""" Copyright, the CVXPY authors Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. """ import numpy as np import scipy.sparse as sp import cvxpy.interface as intf import cvxpy.settings as s from cvxpy.error import SolverError from cvxpy.reductions.solution import Solution, failure_solution from cvxpy.reductions.solvers.qp_solvers.qp_solver import QpSolver from cvxpy.utilities.citations import CITATION_DICT def unpack_highs_options_inplace(solver_opts) -> None: # Users can pass options inside a nested dict -- e.g. to circumvent a name clash highs_options = solver_opts.pop("highs_options", dict()) # merge via update(dict(...)) is needed to avoid silently over-writing options solver_opts.update(dict(**solver_opts, **highs_options)) class HIGHS(QpSolver): """QP interface for the HiGHS solver""" # Note that HiGHS does not support MIQP but supports MILP MIP_CAPABLE = False # Map of HiGHS status to CVXPY status. STATUS_MAP = { "kNotset": s.SOLVER_ERROR, "kModelError": s.SOLVER_ERROR, "kSolveError": s.SOLVER_ERROR, "kOptimal": s.OPTIMAL, "kInfeasible": s.INFEASIBLE, "kUnboundedOrInfeasible": s.INFEASIBLE_OR_UNBOUNDED, "kUnbounded": s.UNBOUNDED, "kObjectiveBound": s.USER_LIMIT, "kObjectiveTarget": s.USER_LIMIT, "kTimeLimit": s.USER_LIMIT, "kIterationLimit": s.USER_LIMIT, "kSolutionLimit": s.USER_LIMIT, } def name(self): return s.HIGHS def import_solver(self) -> None: import highspy highspy def apply(self, problem): """ Construct QP problem data stored in a dictionary. In HiGHS, the QP has the following form minimize 1/2 x' P x + q' x subject to L <= A x <= U """ data, inv_data = super(HIGHS, self).apply(problem) return data, inv_data def invert(self, results, inverse_data): """Returns the solution to the original problem given the inverse_data.""" attr = { s.SOLVE_TIME: results["run_time"], s.EXTRA_STATS: results["info"], } # map solver statuses back to CVXPY statuses status = self.STATUS_MAP.get(results["model_status"], s.UNKNOWN) if status in s.SOLUTION_PRESENT: opt_val = results["info"].objective_function_value + inverse_data[s.OFFSET] primal_vars = { HIGHS.VAR_ID: intf.DEFAULT_INTF.const_to_matrix( np.array(results["solution"].col_value) ) } # add duals if not a MIP. dual_vars = None if not inverse_data[HIGHS.IS_MIP]: dual_vars = {HIGHS.DUAL_VAR_ID: -np.array(results["solution"].row_dual)} attr[s.NUM_ITERS] = ( results["info"].ipm_iteration_count + results["info"].crossover_iteration_count + results["info"].pdlp_iteration_count + results["info"].qp_iteration_count + results["info"].simplex_iteration_count ) sol = Solution(status, opt_val, primal_vars, dual_vars, attr) else: sol = failure_solution(status, attr) return sol def solve_via_data( self, data, warm_start: bool, verbose: bool, solver_opts, solver_cache=None ): """Returns the result of the call to the solver. minimize 1/2 x' P x + q' x subject to A x = b F x <= g in HiGHS, the opt problem format is, minimize 1/2 x' P x + q' x subject to lboundA <= A x <= uboundA Parameters ---------- data : dict Data used by the solver. warm_start : bool Whether to warm_start HiGHS. verbose : bool Should the solver print output? solver_opts : dict Additional arguments for the solver. Returns ------- tuple (status, optimal value, primal, equality dual, inequality dual) """ import highspy as hp # setup problem data inf = hp.Highs().inf P = data[s.P] q = data[s.Q] A = sp.vstack([data[s.A], data[s.F]]).tocsc() data["Ax"] = A uboundA = np.concatenate((data[s.B], data[s.G])) data["u"] = uboundA lboundA = np.concatenate([data[s.B], -inf * np.ones(data[s.G].shape)]) data["l"] = lboundA # setup highs model model = hp.HighsModel() lp = model.lp_ lp.num_col_ = A.shape[1] assert data["n_var"] == A.shape[1] lp.num_row_ = A.shape[0] assert data["n_eq"] + data["n_ineq"] == A.shape[0] # offset already applied in invert() lp.offset_ = 0 lp.col_cost_ = q lp.row_lower_ = lboundA lp.row_upper_ = uboundA assert A.format == "csc" lp.a_matrix_.format_ = hp.MatrixFormat.kColwise lp.a_matrix_.start_ = A.indptr lp.a_matrix_.index_ = A.indices lp.a_matrix_.value_ = A.data # Define Variable bounds lp.col_lower_ = -inf * np.ones(shape=lp.num_col_, dtype=q.dtype) lp.col_upper_ = inf * np.ones(shape=lp.num_col_, dtype=q.dtype) # note that we count actual nonzeros because # parameter values could make the problem linear # (i.e., P.count_nonzero() <= P.nnz) if P.count_nonzero(): hessian = model.hessian_ hessian.dim_ = model.lp_.num_col_ assert P.format == "csc" hessian.format_ = hp.HessianFormat.kSquare hessian.start_ = P.indptr hessian.index_ = P.indices hessian.value_ = P.data solver = hp.Highs() # setup options unpack_highs_options_inplace(solver_opts) solver.setOptionValue("log_to_console", verbose) for name, value in solver_opts.items(): # note that calling setOptionValue directly on the solver # allows one to pass advanced options that aren't available # on the HighOptions class (e.g., presolve_rule_off) if solver.setOptionValue(name, value) == hp.HighsStatus.kError: raise ValueError( f"HIGHS returned status kError for option (name, value): ({name}, {value})" ) solver.passModel(model) if warm_start and solver_cache is not None and self.name() in solver_cache: old_solver, old_data, old_result = solver_cache[self.name()] old_status = self.STATUS_MAP.get(old_result["model_status"], s.SOLVER_ERROR) if old_status in s.SOLUTION_PRESENT: solver.setSolution(old_result["solution"]) # initialize and solve problem try: solver.run() results = { "solution": solver.getSolution(), "basis": solver.getBasis(), "info": solver.getInfo(), "model_status": solver.getModelStatus().name, "run_time": solver.getRunTime(), } except ValueError as e: raise SolverError(e) if solver_cache is not None: solver_cache[self.name()] = (solver, data, results) return results def cite(self, data): """Returns bibtex citation for the solver. Parameters ---------- data : dict Data generated via an apply call. """ return CITATION_DICT["HIGHS"]