Python API Reference¶

Inner PANOC solver¶

class panocpy._panocpy.PANOCSolver

C++ documentation: pa::PANOCSolver

__call__(self: panocpy._panocpy.PANOCSolver, problem: panocpy._panocpy.Problem, Σ: numpy.ndarray[numpy.float64[m, 1]], ε: float, x: numpy.ndarray[numpy.float64[m, 1]], y: numpy.ndarray[numpy.float64[m, 1]]) → Tuple[numpy.ndarray[numpy.float64[m, 1]], numpy.ndarray[numpy.float64[m, 1]], numpy.ndarray[numpy.float64[m, 1]], dict]

Solve.

Parameters

problem – Problem to solve
Σ – Penalty factor
ε – Desired tolerance
x – Initial guess
y – Initial Lagrange multipliers

Returns

Solution \(x\)
Updated Lagrange multipliers \(y\)
Slack variable error \(g(x) - z\)
Statistics

__init__(*args, **kwargs)

Overloaded function.

__init__(self: panocpy._panocpy.PANOCSolver) -> None
__init__(self: panocpy._panocpy.PANOCSolver, panoc_params: panocpy._panocpy.PANOCParams, lbfgs_direction: panocpy._panocpy.LBFGSDirection) -> None
__init__(self: panocpy._panocpy.PANOCSolver, panoc_params: panocpy._panocpy.PANOCParams, lbfgs_params: panocpy._panocpy.LBFGSParams) -> None
__init__(self: panocpy._panocpy.PANOCSolver, panoc_params: panocpy._panocpy.PANOCParams, direction: panocpy._panocpy.PANOCDirection) -> None

set_progress_callback(self: panocpy._panocpy.PANOCSolver, callback: Callable[[panocpy._panocpy.PANOCProgressInfo], None]) → None: Attach a callback that is called on each iteration of the solver.

property direction

property params

class panocpy._panocpy.PANOCParams

C++ documentation: pa::PANOCParams

__init__(*args, **kwargs)

Overloaded function.

__init__(self: panocpy._panocpy.PANOCParams) -> None
__init__(self: panocpy._panocpy.PANOCParams, **kwargs) -> None

to_dict(self: panocpy._panocpy.PANOCParams) → dict

property L_max

property L_min

property Lipschitz

property alternative_linesearch_cond

property lbfgs_stepsize

property max_iter

property max_no_progress

property max_time

property print_interval

property quadratic_upperbound_tolerance_factor

property update_lipschitz_in_linesearch

property τ_min

class panocpy._panocpy.LBFGSParams

C++ documentation: pa::LBFGSParams

__init__(*args, **kwargs)

Overloaded function.

__init__(self: panocpy._panocpy.LBFGSParams) -> None
__init__(self: panocpy._panocpy.LBFGSParams, **kwargs) -> None

to_dict(self: panocpy._panocpy.LBFGSParams) → dict

property cbfgs

property memory

property rescale_when_γ_changes

Inner Structured PANOC solver¶

class panocpy._panocpy.StructuredPANOCLBFGSSolver

C++ documentation: pa::StructuredPANOCLBFGSSolver

__call__(self: panocpy._panocpy.StructuredPANOCLBFGSSolver, problem: panocpy._panocpy.Problem, Σ: numpy.ndarray[numpy.float64[m, 1]], ε: float, x: numpy.ndarray[numpy.float64[m, 1]], y: numpy.ndarray[numpy.float64[m, 1]]) → Tuple[numpy.ndarray[numpy.float64[m, 1]], numpy.ndarray[numpy.float64[m, 1]], numpy.ndarray[numpy.float64[m, 1]], dict]

Solve.

Parameters

problem – Problem to solve
Σ – Penalty factor
ε – Desired tolerance
x – Initial guess
y – Initial Lagrange multipliers

Returns

Solution \(x\)
Updated Lagrange multipliers \(y\)
Slack variable error \(g(x) - z\)
Statistics

__init__(*args, **kwargs)

Overloaded function.

__init__(self: panocpy._panocpy.StructuredPANOCLBFGSSolver) -> None
__init__(self: panocpy._panocpy.StructuredPANOCLBFGSSolver, panoc_params: panocpy._panocpy.StructuredPANOCLBFGSParams, lbfgs_params: panocpy._panocpy.LBFGSParams) -> None

set_progress_callback(self: panocpy._panocpy.StructuredPANOCLBFGSSolver, callback: Callable[[panocpy._panocpy.StructuredPANOCLBFGSProgressInfo], None]) → None: Attach a callback that is called on each iteration of the solver.

property params

class panocpy._panocpy.StructuredPANOCLBFGSParams

C++ documentation: pa::StructuredPANOCLBFGSParams

__init__(self: panocpy._panocpy.StructuredPANOCLBFGSParams, **kwargs) → None

to_dict(self: panocpy._panocpy.StructuredPANOCLBFGSParams) → dict

property L_max

property L_min

property Lipschitz

property alternative_linesearch_cond

property full_augmented_hessian

property hessian_vec_finited_differences

property lbfgs_stepsize

property max_iter

property max_no_progress

property max_time

property nonmonotone_linesearch

property print_interval

property quadratic_upperbound_tolerance_factor

property stop_crit

property update_lipschitz_in_linesearch

property τ_min

Outer ALM solver¶

class panocpy._panocpy.ALMSolver

Main augmented Lagrangian solver.

C++ documentation: pa::ALMSolver

__call__(self: panocpy._panocpy.ALMSolver, problem: panocpy._panocpy.Problem, x: Optional[numpy.ndarray[numpy.float64[m, 1]]] = None, y: Optional[numpy.ndarray[numpy.float64[m, 1]]] = None) → Tuple[numpy.ndarray[numpy.float64[m, 1]], numpy.ndarray[numpy.float64[m, 1]], dict]

Solve.

