Resizing and transposing a TMatrixFSparse - row/col indices are not set

atha.stam · February 16, 2018, 8:47am

I am trying to transpose a sparse matrix. To do that, firstly I am resizing it to a square matrix, as the following code indicates

#include "TMath.h"
#include "TMatrixF.h"
#include "TMatrixFSparse.h"

void TMatrix_Transpose(){
	TMatrixFSparse G(5, 1);
        //TMatrixFSparse G(5, 2);
	
	int   icol[3] = {0, 0, 0};
	int   irow[3] = {1, 2, 3};
	float data[3] = {1.2, 1.3, 1.4};	
	G.SetMatrixArray(3, irow, icol, data);
	
	
	
	G.Print();
	
	G.ResizeTo(5, 5);
	G.T();
	G.ResizeTo(1, 5);
        //G.ResizeTo(2, 5);
	G.Print();
}

I am compiling the code in ACLiC and the output is as expected

5x1 matrix is as follows

     |      0    |
------------------
   0 |          0 
   1 |        1.2 
   2 |        1.3 
   3 |        1.4 
   4 |          0 


1x5 matrix is as follows

     |      0    |      1    |      2    |      3    |      4    |
----------------------------------------------------------------------
   0 |          0         1.2         1.3         1.4           0

The problem appears when I am changing the size of the matrix to

TMatrixFSparse G(5, 2);

I get the following output

5x2 matrix is as follows

     |      0    |      1    |
-------------------------------
   0 |          0           0 
   1 |        1.2           0 
   2 |        1.3           0 
   3 |        1.4           0 
   4 |          0           0 


2x5 matrix is as follows

     |      0    |      1    |      2    |      3    |      4    |
----------------------------------------------------------------------
Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

   0 |          0           0           0           0           0 
Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

   1 |          0           0           0           0           0

So as you can see the transpose matrix is created but it’e empty.
Any idea on what might be the issue?

Thanks in advance!

atha.stam · February 16, 2018, 11:13am

I have found that if a value is set on the second column the code works fine.

#include "TMath.h"
#include "TMatrixF.h"
#include "TMatrixFSparse.h"

void TMatrix_Transpose(){
	TMatrixFSparse G(5, 2);
	
	int   icol[3] = {0, 1, 0};
	int   irow[3] = {1, 2, 3};
	float data[3] = {1.2, 1.3, 1.4};	
	G.SetMatrixArray(3, irow, icol, data);	
	G.Print();	
	G.ResizeTo(5, 5);
	G.T();
	G.ResizeTo(2, 5);
	G.Print();
}

Does this mean that you can only transpose a matrix when it isn’t a sparse one or similarly that there is at least one non-zero element set per each col or row ?

If this is the case is there a work around on that?

kialbert · February 16, 2018, 11:26am

Check the documentation of TMatrixTSparse::ResizeTo. IIRC there’s an optional argument that determines how to deal with zeros.

Cheers,
Kim

atha.stam · February 16, 2018, 11:56am

If I understood correctly, I have to specify the number of non-zero entries, which is 3 in this example. So I changed the Resize() to

G.ResizeTo(5, 5, 3);

and then on the second resizing to

G.ResizeTo(2, 5, 3);

but I still get the same error.

Am I doing something wrong?

kialbert · February 16, 2018, 12:03pm

After a closer look I think it was a red herring.

atha.stam · February 16, 2018, 12:09pm

He he he
No problem.
So no idea?

atha.stam · February 16, 2018, 2:10pm

This is very important because later on, any additional calculations cannot be performed (i.e. multiplication). See for instance the following code

