apply_oldrecursive_differentiation.m

matlab波形优化算法经常要用到的matlab toolbox工具箱:yalmip

function dX = apply_recursive_differentiation(model,x,requested);
dX = [];

% Compute all evaluation-based derivatives df(x)
for i = 1:length(model.evaluation_scheme)
    if isequal(model.evaluation_scheme{i}.group,'eval')
        for j = model.evaluation_scheme{i}.variables
            k = model.evalMap{j}.variableIndex;
            if any(requested(model.evalMap{j}.computes))
                z{i,j} = model.evalMap{j}.properties.derivative(x(k));
            end
        end
    end
end

% Apply chain-rule. This code is horrible
for variable = 1:length(model.linearindicies)
    dx = zeros(length(model.c),1);
    dx(model.linearindicies(variable)) = 1;
        for i = 1:length(model.evaluation_scheme)
            switch model.evaluation_scheme{i}.group
                case 'eval'
                    for j = model.evaluation_scheme{i}.variables
                        k = model.evalMap{j}.variableIndex;
                        r = find(dx(k));
                        if ~isempty(r)%any(dx(k))
                            if any(requested(model.evalMap{j}.computes))
                                if length(model.evalMap{j}.computes) == 1
                                    %dx(model.evalMap{j}.computes) = dx(k)'*model.evalMap{j}.properties.derivative(x(k));
                                    dx(model.evalMap{j}.computes) = dx(k)'*z{i,j};
                                else
                                    %dx(model.evalMap{j}.computes) = dx(k).*model.evalMap{j}.properties.derivative(x(k));
                                    dx(model.evalMap{j}.computes) = dx(k).*z{i,j};
                                end
                            end
                        end
                    end
                case 'monom'

                    computed = model.monomials(model.evaluation_scheme{i}.variables);
                                      
                    for j = computed
                        if requested(j)
                            dp = 0;
                            monomsj = model.monomtable(j,:);
                           
                            for k = find(dx' & monomsj)
                                monoms = monomsj;
                                monoms(k) = 0;
                                r = model.monomtable(j,k);
                                s = find(monoms);
                                dp = dp + r*x(k)^(r-1)*dx(k)*prod((x(s)').^monoms(s));
                            end
                            dx(j) = real(dp);
                        end
                    end

                otherwise
            end
        end
    dX = [dX dx];
end