Parameters

problem – Problem to solve.
y – Initial guess for Lagrange multipliers \(y\)
x – Initial guess for decision variables \(x\)

Returns

Lagrange multipliers \(y\) at the solution
Solution \(x\)
Statistics

__init__(*args, **kwargs)

Overloaded function.

__init__(self: panocpy._panocpy.ALMSolver) -> None

Build an ALM solver using Structured PANOC as inner solver.

__init__(self: panocpy._panocpy.ALMSolver, alm_params: panocpy._panocpy.ALMParams, panoc_solver: panocpy._panocpy.PANOCSolver) -> None

Build an ALM solver using PANOC as inner solver.

__init__(self: panocpy._panocpy.ALMSolver, alm_params: panocpy._panocpy.ALMParams, pga_solver: panocpy._panocpy.PGASolver) -> None

Build an ALM solver using the projected gradient algorithm as inner solver.

__init__(self: panocpy._panocpy.ALMSolver, alm_params: panocpy._panocpy.ALMParams, structuredpanoc_solver: panocpy._panocpy.StructuredPANOCLBFGSSolver) -> None

Build an ALM solver using Structured PANOC as inner solver.

__init__(self: panocpy._panocpy.ALMSolver, alm_params: panocpy._panocpy.ALMParams, inner_solver: panocpy._panocpy.InnerSolver) -> None

Build an ALM solver using a custom inner solver.

__init__(self: panocpy._panocpy.ALMSolver, panoc_solver: panocpy._panocpy.PANOCSolver) -> None

Build an ALM solver using PANOC as inner solver.

__init__(self: panocpy._panocpy.ALMSolver, pga_solver: panocpy._panocpy.PGASolver) -> None

Build an ALM solver using the projected gradient algorithm as inner solver.

__init__(self: panocpy._panocpy.ALMSolver, structuredpanoc_solver: panocpy._panocpy.StructuredPANOCLBFGSSolver) -> None

Build an ALM solver using Structured PANOC as inner solver.

__init__(self: panocpy._panocpy.ALMSolver, inner_solver: panocpy._panocpy.InnerSolver) -> None

Build an ALM solver using a custom inner solver.

__init__(self: panocpy._panocpy.ALMSolver, alm_params: panocpy._panocpy.ALMParams) -> None

Build an ALM solver using Structured PANOC as inner solver.

property inner_solver

property params

class panocpy._panocpy.ALMParams

C++ documentation: pa::ALMParams

__init__(*args, **kwargs)

Overloaded function.

__init__(self: panocpy._panocpy.ALMParams) -> None
__init__(self: panocpy._panocpy.ALMParams, **kwargs) -> None

to_dict(self: panocpy._panocpy.ALMParams) → dict

property M

property max_iter

property max_num_initial_retries

property max_num_retries

property max_time

property max_total_num_retries

property preconditioning

property print_interval

property single_penalty_factor

property Δ

property Δ_lower

property Σ_0

property Σ_0_lower

property Σ_max

property Σ_min

property δ

property ε

property ε_0

property ε_0_increase

property θ

property ρ

property ρ_increase

property σ_0

Problem formulation¶

class panocpy._panocpy.Problem

C++ documentation: pa::Problem

__init__(self: panocpy._panocpy.Problem, n: int, m: int) → None

Parameters

n – Number of unknowns
m – Number of constraints

property C: Box constraints on \(x\)

property D: Box constraints on \(g(x)\)

property f: Objective funcion, \(f(x)\)

property g: Constraint function, \(g(x)\)

property grad_f: Gradient of the objective function, \(\nabla f(x)\)

property grad_g_prod: Gradient of constraint function times vector, \(\nabla g(x)\, v\)

property grad_gi: Gradient vector of the \(i\)-th component of the constriant function, \(\nabla g_i(x)\)

property hess_L: Hessian of the Lagrangian function, \(\nabla^2_{xx} L(x,y)\)

property hess_L_prod: Hessian of the Lagrangian function times vector, \(\nabla^2_{xx} L(x,y)\, v\)

property m: Number of general constraints, dimension of \(g(x)\)

property n: Number of unknowns, dimension of \(x\)

class panocpy._panocpy.ProblemWithParam

C++ documentation: pa::ProblemWithParam

See panocpy._panocpy.Problem for the full documentation.

__init__(self: panocpy._panocpy.ProblemWithParam, n: int, m: int) → None

property C

property D

property f

property g

property grad_f

property grad_g_prod

property grad_gi

property hess_L

property hess_L_prod

property m

property n

property param: Parameter vector \(p\) of the problem

CasADi Interface¶

panocpy.casadi_problem.compile_and_load_problem(cgen: casadi.casadi.CodeGenerator, n: int, m: int, p: int, name: str = 'PANOC_ALM_problem') → Union[panocpy._panocpy.Problem, panocpy._panocpy.ProblemWithParam][source]

Compile the C-code using the given code-generator and load it as a panocpy Problem.

Parameters

cgen – Code generator to generate C-code for the costs and the constraints with.
n – Dimensions of the decision variables (primal dimension).
m – Number of nonlinear constraints (dual dimension).
p – Number of parameters.
name – Optional string description of the problem (used for filename).

Returns

Problem specification that can be passed to the solvers.

panocpy.casadi_problem.compile_and_load_problem_full(cgen: casadi.casadi.CodeGenerator, n: int, m1: int, m2: int, p: int, name: str = 'PANOC_full_problem') → Union[panocpy._panocpy.ProblemFull, panocpy._panocpy.ProblemFullWithParam][source]

Compile the C-code using the given code-generator and load it as a panocpy ProblemFull.

Parameters

cgen – Code generator to generate C-code for the costs and the constraints with.
n – Dimensions of the decision variables (primal dimension).
m1 – Number of ALM constraints (dual dimension 1).
m2 – Number of quadratic penalty constraints (dual dimension 2).
p – Number of parameters.
name – Optional string description of the problem (used for filename).

