.