quara.minimization_algorithm.projected_gradient_descent_backtracking module

class ProjectedGradientDescentBacktracking(func_proj=None)[source]

Bases: quara.minimization_algorithm.projected_gradient_descent.ProjectedGradientDescent

Constructor

Parameters: func_proj (Callable[[np.ndarray], np.ndarray], optional) – function of projection, by default None

is_loss_sufficient()[source]

returns whether the loss is sufficient.

Returns: whether the loss is sufficient.
Return type: bool

is_option_sufficient()[source]

returns whether the option is sufficient.

Returns: whether the option is sufficient.
Return type: bool

optimize(loss_function, loss_function_option, algorithm_option, on_iteration_history=False)[source]

optimizes using specified parameters.

Parameters

loss_function (LossFunction) – Loss Function
loss_function_option (LossFunctionOption) – Loss Function Option
algorithm_option (ProjectedGradientDescentBacktrackingOption) – Projected Gradient Descent Backtracking Algorithm Option
on_iteration_history (bool, optional) – whether to return iteration history, by default False

Returns

the result of the optimization.

Return type

ProjectedGradientDescentBacktrackingResult

Raises

ValueError – when on_value of loss_function is False.
ValueError – when on_gradient of loss_function is False.

class ProjectedGradientDescentBacktrackingOption(on_algo_eq_constraint=True, on_algo_ineq_constraint=True, var_start=None, max_iteration_optimization=1000, max_iteration_proj_physical=100000, mu=None, gamma=0.3, mode_stopping_criterion_gradient_descent='single_difference_loss', num_history_stopping_criterion_gradient_descent=1, mode_proj_order='eq_ineq', eps=None)[source]

Bases: quara.minimization_algorithm.projected_gradient_descent.ProjectedGradientDescentOption

Constructor

Parameters

on_algo_eq_constraint (bool, optional) – whether this algorithm needs on algorithm equality constraint, by default True
on_algo_ineq_constraint (bool, optional) – whether this algorithm needs on algorithm inequality constraint, by default True
var_start (np.ndarray, optional) – initial variable for the algorithm, by default None
max_iteration_optimization (int, optional) – maximun number of iterations of optimization, by default 1000.
max_iteration_proj_physical (int, optional) – maximun number of iterations of projection to physical, by default 100000.
mu (float, optional) – algorithm option mu, by default None
gamma (float, optional) – algorithm option gamma, by default 0.3
mode_stopping_criterion_gradient_descent (str, optional) – mode of stopping criterion for gradient descent, by default “single_difference_loss”
num_history_stopping_criterion_gradient_descent (int, optional) – number of history to be used stopping criterion for gradient descent, by default 1 this must be a integer and greater than or equal to 1.
mode_proj_order (str, optional) – the order in which the projections are performed, by default “eq_ineq”.
eps (float, optional) – algorithm option epsilon, by default None

property gamma: float

returns algorithm option gamma.

Returns: algorithm option gamma.
Return type: float

property mu: float

returns algorithm option mu.

Returns: algorithm option mu.
Return type: float

class ProjectedGradientDescentBacktrackingResult(value, computation_time=None, k=None, fx=None, x=None, y=None, alpha=None, error_values=None)[source]

Bases: quara.minimization_algorithm.projected_gradient_descent.ProjectedGradientDescentResult

Constructor

Parameters

value (np.ndarray) – the result of the minimization.
computation_time (float, optional) – computation time for the minimization, by default None
k (int) –
fx (List[numpy.ndarray]) –
x (List[numpy.ndarray]) –
y (List[numpy.ndarray]) –
alpha (List[float]) –
error_values (List[float]) –

property alpha: List[numpy.ndarray]

return the alpha per iteration.

Returns: the alpha per iteration.
Return type: List[np.ndarray]

property y: List[numpy.ndarray]

return the y per iteration.

Returns: the y per iteration.
Return type: List[np.ndarray]