21.1.2 Creating Sparse Matrices

There are several means to create sparse matrix.

Returned from a function

There are many functions that directly return sparse matrices. These include speye, sprand, diag, etc.

Constructed from matrices or vectors

The function sparse allows a sparse matrix to be constructed from three vectors representing the row, column and data. Alternatively, the function spconvert uses a three column matrix format to allow easy importation of data from elsewhere.

Created and then filled

The function sparse or spalloc can be used to create an empty matrix that is then filled by the user

From a user binary program

The user can directly create the sparse matrix within an oct-file.

There are several basic functions to return specific sparse matrices. For example the sparse identity matrix, is a matrix that is often needed. It therefore has its own function to create it as speye (n) or speye (r, c), which creates an n-by-n or r-by-c sparse identity matrix.

Another typical sparse matrix that is often needed is a random distribution of random elements. The functions sprand and sprandn perform this for uniform and normal random distributions of elements. They have exactly the same calling convention, where sprand (r, c, d), creates an r-by-c sparse matrix with a density of filled elements of d.

Other functions of interest that directly create sparse matrices, are diag or its generalization spdiags, that can take the definition of the diagonals of the matrix and create the sparse matrix that corresponds to this. For example

s = diag (sparse(randn(1,n)), -1);

creates a sparse (n+1)-by-(n+1) sparse matrix with a single diagonal defined.

Function File: [b, c] = spdiags (a)

Function File: b = spdiags (a, c)

Function File: b = spdiags (v, c, a)

Function File: b = spdiags (v, c, m, n)

A generalization of the function diag. Called with a single input argument, the non-zero diagonals c of A are extracted. With two arguments the diagonals to extract are given by the vector c.

The other two forms of spdiags modify the input matrix by replacing the diagonals. They use the columns of v to replace the columns represented by the vector c. If the sparse matrix a is defined then the diagonals of this matrix are replaced. Otherwise a matrix of m by n is created with the diagonals given by v.

Negative values of c represent diagonals below the main diagonal, and positive values of c diagonals above the main diagonal.

For example

spdiags (reshape (1:12, 4, 3), [-1 0 1], 5, 4) ⇒ 5 10 0 0 1 6 11 0 0 2 7 12 0 0 3 8 0 0 0 4

Function File: y = speye (m)

Function File: y = speye (m, n)

Function File: y = speye (sz)

Returns a sparse identity matrix. This is significantly more efficient than sparse (eye (m)) as the full matrix is not constructed.

Called with a single argument a square matrix of size m by m is created. Otherwise a matrix of m by n is created. If called with a single vector argument, this argument is taken to be the size of the matrix to create.

Function File: y = spfun (f,x)

Compute f(x) for the non-zero values of x. This results in a sparse matrix with the same structure as x. The function f can be passed as a string, a function handle or an inline function.

Mapping Function: spmax (x, y, dim)

Mapping Function: [w, iw] = spmax (x)

This function has been deprecated. Use max instead.

Mapping Function: spmin (x, y, dim)

Mapping Function: [w, iw] = spmin (x)

This function has been deprecated. Use min instead.

Function File: y = spones (x)

Replace the non-zero entries of x with ones. This creates a sparse matrix with the same structure as x.

Function File: sprand (m, n, d)

Function File: sprand (s)

Generate a random sparse matrix. The size of the matrix will be m by n, with a density of values given by d. d should be between 0 and 1. Values will be uniformly distributed between 0 and 1.

Note: sometimes the actual density may be a bit smaller than d. This is unlikely to happen for large really sparse matrices.

If called with a single matrix argument, a random sparse matrix is generated wherever the matrix S is non-zero.

See also: sprandn.

Function File: sprandn (m, n, d)

Function File: sprandn (s)

Generate a random sparse matrix. The size of the matrix will be m by n, with a density of values given by d. d should be between 0 and 1. Values will be normally distributed with mean of zero and variance 1.

Note: sometimes the actual density may be a bit smaller than d. This is unlikely to happen for large really sparse matrices.

If called with a single matrix argument, a random sparse matrix is generated wherever the matrix S is non-zero.

See also: sprand.

Function File: sprandsym (n, d)

Function File: sprandsym (s)

Generate a symmetric random sparse matrix. The size of the matrix will be n by n, with a density of values given by d. d should be between 0 and 1. Values will be normally distributed with mean of zero and variance 1.

Note: sometimes the actual density may be a bit smaller than d. This is unlikely to happen for large really sparse matrices.

