Greedy Gradient (GG) optimizer, for local optimum search with constraints.

Follows steepest decent, reducing step size in a finite number of steps on the way. Step directions that violate constraints are projected out or reversed.

Attributes¶

step_max: numpy.ndarray

Maximal step size for each problem variable, shape: (n_vars_float,)

step_min: numpy.ndarray

Minimal step size for each problem variable, shape: (n_vars_float,)

step_div_factor: float

Step size division factor until step_min is reached

f_tol: float

The objective function tolerance

vectorized: bool

Flag for running in vectorized mode

n_max_steps: int

The maximal number of steps without fresh gradient

memory_size: int

The number of memorized visited points

memory: tuple

Memorized data: (x, obj, grad, all_valid), each a numpy.ndarray, shapes: (memory_size, n_vars), (memory_size, n_vars), (memory_size,), (memory_size,)

Public members¶

GG(problem, step_max, step_min, step_div_factor=10.0, ...)[source]

Constructor

initialize(verbosity=0)[source]

Initialize the object.

print_info()[source]

Print solver info, called before solving

solve(verbosity=1)[source]

Run the optimization solver.

finalize(opt_results, verbosity=1)[source]

This function may be called after finishing the optimization.

classmethod new(optimizer_type, *args, **kwargs)[source]

Run-time optimizer factory.

__str__()[source]

Get info string

property initialized

Flag for finished initialization