Pinv causing crash [Matlab]

Prev: parsing a string
Next: uitable: change column width

From: Hugh on 3 Dec 2009 13:23

Hi there,

I get the following error when I run some optimisation code. I assume the error stems from the fact that the matrix is singular/badly scaled.

z =

1

Maximum number of function evaluations exceeded. Increase OPTIONS.MaxFunEvals.
Warning: Matrix is singular, close to singular or badly scaled.
Results may be inaccurate. RCOND = NaN.
> In PW at 22
In CurveGroups at 68
??? Error using ==> svd
Input to SVD must not contain NaN or Inf.

Error in ==> pinv at 29
[U,S,V] = svd(A,0);

Error in ==> PW at 24
stderr=sqrt(diag((u'*u)/(length(Y)-4)*pinv(G'*G)));

Error in ==> CurveGroups at 68
y=PW(xdata2,ydata2,cn(5),cn(6),cn(7),cn(8),cn(9));

That's one thing but is there anyway to get the code to skip this result and carry on to the next set of optimisation parameters?

The full code is shown below:

%% datafitting parameters
tic;
clc
close all
clear screen
%data= dlmread('yield.txt','\t');
data= dlmread('yield2.txt','\t');

ydata=data(2,:);
xdata=data(1,:);

%ydata=data(2:end,:);

%ydata=data(2,:);

%k=size(ydata);

%for i=1:k(1)
% xdata(i,:)=data(1,:);
% end

%x=eyes(2:end,:);oi

%xdata=data(1,:);
xdata2=[2 5 10 30];
%ydata2=[4.3 4.94 5.52 5.95];
ydata2=[4.21 4.6725 5.2375 5.715];

xdata3=data(1,:);
ydata3=data(2,:);
%c0 = [1.0 2 3 4];% starting values

%c0 = [1.0 1 1];
%lbn = [-Inf -Inf -Inf -Inf -Inf -Inf]; % lower bound
%ubn = [Inf Inf Inf Inf Inf Inf]; % upper bound

options = optimset('LargeScale','off','MaxFunEvals',10000000,'TolFun',1e-5,'MaxIter',1000000);

z=0;
Res=cell(180,9);

for (i=0.01:5.01:25.1)

for (j=0.01:5.01:25.1)

for (k=0.01:1.01:5.01)
tic;

c0 = [5.95 1 1 1 i j 0.9 k 0.9 ];
%c0 = [1 1 1 1 1 0.1+i 0.9 0.9 0.9];
%c0 = [4.103594575 21.37812216 -66.97062464 0.079861881 1.81897068 -6.547214299 15.17678637 1.675525953 11.67454445];

%lbn = [-Inf -Inf -Inf -Inf i-0.00000001 j-0.00000001 -Inf k-0.00000001 -Inf]; % lower bound
%ubn = [Inf Inf Inf Inf i+0.00000001 j+0.00000001 Inf k+0.00000001 Inf];
lbn=[];
ubn=[];

z=z+1
% datafitting
%[cn,error]=lsqcurvefit(@PrimePerm,c0,xdata,ydata,lbn,ubn,options);
%[cn,error]=lsqcurvefit(@PrimePerm2,c0,xdata,ydata,lbn,ubn,options);

%[cn,error]=lsqcurvefit(@PrimePerm3,c0,xdata3,ydata3,lbn,ubn,options);
[cn,error,diff,exitf,optdata]=lsqcurvefit(@PrimePerm3,c0,xdata,ydata,lbn,ubn,options);

y=PW(xdata2,ydata2,cn(5),cn(6),cn(7),cn(8),cn(9));
params = y';
%params= [cn(1) cn(2) cn(3) cn(4) ];
params=[params cn(5) cn(6) cn(7) cn(8) cn(9)];
y=CRM(xdata3,params);
error2=(y-ydata3).^2;
error2 = sum(error2);

