I was glad to see at Bartolomeo's talk at ICCOPT that OSQP will support querying differentiation information.
We developed an extension of MOI for this at https://github.com/jump-dev/DiffOpt.jl.
At the moment, no solver implements DiffOpt's interface so the DiffOpt.Optimizer uses the solver to compute primal and dual information and then has to solve a linear system to compute these derivative information.
If package implements the DiffOpt interface then it would allow querying these informations from OSQP directly without needing this DiffOpt.Optimizer which would be faster as OSQP has already done part of the work needing to get these information when solving the problem.
I was glad to see at Bartolomeo's talk at ICCOPT that OSQP will support querying differentiation information. We developed an extension of MOI for this at https://github.com/jump-dev/DiffOpt.jl. At the moment, no solver implements DiffOpt's interface so the
DiffOpt.Optimizer
uses the solver to compute primal and dual information and then has to solve a linear system to compute these derivative information. If package implements the DiffOpt interface then it would allow querying these informations from OSQP directly without needing thisDiffOpt.Optimizer
which would be faster as OSQP has already done part of the work needing to get these information when solving the problem.