If called with a single matrix argument, a random sparse matrix is generated wherever the matrix S is non-zero in its lower triangular part.

See also: sprand, sprandn.

The recommended way for the user to create a sparse matrix, is to create two vectors containing the row and column index of the data and a third vector of the same size containing the data to be stored. For example

  ri = ci = d = [];
  for j = 1:c
    ri = [ri; randperm(r)(1:n)'];
    ci = [ci; j*ones(n,1)];
    d = [d; rand(n,1)];
  endfor
  s = sparse (ri, ci, d, r, c);

creates an r-by-c sparse matrix with a random distribution of n (<r) elements per column. The elements of the vectors do not need to be sorted in any particular order as Octave will sort them prior to storing the data. However, pre-sorting the data will make the creation of the sparse matrix faster.

The function spconvert takes a three or four column real matrix. The first two columns represent the row and column index respectively and the third and four columns, the real and imaginary parts of the sparse matrix. The matrix can contain zero elements and the elements can be sorted in any order. Adding zero elements is a convenient way to define the size of the sparse matrix. For example

s = spconvert ([1 2 3 4; 1 3 4 4; 1 2 3 0]')
⇒ Compressed Column Sparse (rows=4, cols=4, nnz=3)
      (1 , 1) -> 1
      (2 , 3) -> 2
      (3 , 4) -> 3

An example of creating and filling a matrix might be

k = 5;
nz = r * k;
s = spalloc (r, c, nz)
for j = 1:c
  idx = randperm (r);
  s (:, j) = [zeros(r - k, 1); ...
        rand(k, 1)] (idx);
endfor

It should be noted, that due to the way that the Octave assignment functions are written that the assignment will reallocate the memory used by the sparse matrix at each iteration of the above loop. Therefore the spalloc function ignores the nz argument and does not preassign the memory for the matrix. Therefore, it is vitally important that code using to above structure should be vectorized as much as possible to minimize the number of assignments and reduce the number of memory allocations.

Loadable Function: FM = full (SM)

returns a full storage matrix from a sparse, diagonal, permutation matrix or a range.

See also: sparse.

Function File: s = spalloc (r, c, nz)

Returns an empty sparse matrix of size r-by-c. As Octave resizes sparse matrices at the first opportunity, so that no additional space is needed, the argument nz is ignored. This function is provided only for compatibility reasons.

It should be noted that this means that code like

k = 5; nz = r * k; s = spalloc (r, c, nz) for j = 1:c idx = randperm (r); s (:, j) = [zeros(r - k, 1); rand(k, 1)] (idx); endfor

will reallocate memory at each step. It is therefore vitally important that code like this is vectorized as much as possible.

See also: sparse, nzmax.

Loadable Function: s = sparse (a)

Loadable Function: s = sparse (i, j, sv, m, n, nzmax)

Loadable Function: s = sparse (i, j, sv)

Loadable Function: s = sparse (i, j, s, m, n, "unique")

Loadable Function: s = sparse (m, n)

Create a sparse matrix from the full matrix or row, column, value triplets. If a is a full matrix, convert it to a sparse matrix representation, removing all zero values in the process.

Given the integer index vectors i and j, a 1-by-nnz vector of real of complex values sv, overall dimensions m and n of the sparse matrix. The argument nzmax is ignored but accepted for compatibility with MATLAB. If m or n are not specified their values are derived from the maximum index in the vectors i and j as given by m = max (i), n = max (j).

Note: if multiple values are specified with the same i, j indices, the corresponding values in s will be added.

The following are all equivalent:

s = sparse (i, j, s, m, n) s = sparse (i, j, s, m, n, "summation") s = sparse (i, j, s, m, n, "sum")

Given the option "unique". if more than two values are specified for the same i, j indices, the last specified value will be used.

sparse(m, n) is equivalent to sparse ([], [], [], m, n, 0)

If any of sv, i or j are scalars, they are expanded to have a common size.

See also: full.

Function File: x = spconvert (m)

This function converts for a simple sparse matrix format easily produced by other programs into Octave's internal sparse format. The input x is either a 3 or 4 column real matrix, containing the row, column, real and imaginary parts of the elements of the sparse matrix. An element with a zero real and imaginary part can be used to force a particular matrix size.

The above problem of memory reallocation can be avoided in oct-files. However, the construction of a sparse matrix from an oct-file is more complex than can be discussed here, and you are referred to chapter Dynamically Linked Functions, to have a full description of the techniques involved.