Hello,

first of all, thanks for updating the sparsematrix stuff in Eigen 3.1-alpha.

I have a sparsematrix for which i can hardly estimate the number of nonzeros per row, however I can set up the matrix row-by-row (from first to last). Therefore I would use a column index and a value vector, fill them in random order, and afterwards sort the two vectors by increasing index number. Alternatively, I could also use a sparse vector here, but would have a similar problem, as I don't know the nonzeros in advance.

(i) Is it then possible to insert the whole vector at once ?
(ii) if not, is it then at least possible to resize the respective row of the sparsematrix on the fly, such that insertion using multiple 'insert'-commands is not all to expensive?
(iii) in the code (if I remember right) before 'insert' stands something like EIGEN_DONT_INLINE. What does that mean. Would it not be more efficient to inline this function.

thanks in advance,
David

If your non zeros appears only once, then a good strategy is as follow: fill a std::vector<Triplet> where Triplet is a structure storing the i,j indices and the value. During this fill-in you can easily compute the number of rows, columns, and the number of non-zero per columns into a std::vector<int>. Then you create a rows x columns *column-major* sparse matrix, reserve space with the std::vector<int>, insert the elements in the same order as stored in the std::vector<Triplet>, and you're good to go with sorted, packed SparseMatrix !

For beta1, I plan to add a fill method taking an iterator and doing all this logic for you, so basically you would only have to fill the std::vector<Triplet> and pass it to the fill method that will take care of the rest depending on the target storage order and structure of the input.