Returns

ProblemFull specification that can be passed to the solvers.

panocpy.casadi_problem.generate_and_compile_casadi_problem(f: casadi.casadi.Function, g: casadi.casadi.Function, second_order: bool = False, name: str = 'PANOC_ALM_problem') → Union[panocpy._panocpy.Problem, panocpy._panocpy.ProblemWithParam][source]

Compile the objective and constraint functions into a panocpy Problem.

Parameters

f – Objective function.
g – Constraint function.
second_order – Whether to generate functions for evaluating Hessians.
name – Optional string description of the problem (used for filename).

Returns

Problem specification that can be passed to the solvers.

panocpy.casadi_problem.generate_casadi_problem(f: casadi.casadi.Function, g: casadi.casadi.Function, second_order: bool = False, name: str = 'PANOC_ALM_problem') → Tuple[casadi.casadi.CodeGenerator, int, int, int][source]

Convert the objective and constraint functions into a CasADi code generator.

Parameters

f – Objective function.
g – Constraint function.
second_order – Whether to generate functions for evaluating Hessians.
name – Optional string description of the problem (used for filename).

Returns

Code generator that generates the functions and derivatives used by the solvers.
Dimensions of the decision variables (primal dimension).
Number of nonlinear constraints (dual dimension).
Number of parameters.

panocpy.casadi_problem.generate_casadi_problem_full(f: casadi.casadi.Function, g1: casadi.casadi.Function, g2: casadi.casadi.Function, second_order: bool = False, name: str = 'PANOC_ALM_problem') → Tuple[casadi.casadi.CodeGenerator, int, int, int][source]

Convert the objective and constraint functions into a CasADi code generator.

Parameters

f – Objective function.
g1 – ALM constraint function.
g2 – Quadratic penalty constraint function.
second_order – Whether to generate functions for evaluating Hessians.
name – Optional string description of the problem (used for filename).

Returns

Code generator that generates the functions and derivatives used by the solvers.
Dimensions of the decision variables (primal dimension).
Number of nonlinear constraints (dual dimension).
Number of parameters.

All¶

PANOC+ALM solvers

class panocpy._panocpy.ALMParams¶

C++ documentation: pa::ALMParams

__init__(*args, **kwargs)¶

Overloaded function.

__init__(self: panocpy._panocpy.ALMParams) -> None
__init__(self: panocpy._panocpy.ALMParams, **kwargs) -> None

to_dict(self: panocpy._panocpy.ALMParams) → dict¶

property M¶

property max_iter¶

property max_num_initial_retries¶

property max_num_retries¶

property max_time¶

property max_total_num_retries¶

property preconditioning¶

property print_interval¶

property single_penalty_factor¶

property Δ¶

property Δ_lower¶

property Σ_0¶

property Σ_0_lower¶

property Σ_max¶

property Σ_min¶

property δ¶

property ε¶

property ε_0¶

property ε_0_increase¶

property θ¶

property ρ¶

property ρ_increase¶

property σ_0¶

class panocpy._panocpy.ALMSolver¶

Main augmented Lagrangian solver.

C++ documentation: pa::ALMSolver

Solve.

Parameters

problem – Problem to solve.
y – Initial guess for Lagrange multipliers \(y\)
x – Initial guess for decision variables \(x\)

Returns

Lagrange multipliers \(y\) at the solution
Solution \(x\)
Statistics

__init__(*args, **kwargs)¶

Overloaded function.

__init__(self: panocpy._panocpy.ALMSolver) -> None

Build an ALM solver using Structured PANOC as inner solver.

__init__(self: panocpy._panocpy.ALMSolver, alm_params: panocpy._panocpy.ALMParams, panoc_solver: panocpy._panocpy.PANOCSolver) -> None

Build an ALM solver using PANOC as inner solver.

__init__(self: panocpy._panocpy.ALMSolver, alm_params: panocpy._panocpy.ALMParams, pga_solver: panocpy._panocpy.PGASolver) -> None

Build an ALM solver using the projected gradient algorithm as inner solver.

__init__(self: panocpy._panocpy.ALMSolver, alm_params: panocpy._panocpy.ALMParams, structuredpanoc_solver: panocpy._panocpy.StructuredPANOCLBFGSSolver) -> None

Build an ALM solver using Structured PANOC as inner solver.

__init__(self: panocpy._panocpy.ALMSolver, alm_params: panocpy._panocpy.ALMParams, inner_solver: panocpy._panocpy.InnerSolver) -> None

Build an ALM solver using a custom inner solver.

__init__(self: panocpy._panocpy.ALMSolver, panoc_solver: panocpy._panocpy.PANOCSolver) -> None

Build an ALM solver using PANOC as inner solver.

__init__(self: panocpy._panocpy.ALMSolver, pga_solver: panocpy._panocpy.PGASolver) -> None

Build an ALM solver using the projected gradient algorithm as inner solver.

__init__(self: panocpy._panocpy.ALMSolver, structuredpanoc_solver: panocpy._panocpy.StructuredPANOCLBFGSSolver) -> None

Build an ALM solver using Structured PANOC as inner solver.

__init__(self: panocpy._panocpy.ALMSolver, inner_solver: panocpy._panocpy.InnerSolver) -> None

Build an ALM solver using a custom inner solver.

__init__(self: panocpy._panocpy.ALMSolver, alm_params: panocpy._panocpy.ALMParams) -> None

Build an ALM solver using Structured PANOC as inner solver.

property inner_solver¶

property params¶

class panocpy._panocpy.ALMSolverFull¶

Main augmented Lagrangian solver.

C++ documentation: pa::ALMSolverFull

__call__(self: panocpy._panocpy.ALMSolverFull, problem: panocpy._panocpy.ProblemFull, x: Optional[numpy.ndarray[numpy.float64[m, 1]]] = None, y: Optional[numpy.ndarray[numpy.float64[m, 1]]] = None) → Tuple[numpy.ndarray[numpy.float64[m, 1]], numpy.ndarray[numpy.float64[m, 1]], dict]¶

Solve.

