patch-warm-start.md

How to patch warm-start capability into cvxpy interface for many solvers ?

OSQP

1. Navigate to cvxpy/reductions/solvers/qp_solvers/osqp_qpif.py
2. Add the following code snippet to the OSQP class:

# add this to the top of the file
from cvxpy.reductions.solvers import utilities
...
...
class OSQP(QpSolver):
  ....
  ....
    def apply(self, problem):
        ...
        ...
        # Add initial guess.
        data['init_value'] = utilities.stack_vals(problem.variables, default=0.0)
        return data, inv_data
    
    def solve_via_data(self, data, warm_start, verbose, solver_opts):
        ...
        ...
        try:
          solver.setup(P, q, A, lA, uA, verbose=verbose, **solver_opts)
          # add this block
          if warm_start and 'init_value' in data:
              init_x = data['init_value']
              print(f"Using warm start with user-provided initial value: shape={init_x.shape}")
              solver.warm_start(x=init_x, y=None)
        ...
        ...

ProxQP

1. Navigate to cvxpy/reductions/solvers/qp_solvers/prox_qpif.py
2. Add the following code snippet to the PROXQP class:

# add this to the top of the file
from cvxpy.reductions.solvers import utilities
...
...
class PROXQP(QpSolver):
  ....
  ....
    def apply(self, problem):
        ...
        ...
        # Add initial guess.
        data['init_value'] = utilities.stack_vals(problem.variables, default=0.0)
        return data, inv_data
    
    def solve_via_data(self, data, warm_start, verbose, solver_opts):
        ...
        ...
        if warm_start and 'init_value' in data:
                init_x = data['init_value']
                print(f"Using warm start with user-provided initial value: shape={init_x.shape}")
                solver.solve(init_x, None, None)
            else:
                print(f"Cold-starting PROXQP solver")
                solver.solve()
        ...
        ...