%Res(1,:) = [cn params error error2];

toc

% Res(1,:) = [c0 cn params y error error2 toc];
% Res2(1,:) = [exitf];
% Res3(1,:) = [optdata.firstorderopt];
% Res4(1,:) = [optdata.iterations];
% Res5(1,:) = [optdata.funcCount];
% Res6(1,:) = [optdata.cgiterations];
% Res7(z,:) = [optdata.algorithm];
% Res8(z,:) = [optdata.message];
% Res9(z,:) = [Res Res2 Res3 Res4 Res5 Res6];

Inpts(z,:) = [c0 cn params y error error2 toc];
Res{z,1} = exitf;
Res{z,2} = optdata.firstorderopt;
Res{z,3} = optdata.iterations;
Res{z,4} = optdata.funcCount;
%Res6(1,:) = [optdata.cgiterations];
Res{z,5} = optdata.stepsize;
Res{z,6} = optdata.algorithm;
Res{z,7} = optdata.message;
%Res9(z,:) = [Res Res2 Res3 Res4 Res5 Res6];

end
end
end

%optim=find(Res{:,43}==min(Res{:,43}));

toc;

From: Greg Heath on 3 Dec 2009 14:03

On Dec 3, 1:23 pm, "Hugh " <h_a_patie...(a)hotmail.com> wrote:
> Hi there,
>
> I get the following error when I run some optimisation code. I assume the error stems from the fact that the matrix is singular/badly scaled.
>
> z =
>
> 1
>
> Maximum number of function evaluations exceeded. Increase OPTIONS.MaxFunEvals.
> Warning: Matrix is singular, close to singular or badly scaled.
> Results may be inaccurate. RCOND = NaN.> In PW at 22
>
> In CurveGroups at 68
> ??? Error using ==> svd
> Input to SVD must not contain NaN or Inf.
>
> Error in ==> pinv at 29
> [U,S,V] = svd(A,0);
>
> Error in ==> PW at 24
> stderr=sqrt(diag((u'*u)/(length(Y)-4)*pinv(G'*G)));
>
> Error in ==> CurveGroups at 68
> y=PW(xdata2,ydata2,cn(5),cn(6),cn(7),cn(8),cn(9));
>
> That's one thing but is there anyway to get the code to skip this result and carry on to the next set of optimisation parameters?
>
> The full code is shown below:
>
> %% datafitting parameters
> tic;
> clc
> close all
> clear screen
> %data= dlmread('yield.txt','\t');
> data= dlmread('yield2.txt','\t');
>
> ydata=data(2,:);
> xdata=data(1,:);
>
> %ydata=data(2:end,:);
>
> %ydata=data(2,:);
>
> %k=size(ydata);
>
> %for i=1:k(1)
> % xdata(i,:)=data(1,:);
> % end
>
> %x=eyes(2:end,:);oi
>
> %xdata=data(1,:);
> xdata2=[2 5 10 30];
> %ydata2=[4.3 4.94 5.52 5.95];
> ydata2=[4.21 4.6725 5.2375 5.715];
>
> xdata3=data(1,:);