Parameters

problem – Problem to solve.
y – Initial guess for Lagrange multipliers \(y\)
x – Initial guess for decision variables \(x\)

Returns

Lagrange multipliers \(y\) at the solution
Solution \(x\)
Statistics

__init__(self: panocpy._panocpy.ALMSolverFull, arg0: panocpy._panocpy.ALMParams, arg1: panocpy._panocpy.PANOCSolverFull) → None¶: Build an ALM solver using Structured PANOC as inner solver.

property params¶

class panocpy._panocpy.Box¶

C++ documentation: pa::Box

__init__(*args, **kwargs)¶

Overloaded function.

__init__(self: panocpy._panocpy.Box, n: int) -> None

Create an \(n\)-dimensional box at with bounds at \(\pm\infty\) (no constraints).

__init__(self: panocpy._panocpy.Box, ub: numpy.ndarray[numpy.float64[m, 1]], lb: numpy.ndarray[numpy.float64[m, 1]]) -> None

Create a box with the given bounds.

property lowerbound¶

property upperbound¶

class panocpy._panocpy.GAAPGAParams¶

C++ documentation: pa::GAAPGAParams

__init__(*args, **kwargs)¶

Overloaded function.

__init__(self: panocpy._panocpy.GAAPGAParams) -> None
__init__(self: panocpy._panocpy.GAAPGAParams, **kwargs) -> None

to_dict(self: panocpy._panocpy.GAAPGAParams) → dict¶

property L_max¶

property L_min¶

property Lipschitz¶

property full_flush_on_γ_change¶

property limitedqr_mem¶

property max_iter¶

property max_no_progress¶

property max_time¶

property print_interval¶

property quadratic_upperbound_tolerance_factor¶

property stop_crit¶

class panocpy._panocpy.GAAPGAProgressInfo¶

C++ documentation: pa::GAAPGAProgressInfo

__init__(*args, **kwargs)¶

property L¶

property fpr¶

property grad_ψ¶

property grad_ψ_hat¶

property k¶

property norm_sq_p¶

property p¶

property x¶

property x_hat¶

property y¶

property Σ¶

property γ¶

property ε¶

property ψ¶

property ψ_hat¶

class panocpy._panocpy.GAAPGASolver¶

C++ documentation: pa::GAAPGASolver

__call__(self: panocpy._panocpy.GAAPGASolver, problem: panocpy._panocpy.Problem, Σ: numpy.ndarray[numpy.float64[m, 1]], ε: float, x: numpy.ndarray[numpy.float64[m, 1]], y: numpy.ndarray[numpy.float64[m, 1]]) → Tuple[numpy.ndarray[numpy.float64[m, 1]], numpy.ndarray[numpy.float64[m, 1]], numpy.ndarray[numpy.float64[m, 1]], dict]¶

Solve.

Parameters

problem – Problem to solve
Σ – Penalty factor
ε – Desired tolerance
x – Initial guess
y – Initial Lagrange multipliers

Returns

Solution \(x\)
Updated Lagrange multipliers \(y\)
Slack variable error \(g(x) - z\)
Statistics

__init__(self: panocpy._panocpy.GAAPGASolver, arg0: panocpy._panocpy.GAAPGAParams) → None¶

set_progress_callback(self: panocpy._panocpy.GAAPGASolver, callback: Callable[[panocpy._panocpy.GAAPGAProgressInfo], None]) → None¶: Attach a callback that is called on each iteration of the solver.

property params¶

class panocpy._panocpy.InnerSolver¶

__call__(self: panocpy._panocpy.InnerSolver, arg0: panocpy._panocpy.Problem, arg1: numpy.ndarray[numpy.float64[m, 1]], arg2: float, arg3: bool, arg4: numpy.ndarray[numpy.float64[m, 1], flags.writeable], arg5: numpy.ndarray[numpy.float64[m, 1], flags.writeable], arg6: numpy.ndarray[numpy.float64[m, 1], flags.writeable]) → panocpy._panocpy.InnerSolverStats¶

__init__(self: panocpy._panocpy.InnerSolver) → None¶

get_name(self: panocpy._panocpy.InnerSolver) → str¶

get_params(self: panocpy._panocpy.InnerSolver) → object¶

stop(self: panocpy._panocpy.InnerSolver) → None¶

class panocpy._panocpy.InnerSolverStats¶

__init__(self: panocpy._panocpy.InnerSolverStats, arg0: dict) → None¶

class panocpy._panocpy.LBFGSDirection¶

C++ documentation: pa::LBFGSDirection

__init__(self: panocpy._panocpy.LBFGSDirection, params: pa::LBFGSParams) → None¶

apply(self: panocpy._panocpy.LBFGSDirection, arg0: numpy.ndarray[numpy.float64[m, 1]], arg1: numpy.ndarray[numpy.float64[m, 1]], arg2: numpy.ndarray[numpy.float64[m, 1]], arg3: float) → numpy.ndarray[numpy.float64[m, 1]]¶

changed_γ(self: panocpy._panocpy.LBFGSDirection, arg0: float, arg1: float) → None¶

get_name(self: panocpy._panocpy.LBFGSDirection) → str¶

initialize(self: panocpy._panocpy.LBFGSDirection, arg0: numpy.ndarray[numpy.float64[m, 1]], arg1: numpy.ndarray[numpy.float64[m, 1]], arg2: numpy.ndarray[numpy.float64[m, 1]], arg3: numpy.ndarray[numpy.float64[m, 1]]) → None¶

reset(self: panocpy._panocpy.LBFGSDirection) → None¶

update(self: panocpy._panocpy.LBFGSDirection, arg0: numpy.ndarray[numpy.float64[m, 1]], arg1: numpy.ndarray[numpy.float64[m, 1]], arg2: numpy.ndarray[numpy.float64[m, 1]], arg3: numpy.ndarray[numpy.float64[m, 1]], arg4: numpy.ndarray[numpy.float64[m, 1]], arg5: panocpy._panocpy.Box, arg6: float) → bool¶

property params¶

class panocpy._panocpy.LBFGSParams¶

C++ documentation: pa::LBFGSParams

__init__(*args, **kwargs)¶

Overloaded function.

