mf_solver

class MFSolver(device, problem_category='boxqp', batch_size=1000)

Bases: CCVMSolver

Constructor method

Parameters:

• device (str) – The device to use for the solver. Can be “cpu” or “cuda”.

• problem_category (str) – The category of problem to solve. Can be one of “boxqp”. Defaults to “boxqp”.

• batch_size (int) – The number of times to solve a problem instance simultaneously. Defaults to 1000.

Raises:

ValueError – If the problem category is not supported by the solver.

Returns:

The DLSolver object.

Return type:

DLSolver

property parameter_key

The parameters that will be used by the solver when solving the problem.

Note

Setting this parameter after calling tune() will overwrite tuned parameters.

The parameter_key must match the following format:

• key: problem size (the number of variables in the problem).

• value: dict with these keys:

  • pump (float)

  • feedback_scale (float)

  • j (float)

    • The measurement strength

  • S (float or vector of float with size ‘problem_size’)

    • The enforced saturation value

  • dt (float)

    • The time step or the learning rate (old name was lr)

  • noise_ratio (float)

With values, the parameter key might look like this:

{
    20:
        {
            "pump": 2.5,
            "feedback_scale": 400,
            "j": 399,
            "S": 20.0,
            "dt": 0.0025,
            "iterations": 15000
        },
    30:
        {
            "pump": 3.0,
            "feedback_scale": 250,
            "j": 399,
            "S": 20.0,
            "dt": 0.0025,
            "iterations": 15000
        },
}

Raises:

ValueError – If the parameter key does not contain the solver-specific combination of keys described above.

tune(instances, post_processor, g=0.01)

Determines the best parameters for the solver to use by adjusting each parameter over a number of iterations on the problems in the given set of problems instances. The parameter_key attribute of the solver will be updated with the best parameters found.

Parameters:

• instances (list) – A list of problem instances to tune the solver on.

• post_processor (str) – The name of the post processor to use to process the results of the solver. None if no post processing is desired. Defaults to None.

• g (float) – The nonlinearity coefficient. Defaults to 0.01.

MF_SCALING_MULTIPLIER = 0.05

The value used by the MFSolver when calculating a scaling value in super.get_scaling_factor()