> ydata3=data(2,:);
> %c0 = [1.0 2 3 4];% starting values
>
> %c0 = [1.0 1 1];
> %lbn = [-Inf -Inf -Inf -Inf -Inf -Inf]; % lower bound
> %ubn = [Inf Inf Inf Inf Inf Inf]; % upper bound
>
> options = optimset('LargeScale','off','MaxFunEvals',10000000,'TolFun',1e-5,'MaxIter',1000000);
>
> z=0;
> Res=cell(180,9);
>
> for (i=0.01:5.01:25.1)
>
> for (j=0.01:5.01:25.1)
>
> for (k=0.01:1.01:5.01)
> tic;
>
> c0 = [5.95 1 1 1 i j 0.9 k 0.9 ];
> %c0 = [1 1 1 1 1 0.1+i 0.9 0.9 0.9];
> %c0 = [4.103594575 21.37812216 -66.97062464 0..079861881 1.81897068 -6.547214299 15.17678637 1.675525953 11.67454445];
>
> %lbn = [-Inf -Inf -Inf -Inf i-0.00000001 j-0.00000001 -Inf k-0.00000001 -Inf]; % lower bound
> %ubn = [Inf Inf Inf Inf i+0.00000001 j+0.00000001 Inf k+0.00000001 Inf];
> lbn=[];
> ubn=[];
>
> z=z+1
> % datafitting
> %[cn,error]=lsqcurvefit(@PrimePerm,c0,xdata,ydata,lbn,ubn,options);
> %[cn,error]=lsqcurvefit(@PrimePerm2,c0,xdata,ydata,lbn,ubn,options);
>
> %[cn,error]=lsqcurvefit(@PrimePerm3,c0,xdata3,ydata3,lbn,ubn,options);
> [cn,error,diff,exitf,optdata]=lsqcurvefit(@PrimePerm3,c0,xdata,ydata,lbn,ubn,options);
>
> y=PW(xdata2,ydata2,cn(5),cn(6),cn(7),cn(8),cn(9));
> params = y';
> %params= [cn(1) cn(2) cn(3) cn(4) ];
> params=[params cn(5) cn(6) cn(7) cn(8) cn(9)];
> y=CRM(xdata3,params);
> error2=(y-ydata3).^2;
> error2 = sum(error2);
>
> %Res(1,:) = [cn params error error2];
>
> toc
>
> % Res(1,:) = [c0 cn params y error error2 toc];
> % Res2(1,:) = [exitf];
> % Res3(1,:) = [optdata.firstorderopt];
> % Res4(1,:) = [optdata.iterations];
> % Res5(1,:) = [optdata.funcCount];
> % Res6(1,:) = [optdata.cgiterations];
> % Res7(z,:) = [optdata.algorithm];
> % Res8(z,:) = [optdata.message];
> % Res9(z,:) = [Res Res2 Res3 Res4 Res5 Res6];
>
> Inpts(z,:) = [c0 cn params y error error2 toc];
> Res{z,1} = exitf;
> Res{z,2} = optdata.firstorderopt;
> Res{z,3} = optdata.iterations;
> Res{z,4} = optdata.funcCount;
> %Res6(1,:) = [optdata.cgiterations];
> Res{z,5} = optdata.stepsize;
> Res{z,6} = optdata.algorithm;
> Res{z,7} = optdata.message;
> %Res9(z,:) = [Res Res2 Res3 Res4 Res5 Res6];
>
> end
> end
> end
>
> %optim=find(Res{:,43}==min(Res{:,43}));
>
> toc;