__init__(self: panocpy._panocpy.LBFGSParams) -> None
__init__(self: panocpy._panocpy.LBFGSParams, **kwargs) -> None

to_dict(self: panocpy._panocpy.LBFGSParams) → dict¶

property cbfgs¶

property memory¶

property rescale_when_γ_changes¶

class panocpy._panocpy.LBFGSParamsCBFGS¶

C++ documentation: pa::LBFGSParams::cbfgs

__init__(*args, **kwargs)¶

Overloaded function.

__init__(self: panocpy._panocpy.LBFGSParamsCBFGS) -> None
__init__(self: panocpy._panocpy.LBFGSParamsCBFGS, **kwargs) -> None

to_dict(self: panocpy._panocpy.LBFGSParamsCBFGS) → dict¶

property α¶

property ϵ¶

class panocpy._panocpy.LBFGSStepsize¶

C++ documentation: pa::LBFGSStepSize

Members:

BasedOnGradientStepSize

BasedOnCurvature

__init__(self: panocpy._panocpy.LBFGSStepsize, value: int) → None¶

BasedOnCurvature = <LBFGSStepsize.BasedOnCurvature: 1>¶

BasedOnGradientStepSize = <LBFGSStepsize.BasedOnGradientStepSize: 0>¶

property name¶

property value¶

class panocpy._panocpy.LipschitzEstimateParams¶

C++ documentation: pa::LipschitzEstimateParams

__init__(*args, **kwargs)¶

Overloaded function.

__init__(self: panocpy._panocpy.LipschitzEstimateParams) -> None
__init__(self: panocpy._panocpy.LipschitzEstimateParams, **kwargs) -> None

to_dict(self: panocpy._panocpy.LipschitzEstimateParams) → dict¶

property L_0¶

property Lγ_factor¶

property δ¶

property ε¶

class panocpy._panocpy.PANOCDirection¶

Class that provides fast directions for the PANOC algorithm (e.g. L-BFGS)

__init__(self: panocpy._panocpy.PANOCDirection) → None¶

apply(self: panocpy._panocpy.PANOCDirection, arg0: numpy.ndarray[numpy.float64[m, 1]], arg1: numpy.ndarray[numpy.float64[m, 1]], arg2: numpy.ndarray[numpy.float64[m, 1]], arg3: float) → numpy.ndarray[numpy.float64[m, 1]]¶

changed_γ(self: panocpy._panocpy.PANOCDirection, arg0: float, arg1: float) → None¶

get_name(self: panocpy._panocpy.PANOCDirection) → str¶

initialize(self: panocpy._panocpy.PANOCDirection, arg0: numpy.ndarray[numpy.float64[m, 1]], arg1: numpy.ndarray[numpy.float64[m, 1]], arg2: numpy.ndarray[numpy.float64[m, 1]], arg3: numpy.ndarray[numpy.float64[m, 1]]) → None¶

reset(self: panocpy._panocpy.PANOCDirection) → None¶

update(self: panocpy._panocpy.PANOCDirection, arg0: numpy.ndarray[numpy.float64[m, 1]], arg1: numpy.ndarray[numpy.float64[m, 1]], arg2: numpy.ndarray[numpy.float64[m, 1]], arg3: numpy.ndarray[numpy.float64[m, 1]], arg4: numpy.ndarray[numpy.float64[m, 1]], arg5: panocpy._panocpy.Box, arg6: float) → bool¶

class panocpy._panocpy.PANOCParams¶

C++ documentation: pa::PANOCParams

__init__(*args, **kwargs)¶

Overloaded function.

__init__(self: panocpy._panocpy.PANOCParams) -> None
__init__(self: panocpy._panocpy.PANOCParams, **kwargs) -> None

to_dict(self: panocpy._panocpy.PANOCParams) → dict¶

property L_max¶

property L_min¶

property Lipschitz¶

property alternative_linesearch_cond¶

property lbfgs_stepsize¶

property max_iter¶

property max_no_progress¶

property max_time¶

property print_interval¶

property quadratic_upperbound_tolerance_factor¶

property update_lipschitz_in_linesearch¶

property τ_min¶

class panocpy._panocpy.PANOCProgressInfo¶

Data passed to the PANOC progress callback.

C++ documentation: pa::PANOCProgressInfo

__init__(*args, **kwargs)¶

property L¶: Estimate of Lipschitz constant of objective \(L\)

property fpr¶: Fixed-point residual \(\left\|p\right\| / \gamma\)

property grad_ψ¶: Gradient of objective \(\nabla\psi(x)\)

property grad_ψ_hat¶

property k¶: Iteration

property norm_sq_p¶: \(\left\|p\right\|^2\)

property p¶: Projected gradient step \(p\)

property x¶: Decision variable \(x\)

property x_hat¶: Decision variable after projected gradient step \(\hat x\)

property y¶: Lagrange multipliers \(y\)

property Σ¶: Penalty factor \(\Sigma\)

property γ¶: Step size \(\gamma\)

property ε¶: Tolerance reached \(\varepsilon_k\)

property τ¶: Line search parameter \(\tau\)

property φγ¶: Forward-backward envelope \(\varphi_\gamma(x)\)

property ψ¶: Objective value \(\psi(x)\)

property ψ_hat¶

class panocpy._panocpy.PANOCSolver¶

C++ documentation: pa::PANOCSolver

Solve.