#include <iostream>
#include "TMath.h"
#include "TMatrixF.h"
#include "TMatrixFSparse.h"
#include "TMatrixFSym.h"

using namespace std;

void TMatrix_Transpose(){
	TMatrixFSparse G(5, 2);
	
	int   icol[3] = {0, 0, 0};
	int   irow[3] = {1, 2, 3};
	float data[3] = {1.2, 1.3, 1.4};	
	G.SetMatrixArray(3, irow, icol, data);	
	G.Print();	
	//G.Print();
	irow[0] = irow[1] = irow[2] = 0;
	data[0] = 1; data[1] = 2; data[2] = 3;
	TMatrixFSym A(2,data);
	cout << "1\n";
	TMatrixF B(G, TMatrixF::kMult, A);
	B.Print();
	cout << "2\n";
	G.ResizeTo(5, 5, 3);
	cout << "3\n";
	G.T();
	cout << "4\n";
	G.ResizeTo(2, 5, 3);
	G.Print();
	cout << "5\n";
	B *= G;
}

So this code gives exactly the same error

2x5 matrix is as follows

     |      0    |      1    |      2    |      3    |      4    |
----------------------------------------------------------------------
Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

   0 |          0           0           0           0           0 
Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

Error in <operator()(Int_t,Int_t) const>: row/col indices are not set
Info in <operator()>: fNrowIndex = 3 fRowIndex[fNrowIndex-1] = 0

   1 |          0           0           0           0           0 

5
Error in <operator=(const TMatrixTSparse &>: matrices not compatible

Any workaround on that?

atha.stam · February 20, 2018, 10:54am

The only workaround I can think of, is to fill the whole with zeros, and then assign the non-zero values using SetMatrixArray(), but I guess there is no point in that since it contradicts the meaning of the sparsity.

Wile_E_Coyote · February 20, 2018, 11:10am

Maybe you should try some external library, like Eigen and/or Boost and/or some BLAS + LAPACK implementation (e.g. ATLAS).

atha.stam · February 20, 2018, 11:23am

Thanks a lot for your reply.
It seems odd not to do that in root even though there is a nice library for matrices.

In any case I need to have this running on lxplus, which means that I am not able to install it there.

Wile_E_Coyote · February 20, 2018, 11:25am

I guess you should be able to download the source code of any package and build and install everything for yourself (e.g. in ${HOME}/bin, ${HOME}/lib, ${HOME}/include and so on).

atha.stam · February 20, 2018, 11:30am

It could be an option, but I selected to use root in order to avoid building a library, since root provides tools for matrices!

Is there a reason on why the sparse matrix are used like that or it could be a possible bug?

Wile_E_Coyote · February 20, 2018, 11:32am

I guess, you should be able to use these libraries from inside of ROOT, especially if you use ROOT 6 (with ROOT 5 you will probably need to use your macros as precompiled only, i.e. via ACLiC).

atha.stam · February 20, 2018, 12:51pm

Luckily I found that boost is installed on lxplus in /usr/include/boost

The problem is that when I try to run a simple example, I get the following error

Info in <TUnixSystem::ACLiC>: creating shared library /eos/user/a/astamato/Forums/RootForum/./boost_C.so
Error in <ACLiC>: Dictionary generation failed with a core dump!
Info in <ACLiC>: Invoking compiler to check macro's validity
Info in <ACLiC>: The compiler has not found any problem with your macro.
	Probably your macro uses something rootcint can't parse.
	Check http://root.cern.ch/viewvc/branches/v5-34-00-patches/cint/doc/limitati.txt for Cint's limitations.

My code is

#include </usr/include/boost/numeric/ublas/matrix.hpp>
#include </usr/include/boost/numeric/ublas/io.hpp>

int test_boost () {
    using namespace boost::numeric::ublas;
    matrix<double> m (3, 3);
    for (unsigned i = 0; i < m.size1 (); ++ i)
        for (unsigned j = 0; j < m.size2 (); ++ j)
            m (i, j) = 3 * i + j;
    std::cout << m << std::endl;
    return 0;
}

and I am compiling it in ACLiC in root 5.34 doing

.L boost.C++

Any idea?

Wile_E_Coyote · February 20, 2018, 1:02pm

Try to switch to ROOT 6.
Try to “hide” everything that CINT (in ROOT 5) cannot parse:

#if !defined(__CINT__)
#include ... whatever is needed ...
#endif /* !defined(__CINT__) */

system · March 6, 2018, 1:03pm

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.