Re: Project progress

[c.]

On 6 Oct 2013, at 22:45, marco Vassallo <address@hidden> wrote:

> Ops sorry another mistake.
> Hopefully it should never be used because nnz should be an estimation from 
> the upper bound.
> 
> By the way, the idea was to imagine that the sparsity pattern of the matrix 
> is almost "constant"
> with the lines and thus, if we have nz elements in the first i lines, I apply 
> some proportion like
> (total number of nnz) : (number of non zero in the first i line) = (total 
> number of row) : (present number of row : i)
> 
> Obviously this assumption can be wrong, and maybe something like doubling the 
> size could be better.
> 
> Marco         

I don't think you need to guess what the sparsity is ...

If you don't trust nnz just check that you have enough space before writing a 
new row
and if you don't add enough space to fit that row.

But why do you say you don't trust nnz returned by the data() method? 
c.