1

I wrote a python extension in C++ to work with numpy arrays. I have a memory issue.

I have a 3D numpy array with values > 0 before I call the extension. Once I am in the extension I get the numpy array using this function:

PyArrayObject * myArray = NULL;

if (!PyArg_ParseTuple(args, "O!", 
    &PyArray_Type,&myArray))  return NULL;

Using " !O " should borrow the reference to python so that I have directly access to my numpy array.

Then I access the data: 

float * myData = (float *) myArray->data;

int nbFrames =  array -> dimensions[0];
int nbRows =  array -> dimensions[1];
int nbCols =  array -> dimensions[2];

Later I check that values present in myArray are still positive: 

for(int i = 0 ; i < nbFrames; i ++){
    for( int j = 0 ; j < nbRows; j ++){
        for(int k = 0 ; k < nbCols; k++){
            if( myData[ i * nbCols * nbRows + j * nbCols + k ] < 0){
                perror("Value < 0\n");
                exit(1);
            }  
         }
    }
}

And every time I run into the case where the value is < 0. And it is not just "-0.0000", it is rather "-19.73".

So does anyone already encountered this kind of problem or does anyone know where it comes from and how to solve it?

4

1 回答 1

2

For those who will read this question, the answer was provided by sega_sai , in the comments of my question. The trick was to make sure that the array is C contiguous. To do so, you can either use the option "order = 'C' " when creating the array, for instance:

a = np.array([1,2,3,4],order='C')

(for more information see numpy reference: http://docs.scipy.org/doc/numpy/reference/generated/numpy.array.html) or transform it as a c contiguous array by doing:

np.ascontiguousarray(a)

(for mor information see numpy reference http://docs.scipy.org/doc/numpy/reference/generated/numpy.ascontiguousarray.html)

于 2013-05-06T12:31:45.397 回答