Recommender systems - matrix factorization

[Anonymous]

Hello,
I am trying to use the Collaborative Filtering library, and in particular, the Matrix Factorization class. Here is my code:

rank = 5;
GClasses::GRand r(time(NULL));
GClasses::GMatrixFactorization* G = new GClasses::GMatrixFactorization(rank,r);

GClasses::GMatrix GM(0,3);
int row_counter = 0;
for (int i = 0;i < M;i++) {
for (int j = 0;j < N;j++) {
if (MissingEntriesMatrix[i][j]) {
GM.newRow();
GM[row_counter][0] = (double) i;
GM[row_counter][1] = (double) j;
GM[row_counter][2] = (double) RatingsMatrix[i][j];
row_counter++;
}
}
}

G->train(GM);
G->setRegularizer (0.1);
GClasses::GMatrix GP = G->getP();
GClasses::GMatrix GQ = G->getQ();

cout << "number of rows " << GP->rows() << endl;
cout << "number of cols " << GP->cols() << endl;

for (int i = 0;i < GP->rows();i++) {
    for (int j = 0;j < GP->cols();j++) {
        cout << "element (" << i << "," << j << ") is " << *GP[i][j] << endl;
    }
}

It leads to a segmentation fault when printing row#1 of the matrix P. I am not sure why. Any help will be appreciated. The output is pasted below:

number of rows in GP: 6040
number of cols in GP: 6
element (0,0) is 1.75919
element (0,1) is 1.24583
element (0,2) is 1.6636
element (0,3) is 1.61166
element (0,4) is 0.939978
element (0,5) is 1.69074
/prj/icecad/lsficengspool/x/1365184578.3205763: line 8: 16750 Segmentation fault

Thanks,
Harish

[Mike Gashler]

I cannot see any problems by visually inspecting this code. I think it will be necessary to build with debug symbols and run it in a debugger to obtain a call stack. I usually use Valgrind or KDevelop for this purpose.

[Anonymous]

Hi Mike,
I did some fiddling around and the following modification to the code works:

rank = 7;
M = 20;
N = 15;
MFreg = 0.01;
GClasses::GRand r(time(NULL));
GClasses::GMatrixFactorization* G = new GClasses::GMatrixFactorization(rank,r);

GClasses::GMatrix GM(0,3);
int row_counter = 0;
for (int i = 0;i < M;i++) {
    for (int j = 0;j < N;j++) {
        if (MissingEntriesMatrix[i][j]) {
            GM.newRow();
            GM[row_counter][0] = (double) i;
            GM[row_counter][1] = (double) j;
            GM[row_counter][2] = (double) RatingsMatrix[i][j];
            row_counter++;
        }
    }
}


G->train(GM);
G->setRegularizer (MFreg);
GClasses::GMatrix GP = *(G->getP());
GClasses::GMatrix GQ = *(G->getQ());

cout << "number of rows " << GP.rows() << endl;
cout << "number of cols " << GP.cols() << endl;

for (int i = 0;i < GP.rows();i++) {
    for (int j = 0;j < GP.cols();j++) {
        cout << "element (" << i << "," << j << ") is " << GP[i][j] << endl;
    }
}

Basically, declaring GP as a matrix pointer and accessing elements using *GP[i][j] does not work. On the other hand, declaring GP directly as a matrix and accessing elements using GP[i][j] works.

Thanks,
Harish

[Mike Gashler]

I am sorry I was not more help. In hindsight, now I think I understand what the problem was. "GP[i][j]" is the same as "(GP[i][j])", not (*GP[i])[j].

So, if GP is a pointer to a matrix, then "GP[i][j]" evaluates GP as if it were a pointer to an array of matrices, so the whole term means "the jth row of the ith matrix".

If you had replaced "GP[i][j]" with "(GP[i])[j]", that would fixed it too.

Anyway, I am glad you got it to work.

Mike

[Mike Gashler]

(Apparently the asterisk is a formatting character, because it removed most of them from my text. So, the previous comment only makes sense if you insert them before each GP.)