Parameters

problem – Problem to solve
Σ – Penalty factor
ε – Desired tolerance
x – Initial guess
y – Initial Lagrange multipliers

Returns

Solution \(x\)
Updated Lagrange multipliers \(y\)
Slack variable error \(g(x) - z\)
Statistics

__init__(*args, **kwargs)¶

Overloaded function.

__init__(self: panocpy._panocpy.PANOCSolver) -> None
__init__(self: panocpy._panocpy.PANOCSolver, panoc_params: panocpy._panocpy.PANOCParams, lbfgs_direction: panocpy._panocpy.LBFGSDirection) -> None
__init__(self: panocpy._panocpy.PANOCSolver, panoc_params: panocpy._panocpy.PANOCParams, lbfgs_params: panocpy._panocpy.LBFGSParams) -> None
__init__(self: panocpy._panocpy.PANOCSolver, panoc_params: panocpy._panocpy.PANOCParams, direction: panocpy._panocpy.PANOCDirection) -> None

set_progress_callback(self: panocpy._panocpy.PANOCSolver, callback: Callable[[panocpy._panocpy.PANOCProgressInfo], None]) → None¶: Attach a callback that is called on each iteration of the solver.

property direction¶

property params¶

class panocpy._panocpy.PANOCSolverFull¶

C++ documentation: pa::PANOCSolverFull

__call__(self: panocpy._panocpy.PANOCSolverFull, problem: panocpy._panocpy.ProblemFull, Σ1: numpy.ndarray[numpy.float64[m, 1]], Σ2: numpy.ndarray[numpy.float64[m, 1]], ε: float, x: numpy.ndarray[numpy.float64[m, 1]], y: numpy.ndarray[numpy.float64[m, 1]]) → Tuple[numpy.ndarray[numpy.float64[m, 1]], numpy.ndarray[numpy.float64[m, 1]], numpy.ndarray[numpy.float64[m, 1]], numpy.ndarray[numpy.float64[m, 1]], dict]¶

Solve.

Parameters

problem – Problem to solve
Σ1 – Penalty factor
Σ2 – Penalty factor
ε – Desired tolerance
x – Initial guess
y – Initial Lagrange multipliers

Returns

Solution \(x\)
Updated Lagrange multipliers \(y\)
Slack variable error \(g1(x) - z\)
Slack variable error \(g2(x) - z\)
Statistics

__init__(self: panocpy._panocpy.PANOCSolverFull, arg0: panocpy._panocpy.PANOCParams, arg1: panocpy._panocpy.LBFGSParams) → None¶

set_progress_callback(self: panocpy._panocpy.PANOCSolverFull, callback: Callable[[pa::PANOCFullProgressInfo], None]) → panocpy._panocpy.PANOCSolverFull¶: Attach a callback that is called on each iteration of the solver.

property direction¶

property params¶

class panocpy._panocpy.PANOCStopCrit¶

C++ documentation: pa::PANOCStopCrit

Members:

ApproxKKT

ProjGradNorm

ProjGradUnitNorm

FPRNorm

__init__(self: panocpy._panocpy.PANOCStopCrit, value: int) → None¶

ApproxKKT = <PANOCStopCrit.ApproxKKT: 0>¶

FPRNorm = <PANOCStopCrit.FPRNorm: 3>¶

ProjGradNorm = <PANOCStopCrit.ProjGradNorm: 1>¶

ProjGradUnitNorm = <PANOCStopCrit.ProjGradUnitNorm: 2>¶

property name¶

property value¶

class panocpy._panocpy.PGAParams¶

C++ documentation: pa::PGAParams

__init__(*args, **kwargs)¶

Overloaded function.

__init__(self: panocpy._panocpy.PGAParams) -> None
__init__(self: panocpy._panocpy.PGAParams, **kwargs) -> None

to_dict(self: panocpy._panocpy.PGAParams) → dict¶

property L_max¶

property L_min¶

property Lipschitz¶

property max_iter¶

property max_time¶

property print_interval¶

property quadratic_upperbound_tolerance_factor¶

property stop_crit¶

class panocpy._panocpy.PGAProgressInfo¶

C++ documentation: pa::PGAProgressInfo

__init__(*args, **kwargs)¶

property L¶

property fpr¶

property grad_ψ¶

property grad_ψ_hat¶

property k¶

property norm_sq_p¶

property p¶

property x¶

property x_hat¶

property y¶

property Σ¶

property γ¶

property ε¶

property ψ¶

property ψ_hat¶

class panocpy._panocpy.PGASolver¶

C++ documentation: pa::PGASolver

__call__(self: panocpy._panocpy.PGASolver, problem: panocpy._panocpy.Problem, Σ: numpy.ndarray[numpy.float64[m, 1]], ε: float, x: numpy.ndarray[numpy.float64[m, 1]], y: numpy.ndarray[numpy.float64[m, 1]]) → Tuple[numpy.ndarray[numpy.float64[m, 1]], numpy.ndarray[numpy.float64[m, 1]], numpy.ndarray[numpy.float64[m, 1]], dict]¶

Solve.

Parameters

problem – Problem to solve
Σ – Penalty factor
ε – Desired tolerance
x – Initial guess
y – Initial Lagrange multipliers

Returns

Solution \(x\)
Updated Lagrange multipliers \(y\)
Slack variable error \(g(x) - z\)
Statistics

__init__(self: panocpy._panocpy.PGASolver, arg0: panocpy._panocpy.PGAParams) → None¶

set_progress_callback(self: panocpy._panocpy.PGASolver, callback: Callable[[panocpy._panocpy.PGAProgressInfo], None]) → None¶: Attach a callback that is called on each iteration of the solver.