Try replacing +/1 inf with large numerical values.

Hope this helps.

Greg

From: Greg Heath on 3 Dec 2009 14:04

On Dec 3, 2:03 pm, Greg Heath <he...(a)alumni.brown.edu> wrote:
> On Dec 3, 1:23 pm, "Hugh " <h_a_patie...(a)hotmail.com> wrote:
>
>
>
>
>
> > Hi there,
>
> > I get the following error when I run some optimisation code. I assume the error stems from the fact that the matrix is singular/badly scaled.
>
> > z =
>
> > 1
>
> > Maximum number of function evaluations exceeded. Increase OPTIONS.MaxFunEvals.
> > Warning: Matrix is singular, close to singular or badly scaled.
> > Results may be inaccurate. RCOND = NaN.> In PW at 22
>
> > In CurveGroups at 68
> > ??? Error using ==> svd
> > Input to SVD must not contain NaN or Inf.
>
> > Error in ==> pinv at 29
> > [U,S,V] = svd(A,0);
>
> > Error in ==> PW at 24
> > stderr=sqrt(diag((u'*u)/(length(Y)-4)*pinv(G'*G)));
>
> > Error in ==> CurveGroups at 68
> > y=PW(xdata2,ydata2,cn(5),cn(6),cn(7),cn(8),cn(9));
>
> > That's one thing but is there anyway to get the code to skip this result and carry on to the next set of optimisation parameters?
>
> > The full code is shown below:
>
> > %% datafitting parameters
> > tic;
> > clc
> > close all
> > clear screen
> > %data= dlmread('yield.txt','\t');
> > data= dlmread('yield2.txt','\t');
>
> > ydata=data(2,:);
> > xdata=data(1,:);
>
> > %ydata=data(2:end,:);
>
> > %ydata=data(2,:);
>
> > %k=size(ydata);
>
> > %for i=1:k(1)
> > % xdata(i,:)=data(1,:);
> > % end
>
> > %x=eyes(2:end,:);oi
>
> > %xdata=data(1,:);
> > xdata2=[2 5 10 30];
> > %ydata2=[4.3 4.94 5.52 5.95];
> > ydata2=[4.21 4.6725 5.2375 5.715];
>
> > xdata3=data(1,:);
> > ydata3=data(2,:);
> > %c0 = [1.0 2 3 4];% starting values
>
> > %c0 = [1.0 1 1];
> > %lbn = [-Inf -Inf -Inf -Inf -Inf -Inf]; % lower bound
> > %ubn = [Inf Inf Inf Inf Inf Inf]; % upper bound
>
> > options = optimset('LargeScale','off','MaxFunEvals',10000000,'TolFun',1e-5,'MaxIter',1000000);
>
> > z=0;
> > Res=cell(180,9);
>
> > for (i=0.01:5.01:25.1)
>
> > for (j=0.01:5.01:25.1)
>
> > for (k=0.01:1.01:5.01)
> > tic;
>
> > c0 = [5.95 1 1 1 i j 0.9 k 0.9 ];
> > %c0 = [1 1 1 1 1 0.1+i 0.9 0.9 0.9];
> > %c0 = [4.103594575 21.37812216 -66.97062464 0.079861881 1.81897068 -6.547214299 15.17678637 1.675525953 11.67454445];
>
> > %lbn = [-Inf -Inf -Inf -Inf i-0.00000001 j-0.00000001 -Inf k-0.00000001 -Inf]; % lower bound
> > %ubn = [Inf Inf Inf Inf i+0.00000001 j+0.00000001 Inf k+0.00000001 Inf];
> > lbn=[];
> > ubn=[];
>
> > z=z+1
> > % datafitting
> > %[cn,error]=lsqcurvefit(@PrimePerm,c0,xdata,ydata,lbn,ubn,options);
> > %[cn,error]=lsqcurvefit(@PrimePerm2,c0,xdata,ydata,lbn,ubn,options);
>
> > %[cn,error]=lsqcurvefit(@PrimePerm3,c0,xdata3,ydata3,lbn,ubn,options);
> > [cn,error,diff,exitf,optdata]=lsqcurvefit(@PrimePerm3,c0,xdata,ydata,lbn,ubn,options);
>
> > y=PW(xdata2,ydata2,cn(5),cn(6),cn(7),cn(8),cn(9));
> > params = y';
> > %params= [cn(1) cn(2) cn(3) cn(4) ];
> > params=[params cn(5) cn(6) cn(7) cn(8) cn(9)];
> > y=CRM(xdata3,params);
> > error2=(y-ydata3).^2;
> > error2 = sum(error2);
>
> > %Res(1,:) = [cn params error error2];
>
> > toc
>
> > % Res(1,:) = [c0 cn params y error error2 toc];
> > % Res2(1,:) = [exitf];
> > % Res3(1,:) = [optdata.firstorderopt];
> > % Res4(1,:) = [optdata.iterations];
> > % Res5(1,:) = [optdata.funcCount];
> > % Res6(1,:) = [optdata.cgiterations];
> > % Res7(z,:) = [optdata.algorithm];
> > % Res8(z,:) = [optdata.message];
> > % Res9(z,:) = [Res Res2 Res3 Res4 Res5 Res6];
>
> > Inpts(z,:) = [c0 cn params y error error2 toc];
> > Res{z,1} = exitf;
> > Res{z,2} = optdata.firstorderopt;
> > Res{z,3} = optdata.iterations;
> > Res{z,4} = optdata.funcCount;
> > %Res6(1,:) = [optdata.cgiterations];
> > Res{z,5} = optdata.stepsize;
> > Res{z,6} = optdata.algorithm;
> > Res{z,7} = optdata.message;
> > %Res9(z,:) = [Res Res2 Res3 Res4 Res5 Res6];
>
> > end
> > end
> > end
>
> > %optim=find(Res{:,43}==min(Res{:,43}));
>
> > toc;
>
> Try replacing +/1 inf with large numerical values.

