Thread Subject:

Again about inv function

Dear all,
I have a problem about how to use inv function

Here is my data:

A=[0,0290478678447601 -0,0141099913256855 -0,000769229870092528 0,000834888891356351 -0,000767833744239176 0,000447557353876793 -0,000341233167625316 0,000213594979723706 0,000405975161430466 -0,000347763223565095 -0,00607388361880634
-0,0141099913256855 0,0290281770724332 -0,0141102570779875 -0,000765746512551112 0,000839259102833237 -0,000765866696806914 0,000477892639136813 -0,000352414875015785 0,000191134961756833 0,000431471009695488 -0,000282781975670332
-0,000769229870092528 -0,0141102570779875 0,0290405944632981 -0,0141283489207921 -0,000739653617576184 0,000822134312336111 -0,000753946667740606 0,000464069879338386 -0,000347165016618071 0,000191210163393539 0,00439484543546254
0,000834888891356351 -0,000765746512551112 -0,0141283489207921 0,0290664671496478 -0,0141671403997143 -0,000714954802116445 0,000799958491511303 -0,000732003037417882 0,000459933876232131 -0,000351282627683021 -0,00334640849289780
-0,000767833744239176 0,000839259102833237 -0,000739653617576184 -0,0141671403997143 0,0291200633432482 -0,0142035255464521 -0,000697561651568423 0,000773784490655982 -0,000715379360351614 0,000453711674401027 0,00500862063732566
0,000447557353876793 -0,000765866696806914 0,000822134312336111 -0,000714954802116445 -0,0142035255464521 0,0291435500341477 -0,0142224502249874 -0,000677566047258281 0,000768360412858587 -0,000717555013319865 -0,00192100699517450
-0,000341233167625316 0,000477892639136813 -0,000753946667740606 0,000799958491511303 -0,000697561651568423 -0,0142224502249874 0,0291074829602388 -0,0142178936743960 -0,000637818512866281 0,000728832732028223 -0,00658032116618745
0,000213594979723706 -0,000352414875015785 0,000464069879338386 -0,000732003037417882 0,000773784490655982 -0,000677566047258281 -0,0142178936743960 0,0291161819455991 -0,0142367601462918 -0,000623044326656091 -0,00835441975655993
0,000405975161430466 0,000191134961756833 -0,000347165016618071 0,000459933876232131 -0,000715379360351614 0,000768360412858587 -0,000637818512866281 -0,0142367601462918 0,0290930457701177 -0,0142077988602857 0,000307385394330950
-0,000347763223565095 0,000431471009695488 0,000191210163393539 -0,000351282627683021 0,000453711674401027 -0,000717555013319865 0,000728832732028223 -0,000623044326656091 -0,0142077988602857 0,0290600392611528 -0,00729057488009133
-0,00607388361880634 -0,000282781975670332 0,00439484543546254 -0,00334640849289780 0,00500862063732566 -0,00192100699517450 -0,00658032116618745 -0,00835441975655993 0,000307385394330950 -0,00729057488009133 2187];

b=7,25152487284537e-06;

Using inv I can make var_cov=b*inv(A)
but if I want to use more accurate / or \ how do I have to procede?

Thanks a lot in advance to everyone

Peter

"Peter Ganzo " <mgnotar@gmail.com> wrote in message <ia6r27$juh$1@fred.mathworks.com>...
> Dear all,
> I have a problem about how to use inv function
>
> Here is my data:
>
> A=[0,0290478678447601 -0,0141099913256855 -0,000769229870092528 0,000834888891356351 -0,000767833744239176 0,000447557353876793 -0,000341233167625316 0,000213594979723706 0,000405975161430466 -0,000347763223565095 -0,00607388361880634
> -0,0141099913256855 0,0290281770724332 -0,0141102570779875 -0,000765746512551112 0,000839259102833237 -0,000765866696806914 0,000477892639136813 -0,000352414875015785 0,000191134961756833 0,000431471009695488 -0,000282781975670332
> -0,000769229870092528 -0,0141102570779875 0,0290405944632981 -0,0141283489207921 -0,000739653617576184 0,000822134312336111 -0,000753946667740606 0,000464069879338386 -0,000347165016618071 0,000191210163393539 0,00439484543546254
> 0,000834888891356351 -0,000765746512551112 -0,0141283489207921 0,0290664671496478 -0,0141671403997143 -0,000714954802116445 0,000799958491511303 -0,000732003037417882 0,000459933876232131 -0,000351282627683021 -0,00334640849289780
> -0,000767833744239176 0,000839259102833237 -0,000739653617576184 -0,0141671403997143 0,0291200633432482 -0,0142035255464521 -0,000697561651568423 0,000773784490655982 -0,000715379360351614 0,000453711674401027 0,00500862063732566
> 0,000447557353876793 -0,000765866696806914 0,000822134312336111 -0,000714954802116445 -0,0142035255464521 0,0291435500341477 -0,0142224502249874 -0,000677566047258281 0,000768360412858587 -0,000717555013319865 -0,00192100699517450
> -0,000341233167625316 0,000477892639136813 -0,000753946667740606 0,000799958491511303 -0,000697561651568423 -0,0142224502249874 0,0291074829602388 -0,0142178936743960 -0,000637818512866281 0,000728832732028223 -0,00658032116618745
> 0,000213594979723706 -0,000352414875015785 0,000464069879338386 -0,000732003037417882 0,000773784490655982 -0,000677566047258281 -0,0142178936743960 0,0291161819455991 -0,0142367601462918 -0,000623044326656091 -0,00835441975655993
> 0,000405975161430466 0,000191134961756833 -0,000347165016618071 0,000459933876232131 -0,000715379360351614 0,000768360412858587 -0,000637818512866281 -0,0142367601462918 0,0290930457701177 -0,0142077988602857 0,000307385394330950
> -0,000347763223565095 0,000431471009695488 0,000191210163393539 -0,000351282627683021 0,000453711674401027 -0,000717555013319865 0,000728832732028223 -0,000623044326656091 -0,0142077988602857 0,0290600392611528 -0,00729057488009133
> -0,00607388361880634 -0,000282781975670332 0,00439484543546254 -0,00334640849289780 0,00500862063732566 -0,00192100699517450 -0,00658032116618745 -0,00835441975655993 0,000307385394330950 -0,00729057488009133 2187];
>
> b=7,25152487284537e-06;
>
> Using inv I can make var_cov=b*inv(A)
> but if I want to use more accurate / or \ how do I have to procede?
>
> Thanks a lot in advance to everyone
>
> Peter