property params¶

class panocpy._panocpy.Problem¶

C++ documentation: pa::Problem

__init__(self: panocpy._panocpy.Problem, n: int, m: int) → None¶

Parameters

n – Number of unknowns
m – Number of constraints

property C¶: Box constraints on \(x\)

property D¶: Box constraints on \(g(x)\)

property f¶: Objective funcion, \(f(x)\)

property g¶: Constraint function, \(g(x)\)

property grad_f¶: Gradient of the objective function, \(\nabla f(x)\)

property grad_g_prod¶: Gradient of constraint function times vector, \(\nabla g(x)\, v\)

property grad_gi¶: Gradient vector of the \(i\)-th component of the constriant function, \(\nabla g_i(x)\)

property hess_L¶: Hessian of the Lagrangian function, \(\nabla^2_{xx} L(x,y)\)

property hess_L_prod¶: Hessian of the Lagrangian function times vector, \(\nabla^2_{xx} L(x,y)\, v\)

property m¶: Number of general constraints, dimension of \(g(x)\)

property n¶: Number of unknowns, dimension of \(x\)

class panocpy._panocpy.ProblemFull¶

C++ documentation: pa::ProblemFull

__init__(self: panocpy._panocpy.ProblemFull, n: int, m1: int, m2: int) → None¶

Parameters

n – Number of unknowns
m1 – Number of ALM constraints
m2 – Number of quadratic penalty constraints

property C¶: Box constraints on \(x\)

property D1¶: Box constraints on \(g1(x)\)

property D2¶: Box constraints on \(g2(x)\)

property f¶: Objective funcion, \(f(x)\)

property g1¶: Constraint function, \(g1(x)\)

property g2¶: Constraint function, \(g2(x)\)

property grad_f¶: Gradient of the objective function, \(\nabla f(x)\)

property grad_g1_prod¶: Gradient of constraint function times vector, \(\nabla g1(x)\, v\)

property grad_g1i¶: Gradient vector of the \(i\)-th component of the constraint function, \(\nabla g1_i(x)\)

property grad_g2_prod¶: Gradient of constraint function times vector, \(\nabla g2(x)\, v\)

property grad_g2i¶: Gradient vector of the \(i\)-th component of the constraint function, \(\nabla g2_i(x)\)

property hess_L¶: Hessian of the Lagrangian function, \(\nabla^2_{xx} L(x,y)\)

property hess_L_prod¶: Hessian of the Lagrangian function times vector, \(\nabla^2_{xx} L(x,y)\, v\)

property m1¶: Number of general constraints, dimension of \(g1(x)\)

property m2¶: Number of general constraints, dimension of \(g2(x)\)

property n¶: Number of unknowns, dimension of \(x\)

class panocpy._panocpy.ProblemFullWithParam¶

C++ documentation: pa::ProblemFullWithParam

__init__(self: panocpy._panocpy.ProblemFullWithParam, n: int, m1: int, m2: int) → None¶

Parameters

n – Number of unknowns
m1 – Number of ALM constraints
m2 – Number of quadratic penalty constraints

property C¶: Box constraints on \(x\)

property D1¶: Box constraints on \(g1(x)\)

property D2¶: Box constraints on \(g2(x)\)

property f¶: Objective funcion, \(f(x)\)

property g1¶: Constraint function, \(g1(x)\)

property g2¶: Constraint function, \(g2(x)\)

property grad_f¶: Gradient of the objective function, \(\nabla f(x)\)

property grad_g1_prod¶: Gradient of constraint function times vector, \(\nabla g1(x)\, v\)

property grad_g1i¶: Gradient vector of the \(i\)-th component of the constraint function, \(\nabla g1_i(x)\)

property grad_g2_prod¶: Gradient of constraint function times vector, \(\nabla g2(x)\, v\)

property grad_g2i¶: Gradient vector of the \(i\)-th component of the constraint function, \(\nabla g2_i(x)\)

property hess_L¶: Hessian of the Lagrangian function, \(\nabla^2_{xx} L(x,y)\)

property hess_L_prod¶: Hessian of the Lagrangian function times vector, \(\nabla^2_{xx} L(x,y)\, v\)

property m1¶: Number of general constraints, dimension of \(g1(x)\)

property m2¶: Number of general constraints, dimension of \(g2(x)\)

property n¶: Number of unknowns, dimension of \(x\)

property param¶: Parameter vector \(p\) of the problem

class panocpy._panocpy.ProblemWithParam¶

C++ documentation: pa::ProblemWithParam

See panocpy._panocpy.Problem for the full documentation.

__init__(self: panocpy._panocpy.ProblemWithParam, n: int, m: int) → None¶

property C¶

property D¶

property f¶

property g¶

property grad_f¶

property grad_g_prod¶

property grad_gi¶

property hess_L¶

property hess_L_prod¶

property m¶

property n¶

property param¶: Parameter vector \(p\) of the problem

class panocpy._panocpy.SolverStatus¶

C++ documentation: pa::SolverStatus

Members:

Unknown : Initial value

Converged : Converged and reached given tolerance

MaxTime : Maximum allowed execution time exceeded

MaxIter : Maximum number of iterations exceeded

NotFinite : Intermediate results were infinite or not-a-number

NoProgress : No progress was made in the last iteration

Interrupted : Solver was interrupted by the user

__init__(self: panocpy._panocpy.SolverStatus, value: int) → None¶

Converged = <SolverStatus.Converged: 1>¶

Interrupted = <SolverStatus.Interrupted: 6>¶

MaxIter = <SolverStatus.MaxIter: 3>¶

MaxTime = <SolverStatus.MaxTime: 2>¶

NoProgress = <SolverStatus.NoProgress: 5>¶

NotFinite = <SolverStatus.NotFinite: 4>¶

Unknown = <SolverStatus.Unknown: 0>¶

property name¶

property value¶

class panocpy._panocpy.StructuredPANOCLBFGSParams¶

C++ documentation: pa::StructuredPANOCLBFGSParams

__init__(self: panocpy._panocpy.StructuredPANOCLBFGSParams, **kwargs) → None¶

to_dict(self: panocpy._panocpy.StructuredPANOCLBFGSParams) → dict¶

property L_max¶

property L_min¶

property Lipschitz¶

property alternative_linesearch_cond¶

property full_augmented_hessian¶

property hessian_vec_finited_differences¶

property lbfgs_stepsize¶

property max_iter¶

property max_no_progress¶

property max_time¶

property nonmonotone_linesearch¶

property print_interval¶

property quadratic_upperbound_tolerance_factor¶

property stop_crit¶

property update_lipschitz_in_linesearch¶

property τ_min¶

class panocpy._panocpy.StructuredPANOCLBFGSProgressInfo¶

Data passed to the structured PANOC progress callback.