+/- inf

Hope this helps.

Greg- Hide quoted text -
>
> - Show quoted text -

From: Hugh on 9 Dec 2009 14:00

Greg Heath <heath(a)alumni.brown.edu> wrote in message <1041a733-c6b6-4228-8513-adec56e019dc(a)p32g2000vbi.googlegroups.com>...
> On Dec 3, 1:23?pm, "Hugh " <h_a_patie...(a)hotmail.com> wrote:
> > Hi there,
> >
> > I get the following error when I run some optimisation code. I assume the error stems from the fact that the matrix is singular/badly scaled.
> >
> > z =
> >
> > ? ? ?1
> >
> > Maximum number of function evaluations exceeded. Increase OPTIONS.MaxFunEvals.
> > Warning: Matrix is singular, close to singular or badly scaled.
> > ? ? ? ? ?Results may be inaccurate. RCOND = NaN.> In PW at 22
> >
> > ? In CurveGroups at 68
> > ??? Error using ==> svd
> > Input to SVD must not contain NaN or Inf.
> >
> > Error in ==> pinv at 29
> > ? ?[U,S,V] = svd(A,0);
> >
> > Error in ==> PW at 24
> > ? ? ? ? ?stderr=sqrt(diag((u'*u)/(length(Y)-4)*pinv(G'*G)));
> >
> > Error in ==> CurveGroups at 68
> > y=PW(xdata2,ydata2,cn(5),cn(6),cn(7),cn(8),cn(9));
> >
> > That's one thing but is there anyway to get the code to skip this result and carry on to the next set of optimisation parameters?
> >
> > The full code is shown below:
> >
> > %% datafitting parameters
> > tic;
> > clc
> > close all
> > clear screen
> > %data= dlmread('yield.txt','\t');
> > data= dlmread('yield2.txt','\t');
> >
> > ydata=data(2,:);
> > xdata=data(1,:);
> >
> > %ydata=data(2:end,:);
> >
> > %ydata=data(2,:);
> >
> > %k=size(ydata);
> >
> > %for i=1:k(1)
> > ?% ? ? xdata(i,:)=data(1,:);
> > ?% end
> >
> > %x=eyes(2:end,:);oi
> >
> > %xdata=data(1,:);
> > xdata2=[2 5 10 30];
> > %ydata2=[4.3 4.94 5.52 5.95];
> > ydata2=[4.21 4.6725 5.2375 5.715];
> >
> > xdata3=data(1,:);
> > ydata3=data(2,:);
> > %c0 = [1.0 2 3 4];% starting values
> >
> > %c0 = [1.0 1 1];
> > %lbn = [-Inf -Inf -Inf -Inf -Inf -Inf]; ? ? ? ? ? ? ? ? ? ? ?% lower bound
> > %ubn = [Inf Inf Inf Inf Inf Inf]; ? ? ? ? ? ? ? ? ? ? ?% upper bound
> >
> > options = optimset('LargeScale','off','MaxFunEvals',10000000,'TolFun',1e-5,'MaxIter',?1000000);
> >
> > z=0;
> > Res=cell(180,9);
> >
> > for (i=0.01:5.01:25.1)
> >
> > ? ? for (j=0.01:5.01:25.1)
> >
> > ? ? ? ? ?for (k=0.01:1.01:5.01)
> > tic; ? ? ? ? ?
> >
> > c0 = [5.95 1 1 1 i j 0.9 k 0.9 ];
> > %c0 = [1 1 1 1 1 0.1+i 0.9 0.9 0.9];
> > %c0 = [4.103594575 ? ? ?21.37812216 ? ? -66.97062464 ? ?0.079861881 ? ? 1.81897068 ? ? ?-6.547214299 ? ?15.17678637 ? ? 1.675525953 ? ? 11.67454445];
> >
