Example of using the QUBO solver on a pool of small random instances.

# pylint: disable=invalid-name

# Maths
import numpy as np

from qtealeaves.optimization import QUBOConvergenceParameter, QUBOSolver

# qtealeaves
from qtealeaves.tensors import TensorBackend


def main(qubo_size=8, n_instances=10, input_folder=None, output_folder=None):
    """Example of a QUBO solver routine on random instances"""

    # Creating a small pool of small random (integer) QUBO problems
    np.random.seed(11)
    qubo_problems = [
        np.random.randint(-50, 50, (qubo_size, qubo_size)) for _ in range(n_instances)
    ]
    qubo_problems = [qubo + qubo.T for qubo in qubo_problems]

    # Solving the QUBOs and benchmarking with brute-force
    print("\n\n")
    separator = "-" * 60
    for jj, qprob in enumerate(qubo_problems):
        my_solver = QUBOSolver(qprob)
        print(separator)
        print(f"Solving QUBO problem instance n° {jj + 1} . . . ")

        ## Brute-force for benchmarking
        my_solver.solve("brute-force")
        bf_config = my_solver.solution[0]
        bf_cost = my_solver.cost[0]

        print(f"  Brute-force solution in {my_solver.time_to_solution} s")
        print(f"\t Cost --> {bf_cost}")
        print(f"\t Solution --> {bf_config}")

        ## Tensor-network based solver
        ## Step 1: define I/O folders
        in_folder = input_folder or "./Tensor_Network_Simulation_Input/"
        out_folder = output_folder or "./Tensor_Network_Simulation_Output/"

        ## Tensor-network based solver
        ## Step 2: define user's parameters for solving the QUBO
        my_conv_params = QUBOConvergenceParameter(
            max_bond_dimension=8,
            max_iter=10,
            enable_spacelink_expansion=True,
            transverse_field_ratio=1e9,
            tn_input_folder_path=in_folder,
            tn_output_folder_path=out_folder,
            tn_io_extra_tag=f"{jj + 1}",
            data_type="C",
            device="cpu",
        )

        ## Tensor-network based solver
        ## Step 3: define the TN backend to be used
        ##         it must match parameters in QUBOConvergenceParameter
        my_tn_backend = TensorBackend(device="cpu", dtype=np.complex64)

        ## Tensor-network based solver
        ## Step 4: call the TN-based solver
        my_solver.solve(
            tn_convergence_parameters=my_conv_params, tensor_backend=my_tn_backend
        )

        ## Tensor-network based solver
        ## Step 5: process results
        print(f"  Tensor-network solution in {my_solver.time_to_solution} s")
        print(f"\t Cost --> {my_solver.cost[0]}")
        print(f"\t Solution --> {my_solver.solution[0]}")
        print(f"Solving QUBO problem instance n° {jj + 1} . . . done")
        print(separator, "\n\n")

    # Output
    print(f"\nExample `{__file__}` ran successfully; no asserts implemented.")


if __name__ == "__main__":
    main()