C++ documentation: pa::StructuredPANOCLBFGSProgressInfo

__init__(*args, **kwargs)¶

property L¶

property fpr¶

property grad_ψ¶

property grad_ψ_hat¶

property k¶

property norm_sq_p¶

property p¶

property x¶

property x_hat¶

property y¶

property Σ¶

property γ¶

property ε¶

property τ¶

property φγ¶

property ψ¶

property ψ_hat¶

class panocpy._panocpy.StructuredPANOCLBFGSSolver¶

C++ documentation: pa::StructuredPANOCLBFGSSolver

Solve.

Parameters

problem – Problem to solve
Σ – Penalty factor
ε – Desired tolerance
x – Initial guess
y – Initial Lagrange multipliers

Returns

Solution \(x\)
Updated Lagrange multipliers \(y\)
Slack variable error \(g(x) - z\)
Statistics

__init__(*args, **kwargs)¶

Overloaded function.

__init__(self: panocpy._panocpy.StructuredPANOCLBFGSSolver) -> None
__init__(self: panocpy._panocpy.StructuredPANOCLBFGSSolver, panoc_params: panocpy._panocpy.StructuredPANOCLBFGSParams, lbfgs_params: panocpy._panocpy.LBFGSParams) -> None

set_progress_callback(self: panocpy._panocpy.StructuredPANOCLBFGSSolver, callback: Callable[[panocpy._panocpy.StructuredPANOCLBFGSProgressInfo], None]) → None¶: Attach a callback that is called on each iteration of the solver.

property params¶

panocpy._panocpy.load_casadi_problem(so_name: str, n: int, m: int, second_order: bool = False) → panocpy._panocpy.Problem¶

Load a compiled CasADi problem without parameters.

C++ documentation: pa::load_CasADi_problem()

panocpy._panocpy.load_casadi_problem_full(so_name: str, n: int, m1: int, m2: int, second_order: bool = False) → panocpy._panocpy.ProblemFull¶

Load a compiled CasADi problem without parameters.

C++ documentation: pa::load_CasADi_problem()

panocpy._panocpy.load_casadi_problem_full_with_param(so_name: str, n: int, m1: int, m2: int, second_order: bool = False) → panocpy._panocpy.ProblemFullWithParam¶

Load a compiled CasADi problem with parameters.

C++ documentation: pa::load_CasADi_problem_with_param()

panocpy._panocpy.load_casadi_problem_with_param(so_name: str, n: int, m: int, second_order: bool = False) → panocpy._panocpy.ProblemWithParam¶

Load a compiled CasADi problem with parameters.

C++ documentation: pa::load_CasADi_problem_with_param()

panocpy._panocpy.panoc(ψ: Callable[[numpy.ndarray[numpy.float64[m, 1]]], float], grad_ψ: Callable[[numpy.ndarray[numpy.float64[m, 1]]], numpy.ndarray[numpy.float64[m, 1]]], C: panocpy._panocpy.Box, x0: Optional[numpy.ndarray[numpy.float64[m, 1]]] = None, ε: float = 1e-08, params: panocpy._panocpy.PANOCParams = <panocpy._panocpy.PANOCParams object at 0x7f059bce38b0>, lbfgs_params: panocpy._panocpy.LBFGSParams = <panocpy._panocpy.LBFGSParams object at 0x7f059bce3c70>) → Tuple[numpy.ndarray[numpy.float64[m, 1]], dict]¶

Compile the C-code using the given code-generator and load it as a panocpy Problem.

Parameters

cgen – Code generator to generate C-code for the costs and the constraints with.
n – Dimensions of the decision variables (primal dimension).
m – Number of nonlinear constraints (dual dimension).
p – Number of parameters.
name – Optional string description of the problem (used for filename).

Returns

Problem specification that can be passed to the solvers.

Compile the C-code using the given code-generator and load it as a panocpy ProblemFull.

Parameters

cgen – Code generator to generate C-code for the costs and the constraints with.
n – Dimensions of the decision variables (primal dimension).
m1 – Number of ALM constraints (dual dimension 1).
m2 – Number of quadratic penalty constraints (dual dimension 2).
p – Number of parameters.
name – Optional string description of the problem (used for filename).

Returns

ProblemFull specification that can be passed to the solvers.

Compile the objective and constraint functions into a panocpy Problem.

Parameters

f – Objective function.
g – Constraint function.
second_order – Whether to generate functions for evaluating Hessians.
name – Optional string description of the problem (used for filename).

Returns

Problem specification that can be passed to the solvers.

Convert the objective and constraint functions into a CasADi code generator.

Parameters

f – Objective function.
g – Constraint function.
second_order – Whether to generate functions for evaluating Hessians.
name – Optional string description of the problem (used for filename).

Returns

Code generator that generates the functions and derivatives used by the solvers.
Dimensions of the decision variables (primal dimension).
Number of nonlinear constraints (dual dimension).
Number of parameters.

Convert the objective and constraint functions into a CasADi code generator.

Parameters

f – Objective function.
g1 – ALM constraint function.
g2 – Quadratic penalty constraint function.
second_order – Whether to generate functions for evaluating Hessians.
name – Optional string description of the problem (used for filename).

Returns

Code generator that generates the functions and derivatives used by the solvers.
Dimensions of the decision variables (primal dimension).
Number of nonlinear constraints (dual dimension).
Number of parameters.