jaxopt.BoxCDQP

class jaxopt.BoxCDQP(maxiter=500, tol=0.0001, verbose=0, implicit_diff=True, implicit_diff_solve=None, jit='auto', unroll='auto')[source]

Coordinate descent solver for box-constrained QPs.

This solver minimizes:

0.5 <x, Qx> + <c, x> subject to l <= x <= u
Parameters

  • maxiter (int) –

  • tol (float) –

  • verbose (int) –

  • implicit_diff (bool) –

  • implicit_diff_solve (Optional[Callable]) –

  • jit (Union[str, bool]) –

  • unroll (Union[str, bool]) –

maxiter

maximum number of coordinate descent iterations.

Type

int

tol

tolerance to use.

Type

float

verbose

whether to print error on every iteration or not. Warning: verbose=True will automatically disable jit.

Type

int

implicit_diff

whether to enable implicit diff or autodiff of unrolled iterations.

Type

bool

implicit_diff_solve

the linear system solver to use.

Type

Optional[Callable]

jit

whether to JIT-compile the optimization loop (default: “auto”).

Type

Union[str, bool]

unroll

whether to unroll the optimization loop (default: “auto”).

Type

Union[str, bool]

__init__(maxiter=500, tol=0.0001, verbose=0, implicit_diff=True, implicit_diff_solve=None, jit='auto', unroll='auto')
Parameters

  • maxiter (int) –

  • tol (float) –

  • verbose (int) –

  • implicit_diff (bool) –

  • implicit_diff_solve (Optional[Callable]) –

  • jit (Union[str, bool]) –

  • unroll (Union[str, bool]) –

Return type

None

Methods

__init__([maxiter, tol, verbose, ...])

attribute_names()

attribute_values()

init_state(init_params[, params_obj, ...])

Initialize the solver state.

l2_optimality_error(params, *args, **kwargs)

Computes the L2 optimality error.

optimality_fun(sol, params_obj, params_ineq)

rtype

Array

run(init_params, *args, **kwargs)

Runs the optimization loop.

update(params, state, params_obj, params_ineq)

Performs one epoch of coordinate descent.

Attributes

implicit_diff

implicit_diff_solve

jit

maxiter

tol

unroll

verbose

init_state(init_params, params_obj=None, params_ineq=None)[source]

Initialize the solver state.

Parameters

  • init_params (Array) – array containing the initial parameters.

  • params_obj (Optional[Tuple[Array, Array]]) – Tuple of arrays (Q, c).

  • params_ineq (Optional[Tuple[Array, Array]]) – Tuple of arrays (l, u).

Return type

BoxCDQPState

Returns

state

l2_optimality_error(params, *args, **kwargs)

Computes the L2 optimality error.

run(init_params, *args, **kwargs)

Runs the optimization loop.

Parameters

  • init_params (Any) – pytree containing the initial parameters.

  • *args – additional positional arguments to be passed to the update method.

  • **kwargs – additional keyword arguments to be passed to the update method.

Return type

OptStep

Returns

(params, state)

update(params, state, params_obj, params_ineq)[source]

Performs one epoch of coordinate descent.

Parameters

  • params (Array) – array containing the parameters.

  • state (NamedTuple) – named tuple containing the solver state.

  • params_obj (Tuple[Array, Array]) – Tuple of arrays (Q, c).

  • params_ineq (Tuple[Array, Array]) – Tuple of arrays (l, u).

Return type

OptStep

Returns

(params, state)