williamesamuelson / AffineRayleighOptimization.jl

MIT License
0 stars 0 forks source link

improving sparse solver #7

Open williamesamuelson opened 4 months ago

williamesamuelson commented 4 months ago

It seems like sparsity does not make a difference in the initial benchmark tests. In the plots below, the x-axis is $n$ (number of columns of $C$), and the y-axis is time. With RQ_SPARSE (d) I mean RQ_SPARSE supplied with the matrices converted to regular matrices/vectors.

Both $C$ and $Q$ are sparse here (using sprand(..., sparsity)), but $b$ does not have a lot of zeros.

Do the matrices have to be even sparser to make a big difference?

benchmark_k=n_over_10