91 lines
2.7 KiB
Mathematica
91 lines
2.7 KiB
Mathematica
|
function [f,df,xevaledout,dx] = fmincon_fun(x,model)
|
||
|
|
|
||
|
|
global latest_xevaled
|
||
|
|
global latest_x_xevaled
|
||
|
|
|
||
|
|
% Apply the precomputed evaluation scheme (if necessary)
|
||
|
|
xevaled = zeros(1,length(model.c));
|
||
|
|
xevaled(model.linearindicies) = x;
|
||
|
|
if ~model.SimpleLinearObjective
|
||
|
|
if isequal(x,latest_x_xevaled)
|
||
|
|
xevaled = latest_xevaled;
|
||
|
|
else
|
||
|
|
xevaled = apply_recursive_evaluation(model,xevaled);
|
||
|
|
latest_xevaled = xevaled;
|
||
|
|
latest_x_xevaled = x;
|
||
|
|
end
|
||
|
|
end
|
||
|
|
xevaledout=xevaled;
|
||
|
|
|
||
|
|
xevaled = xevaled(:);
|
||
|
|
if model.SimpleLinearObjective
|
||
|
|
f = model.f + model.c'*xevaled;
|
||
|
|
else
|
||
|
|
f = model.f + (model.c'+xevaled'*model.Q)*xevaled;
|
||
|
|
if isnan(f)
|
||
|
|
f = inf;
|
||
|
|
end
|
||
|
|
end
|
||
|
|
f=full(f);
|
||
|
|
df = [];
|
||
|
|
dx = [];
|
||
|
|
if nargout==1 || ~model.derivative_available
|
||
|
|
return
|
||
|
|
elseif model.SimpleLinearObjective
|
||
|
|
df = full(model.c(model.linearindicies));
|
||
|
|
elseif model.SimpleQuadraticObjective
|
||
|
|
df = full(model.c(model.linearindicies) + 2*model.Q(model.linearindicies,model.linearindicies)*x);
|
||
|
|
elseif model.SimpleNonlinearObjective
|
||
|
|
requested = model.c | any(model.Q,2);
|
||
|
|
[i,j,k] = find((model.deppattern(find(requested),:)));
|
||
|
|
requested(j) = 1;
|
||
|
|
df = [];
|
||
|
|
n = length(model.c);
|
||
|
|
linearindicies = model.linearindicies;
|
||
|
|
mtNonlinear = model.monomtable(model.nonlinearindicies,:);
|
||
|
|
xevaled = zeros(1,n);
|
||
|
|
xevaled(linearindicies) = x;
|
||
|
|
X = repmat(xevaled,size(mtNonlinear,1),1);
|
||
|
|
r = find(mtNonlinear);
|
||
|
|
used = find(any(mtNonlinear,1));
|
||
|
|
mtCompressed = mtNonlinear(:,used);
|
||
|
|
rCompressed = find(mtCompressed);
|
||
|
|
X = X(r);
|
||
|
|
Xones = ones(size(mtNonlinear,1),size(mtCompressed,2));
|
||
|
|
for i = 1:length(linearindicies)
|
||
|
|
if requested(i)
|
||
|
|
mt = mtNonlinear;
|
||
|
|
oldpower = mtNonlinear(:,linearindicies(i));
|
||
|
|
mt(:,linearindicies(i)) = mt(:,linearindicies(i))-1;
|
||
|
|
Z = X.^mt(r);
|
||
|
|
XX = Xones;
|
||
|
|
XX(rCompressed) = Z;
|
||
|
|
xevaledNonLinear = prod(XX,2);
|
||
|
|
xevaledNonLinear = xevaledNonLinear(:)'.*oldpower';xevaledNonLinear(isnan(xevaledNonLinear))=0;
|
||
|
|
dx = zeros(1,n);
|
||
|
|
dx(linearindicies(i)) = 1;
|
||
|
|
dx(model.nonlinearindicies) = xevaledNonLinear;
|
||
|
|
df = [df;model.c'*dx'];
|
||
|
|
else
|
||
|
|
df = [df;zeros(1,length(n))];
|
||
|
|
end
|
||
|
|
end
|
||
|
|
df = real(df + 2*model.Q(model.linearindicies,model.linearindicies)*x);
|
||
|
|
df = full(df);
|
||
|
|
elseif nargout > 1
|
||
|
|
requested = model.c | any(model.Q,2);
|
||
|
|
[i,j,k] = find((model.deppattern(find(requested),:)));
|
||
|
|
requested(j) = 1;
|
||
|
|
dx = apply_recursive_differentiation(model,xevaled,requested,model.frecursivederivativeprecompute);
|
||
|
|
df = model.c'*dx+2*xevaled'*model.Q*dx;
|
||
|
|
df = full(df);
|
||
|
|
else
|
||
|
|
df = [];
|
||
|
|
end
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|