Hi Peter,

Remember that if you wish to get good answer, the formulation of your question should also be good. I am afraid that your question is not good. Before you ask, you have to formulate your problem. Instead of that you wrote a solution
      var_cov = b*inv(A).
It means that var_cov would be a solution of the equation
      var_cov*A =b,
what is due to given sizes of the matrix A and the scalar b impossible.

A hint: formulate your problem and not a solution (which is wrong).

Mira

> Hi Peter,
>
> Remember that if you wish to get good answer, the formulation of your question should also be good. I am afraid that your question is not good. Before you ask, you have to formulate your problem. Instead of that you wrote a solution
> var_cov = b*inv(A).

> It means that var_cov would be a solution of the equation
> var_cov*A =b,
> what is due to given sizes of the matrix A and the scalar b impossible.

Probably the correct equation is var_cov*A =b*eye(size(A)), and Peter's question makes perfectly sense.

Now I have heard and read that INV(A) is poorer than A\eye(size(A)) (not check myself). That begs the question: why TMW don't bother to implement INV(A) as A\eye(size(A)) instead of issuing the non-sense warning?

Bruno

"Bruno Luong" <b.luong@fogale.findmycountry> wrote in message <ia72k6$ker$1@fred.mathworks.com>...
>
> > Hi Peter,
> >
> > Remember that if you wish to get good answer, the formulation of your question should also be good. I am afraid that your question is not good. Before you ask, you have to formulate your problem. Instead of that you wrote a solution
> > var_cov = b*inv(A).
>
> > It means that var_cov would be a solution of the equation
> > var_cov*A =b,
> > what is due to given sizes of the matrix A and the scalar b impossible.
>
> Probably the correct equation is var_cov*A =b*eye(size(A)), and Peter's question makes perfectly sense.
>
> Now I have heard and read that INV(A) is poorer than A\eye(size(A)) (not check myself). That begs the question: why TMW don't bother to implement INV(A) as A\eye(size(A)) instead of issuing the non-sense warning?
>
> Bruno

Thanks Miroslav and Bruno

Bruno your solution works well

var_cov = b*eye(size((A),1)) / A;

I am agreement on you question:
>why TMW don't bother to implement INV(A) as A\eye(size(A)) instead of issuing >the non-sense warning?

Cheers

Peter

"Bruno Luong" <b.luong@fogale.findmycountry> wrote in message
news:ia72k6$ker$1@fred.mathworks.com...
>
>> Hi Peter,
>>
>> Remember that if you wish to get good answer, the formulation of your
>> question should also be good. I am afraid that your question is not good.
>> Before you ask, you have to formulate your problem. Instead of that you
>> wrote a solution var_cov = b*inv(A).
>
>> It means that var_cov would be a solution of the equation
>> var_cov*A =b,
>> what is due to given sizes of the matrix A and the scalar b impossible.
>
> Probably the correct equation is var_cov*A =b*eye(size(A)), and Peter's
> question makes perfectly sense.
>
> Now I have heard and read that INV(A) is poorer than A\eye(size(A)) (not
> check myself). That begs the question: why TMW don't bother to implement
> INV(A) as A\eye(size(A)) instead of issuing the non-sense warning?

Because constructing an identity matrix the same size as the input matrix
would dramatically increase the memory consumption of INV. This could lead
to INV failing with an out-of-memory error if you called it on a moderately
large matrix.

--
Steve Lord
slord@mathworks.com
comp.soft-sys.matlab (CSSM) FAQ: http://matlabwiki.mathworks.com/MATLAB_FAQ
To contact Technical Support use the Contact Us link on
http://www.mathworks.com

"Steven_Lord" <slord@mathworks.com> wrote in message <ia88ol$se0$1@fred.mathworks.com>...

>
> Because constructing an identity matrix the same size as the input matrix
> would dramatically increase the memory consumption of INV. This could lead
> to INV failing with an out-of-memory error if you called it on a moderately
> large matrix.

speye?

Bruno

Allow me to elaborate more, here is what I observe, the MLINT message is on the first method "INV(A) is slow and inaccurate. Use A\b for INV(A)*B":

A=rand(3);
c=rand(3,10000);

%% METHOD 1: TAGGED AS BAD
tic
B=inv(A)*c;
toc

%% METHOD 2: MLINT SUGGESTION
tic
B=A\c;
toc

%% METHOD 3
tic
B=A\speye(size(A));
B=B*c;
toc

What I observe is Method 1 & 3 is faster than 2 (two to three times). So I conclude that method 1 with INV() must be "less accurate" according to MLINT. Method 3 is not warned by MLINT. So why TMW prefers to issue warning on Method 1 rather than implement INV(A) as A\speye(size(A))?

Bruno