CppAD: A C++ Algorithmic Differentiation Package
20130918
|
void CppAD::color_general_cppad | ( | VectorSet & | pattern, |
const VectorSize & | row, | ||
const VectorSize & | col, | ||
CppAD::vector< size_t > & | color | ||
) |
Determine which rows of a general sparse matrix can be computed together; i.e., do not have non-zero overlapping values.
VectorSize | is a simple vector class with elements of type size_t . |
VectorSet | is an unspecified type with the exception that it must support the operations under pattern and the following operations where p is a VectorSet object: VectorSet p Constructs a new vector of sets object. p.resize(ns, ne) resizes to ns sets with elements between zero ne . All of the ns sets are initially empty. p.add_element(s, e) add element e to set with index s . |
pattern | [in] Is a representation of the sparsity pattern for the matrix. Note that color_general does not change the values in pattern , but it is not const because its iterator facility modifies some of its internal data. m = pattern.n_set() sets m to the number of rows in the sparse matrix. All of the row indices are less than this value. n = pattern.end() sets n to the number of columns in the sparse matrix. All of the column indices are less than this value. pattern.begin(i) instructs the iterator facility to start iterating over columns in the i-th row of the sparsity pattern. j = pattern.next_element() Sets j to the next possibly non-zero column in the row specified by the previous call to pattern.begin . If there are no more such columns, the value pattern.end() is returned. |
row | [in] is a vector specifying which row indices to compute. |
col | [in] is a vector, with the same size as row , that specifies which column indices to compute. For each valid index k , the index pair (row[k], col[k]) must be present in the sparsity pattern. It may be that some entries in the sparsity pattern do not need to be computed; i.e, do not appear in the set of (row[k], col[k]) entries. |
color | [out] is a vector with size m . The input value of its elements does not matter. Upon return, it is a coloring for the rows of the sparse matrix. If for come i , color[i] == m , then the i-th row does not appear in the vector row . Otherwise, color[i] < m . Suppose two differen rows, i != r have the same color and column index j is such that both of the pairs (i, j) and (r, j) appear in the sparsity pattern. It follows that neither of these pairs appear in the set of (row[k], col[k]) entries. This routine tries to minimize, with respect to the choice of colors, the maximum, with respct to k , of color[ row[k] ] . |
Definition at line 105 of file color_general.hpp.
Referenced by CppAD::ADFun< Base >::SparseHessianCompute(), CppAD::ADFun< Base >::SparseJacobianFor(), and CppAD::ADFun< Base >::SparseJacobianRev().