> > %lbn = [-Inf -Inf -Inf -Inf i-0.00000001 j-0.00000001 -Inf k-0.00000001 -Inf]; ? ? ? ? ? ? ? ? ? ? ?% lower bound
> > %ubn = [Inf Inf Inf Inf i+0.00000001 j+0.00000001 Inf k+0.00000001 Inf];
> > lbn=[];
> > ubn=[];
> >
> > z=z+1
> > % datafitting
> > %[cn,error]=lsqcurvefit(@PrimePerm,c0,xdata,ydata,lbn,ubn,options);
> > %[cn,error]=lsqcurvefit(@PrimePerm2,c0,xdata,ydata,lbn,ubn,options);
> >
> > %[cn,error]=lsqcurvefit(@PrimePerm3,c0,xdata3,ydata3,lbn,ubn,options);
> > [cn,error,diff,exitf,optdata]=lsqcurvefit(@PrimePerm3,c0,xdata,ydata,lbn,ub?n,options);
> >
> > y=PW(xdata2,ydata2,cn(5),cn(6),cn(7),cn(8),cn(9));
> > params = y';
> > %params= [cn(1) cn(2) cn(3) cn(4) ];
> > params=[params cn(5) cn(6) cn(7) cn(8) cn(9)];
> > y=CRM(xdata3,params);
> > error2=(y-ydata3).^2;
> > error2 = sum(error2);
> >
> > %Res(1,:) = [cn params error error2];
> >
> > ?toc ? ?
> >
> > % Res(1,:) = [c0 cn params y error error2 toc];
> > % Res2(1,:) = [exitf];
> > % Res3(1,:) = [optdata.firstorderopt];
> > % Res4(1,:) = [optdata.iterations];
> > % Res5(1,:) = [optdata.funcCount];
> > % Res6(1,:) = [optdata.cgiterations];
> > % Res7(z,:) = [optdata.algorithm];
> > % Res8(z,:) = [optdata.message];
> > % Res9(z,:) = [Res Res2 Res3 Res4 Res5 Res6];
> >
> > ?Inpts(z,:) = [c0 cn params y error error2 toc];
> > ?Res{z,1} = exitf;
> > ?Res{z,2} = optdata.firstorderopt;
> > ?Res{z,3} = optdata.iterations;
> > ?Res{z,4} = optdata.funcCount;
> > ?%Res6(1,:) = [optdata.cgiterations];
> > ?Res{z,5} = optdata.stepsize;
> > ?Res{z,6} = optdata.algorithm;
> > ?Res{z,7} = optdata.message;
> > ?%Res9(z,:) = [Res Res2 Res3 Res4 Res5 Res6];
> >
> > ? ? ? ? ?end
> > ? ? end
> > end
> >
> > %optim=find(Res{:,43}==min(Res{:,43}));
> >
> > toc;
>
> Try replacing +/1 inf with large numerical values.
>
> Hope this helps.
>
> Greg

Ok so canI use this function to test for Inf:

http://www.mathworks.com/access/helpdesk/help/toolbox/compiler/staticboolisinfdoublex.html

In which case where does the c++ header file get used?

If not is there another function I can use to test for the appearance of Inf values?

Also what would be a suitable large number to replace Inf with when I find it?
Can I check to see what type of variable is being used for example double and then just put the highest legitimate value for that type of variable.

If so how do I test what type a given varoable has and how do I find out what the maximum value for they type is??

|
Pages: 1
Prev: parsing a string
Next: uitable: change column width