Dynamic-Calibration/utils/YALMIP-master/@sdpvar/log.m

124 lines
2.5 KiB
Matlab
Executable File

function varargout = log(varargin)
%LOG (overloaded)
switch class(varargin{1})
case 'double'
error('Overloaded SDPVAR/LOG CALLED WITH DOUBLE. Report error')
case 'sdpvar'
% Try to detect logsumexp construction etc
varargout{1} = check_for_special_cases(varargin{:});
% Nope, then just define this logarithm
if isempty(varargout{1})
varargout{1} = InstantiateElementWise(mfilename,varargin{:});
end
case 'char'
X = varargin{3};
F = (X >= 1e-8);
operator = struct('convexity','concave','monotonicity','increasing','definiteness','none','model','callback');
operator.convexhull = @convexhull;
operator.bounds = @bounds;
operator.domain = [0 inf];
operator.derivative = @(x)(1./(abs(x)+eps));
operator.inverse = @(x)(exp(x));
varargout{1} = F;
varargout{2} = operator;
varargout{3} = X;
otherwise
error('SDPVAR/LOG called with CHAR argument?');
end
function [L,U] = bounds(xL,xU)
if xL <= 0
% The variable is not bounded enough yet
L = -inf;
else
L = log(xL);
end
if xU < 0
% This is an infeasible problem
L = inf;
U = -inf;
else
U = log(xU);
end
function [Ax, Ay, b, K] = convexhull(xL,xU)
K = [];
if xL <= 0
fL = inf;
else
fL = log(xL);
end
fU = log(xU);
dfL = 1/(xL);
dfU = 1/(xU);
%xM = (xU - xL)/(fU-fL);
xM = (xL + xU)/2;
fM = log(xM);
dfM = 1/xM;
[Ax,Ay,b] = convexhullConcave(xL,xM,xU,fL,fM,fU,dfL,dfM,dfU);
remove = isinf(b) | isinf(Ax) | isnan(b);
if any(remove)
remove = find(remove);
Ax(remove)=[];
b(remove)=[];
Ay(remove)=[];
end
function f = check_for_special_cases(x)
f = [];
% Check for log(1+x)
base = getbase(x);
if all(base(:,1)==1)
f = slog(x-1);
return;
end
% Check if user is constructing log(sum(exp(x)))
if base(1)~=0
return
end
if ~all(base(2:end)==1)
return
end
modelst = yalmip('extstruct',getvariables(x));
if isempty(modelst)
return;
end
if length(modelst)==1
models{1} = modelst;
else
models = modelst;
end
% LOG(DET(X))
if length(models)==1
if strcmp(models{1}.fcn,'det_internal')
n = length(models{1}.arg{1});
try
f = logdet(reshape(models{1}.arg{1},sqrt(n),sqrt(n)));
catch
end
return
end
end
% LOG(EXP(x1)+...+EXP(xn))
for i = 1:length(models)
if ~strcmp(models{i}.fcn,'exp')
return
end
end
p = [];
for i = 1:length(models)
p = [p;models{i}.arg{1}];
end
f = logsumexp(p);