Closed ZedongPeng closed 2 years ago
The primal and dual starting points are initialized to zero here. You'll need to modify the code to set a different starting point (current_primal_solution
and current_dual_solution
). If you do this, also be sure to set current_dual_product
to constraint_matrix' * current_dual_solution
.
How will you select the starting points?
Hi Lubin. Thanks for your reply and sorry for the late response. I am trying to solve a large-scale bidding optimization model. The model needs to be solved repeatedly thousands of times every day(with different input data). Since the formulation of the model is deterministic and only the input data will change, I am thinking about using ML or RL to obtain a good start point.
It seems like you're trying to use this code to do something useful, which was not intended :). However, I'm pleased to say that a C++ implementation of the PDLP algorithm is now available in OR-Tools. This code supports initial starting points. I'll be happy to answer further questions on that repository.
Thank you a lot.
Expected Behavior
Can we set a starting point for in the FirstOrderLp.jl solver? Or warm-start the dual variables?
In the paper, it's mentioned that