liblaf.peach.optim.objective

Classes:

• Objective –

Objective

Bases: FunctionWrapper

              flowchart TD
              liblaf.peach.optim.objective.Objective[Objective]
              liblaf.peach.functools._wrapper.FunctionWrapper[FunctionWrapper]

                              liblaf.peach.functools._wrapper.FunctionWrapper --> liblaf.peach.optim.objective.Objective
                


              click liblaf.peach.optim.objective.Objective href "" "liblaf.peach.optim.objective.Objective"
              click liblaf.peach.functools._wrapper.FunctionWrapper href "" "liblaf.peach.functools._wrapper.FunctionWrapper"
            

Parameters:

• structure

  (Structure[PyTree] | None, default: None ) –

Returned by:

• Reference Liblaf peach
  • testing  testing  rosen_objective
  • optim abc  abc  SetupResult  objective

Used by:

• Reference Liblaf peach optim
  •  optim
    •  LineSearch  search
    •  LineSearchCollisionRepulsionThreshold  search
    •  LineSearchMin  search
    •  LineSearchNaive  search
    •  LineSearchSingleNewton  search
    •  Optax
      •  init
      •  minimize
      •  postprocess
      •  step
      •  terminate
      •  update_stats
    •  Optimizer
      •  init
      •  minimize
      •  postprocess
      •  step
      •  terminate
      •  update_stats
    •  PNCG
      •  init
      •  minimize
      •  postprocess
      •  step
      •  terminate
      •  update_stats
    •  ScipyOptimizer
      •  init
      •  minimize
      •  postprocess
      •  step
      •  terminate
      •  update_stats
  • abc  abc
    •  Optimizer
      •  init
      •  minimize
      •  postprocess
      •  step
      •  terminate
      •  update_stats
    •  SetupResult
  • linesearch  linesearch
    •  LineSearch  search
    •  LineSearchCollisionRepulsionThreshold  search
    •  LineSearchMin  search
    •  LineSearchNaive  search
    •  LineSearchSingleNewton  search
  • optax  optax  Optax
    •  init
    •  minimize
    •  postprocess
    •  step
    •  terminate
    •  update_stats
  • pncg  pncg  PNCG
    •  init
    •  minimize
    •  postprocess
    •  step
    •  terminate
    •  update_stats
  • scipy  scipy  ScipyOptimizer
    •  init
    •  minimize
    •  postprocess
    •  step
    •  terminate
    •  update_stats

Methods:

• flatten –
• jit –
• partial –
• timer –
• timer_finish –
• with_aux –

Attributes:

• fun –

  X -> Scalar

• grad –

  X -> X

• grad_and_hess_diag –

  X -> X, X

• hess –

  X -> H

• hess_diag –

  X -> X

• hess_prod –

  X, P -> X

• hess_quad –

  X, P -> Scalar

• structure (Structure[PyTree] | None) –
• value_and_grad –

  X -> Scalar, X

fun class-attribute instance-attribute

fun = FunctionDescriptor(
    n_outputs=1, unflatten_inputs=(0,), flatten_outputs=()
)

X -> Scalar

grad class-attribute instance-attribute

grad = FunctionDescriptor(
    n_outputs=1, unflatten_inputs=(0,), flatten_outputs=(0,)
)

X -> X

grad_and_hess_diag class-attribute instance-attribute

grad_and_hess_diag = FunctionDescriptor(
    n_outputs=2,
    unflatten_inputs=(0,),
    flatten_outputs=(0, 1),
)

X -> X, X

hess class-attribute instance-attribute

hess = FunctionDescriptor(
    n_outputs=1, unflatten_inputs=(0,), flatten_outputs=(0,)
)

X -> H

hess_diag class-attribute instance-attribute

hess_diag = FunctionDescriptor(
    n_outputs=1, unflatten_inputs=(0,), flatten_outputs=(0,)
)

X -> X

hess_prod class-attribute instance-attribute

hess_prod = FunctionDescriptor(
    n_outputs=1,
    unflatten_inputs=(0, 1),
    flatten_outputs=(0,),
)

X, P -> X

hess_quad class-attribute instance-attribute

hess_quad = FunctionDescriptor(
    n_outputs=1, unflatten_inputs=(0, 1), flatten_outputs=()
)

X, P -> Scalar

structure class-attribute instance-attribute

structure: Structure[PyTree] | None = field(
    default=None, kw_only=True
)

value_and_grad class-attribute instance-attribute

value_and_grad = FunctionDescriptor(
    n_outputs=2, unflatten_inputs=(0,), flatten_outputs=(1,)
)

X -> Scalar, X

flatten

flatten(
    params: PyTree,
    *,
    constraints: Iterable[Constraint] = (),
) -> tuple[Self, Shaped[Array, " free"], list[Constraint]]

Source code in src/liblaf/peach/functools/_wrapper.py

def flatten(
    self, params: PyTree, *, constraints: Iterable[Constraint] = ()
) -> tuple[Self, Shaped[Array, " free"], list[Constraint]]:
    fixed_constr: list[FixedConstraint] = []
    other_constr: list[Constraint] = []
    for c in constraints:
        if isinstance(c, FixedConstraint):
            fixed_constr.append(c)
        else:
            other_constr.append(c)
    if len(fixed_constr) > 1:
        raise NotImplementedError
    fixed_mask: PyTree | None = fixed_constr[0].mask if fixed_constr else None
    params_flat: Shaped[Array, " free"]
    structure: Structure[PyTree]
    params_flat, structure = tree.flatten(params, fixed_mask=fixed_mask)
    self_new: Self = attrs.evolve(self, flatten=True, structure=structure)
    constr_flat: list[Constraint] = [
        constr.flatten(structure) for constr in other_constr
    ]
    return self_new, params_flat, constr_flat

jit

jit(enable: bool = True) -> Self

Source code in src/liblaf/peach/functools/_wrapper.py

def jit(self, enable: bool = True) -> Self:  # noqa: FBT001, FBT002
    return attrs.evolve(self, jit=enable)

partial

partial(*args: Any, **kwargs: Any) -> Self

Source code in src/liblaf/peach/functools/_wrapper.py

def partial(self, *args: Any, **kwargs: Any) -> Self:
    return attrs.evolve(
        self, args=(*self._args, *args), kwargs={**self._kwargs, **kwargs}
    )

timer

timer(enable: bool = True) -> Self

Source code in src/liblaf/peach/functools/_wrapper.py

def timer(self, enable: bool = True) -> Self:  # noqa: FBT001, FBT002
    return attrs.evolve(self, timer=enable)

timer_finish

timer_finish() -> None

Source code in src/liblaf/peach/functools/_wrapper.py

def timer_finish(self) -> None:
    _logging_hide = True
    for name, member in inspect.getmembers(self):
        if name.startswith("_"):
            continue
        timer: grapes.BaseTimer | None = grapes.get_timer(member, None)
        if timer is not None and len(timer) > 0:
            timer.finish()

with_aux

with_aux(enable: bool = True) -> Self

Source code in src/liblaf/peach/functools/_wrapper.py

def with_aux(self, enable: bool = True) -> Self:  # noqa: FBT001, FBT002
    return attrs.evolve(self, with_aux=enable)