"""Base history class."""
import numbers
import time
from abc import ABC, abstractmethod
from typing import Dict, Sequence, Tuple, Union
import numpy as np
from ..C import (
FVAL,
GRAD,
HESS,
MODE_FUN,
MODE_RES,
RES,
SRES,
TIME,
ModeType,
X,
)
from ..util import (
fval_to_chi2,
grad_to_schi2,
res_to_fval,
sres_to_fim,
sres_to_grad,
)
from .options import HistoryOptions
from .util import MaybeArray, ResultDict
[docs]class HistoryBase(ABC):
"""Abstract base class for histories."""
# values calculated by the objective function
RESULT_KEYS = (FVAL, GRAD, HESS, RES, SRES)
# all possible history entries
ALL_KEYS = (X, *RESULT_KEYS, TIME)
[docs] def __init__(self, options: HistoryOptions = None):
if options is None:
options = HistoryOptions()
options = HistoryOptions.assert_instance(options)
self.options: HistoryOptions = options
[docs] @abstractmethod
def update(
self,
x: np.ndarray,
sensi_orders: Tuple[int, ...],
mode: ModeType,
result: ResultDict,
) -> None:
"""
Update history after a function evaluation.
Parameters
----------
x:
The parameter vector.
sensi_orders:
The sensitivity orders computed.
mode:
The objective function mode computed (function value or residuals).
result:
The objective function values for parameters `x`, sensitivities
`sensi_orders` and mode `mode`.
"""
[docs] def finalize(
self,
message: str = None,
exitflag: str = None,
) -> None:
"""
Finalize history. Called after a run. Default: Do nothing.
Parameters
----------
message:
Optimizer message to be saved.
exitflag:
Optimizer exitflag to be saved.
"""
@abstractmethod
def __len__(self) -> int:
"""Define length by number of stored entries in the history."""
@property
@abstractmethod
def n_fval(self) -> int:
"""Return number of function evaluations."""
@property
@abstractmethod
def n_grad(self) -> int:
"""Return number of gradient evaluations."""
@property
@abstractmethod
def n_hess(self) -> int:
"""Return number of Hessian evaluations."""
@property
@abstractmethod
def n_res(self) -> int:
"""Return number of residual evaluations."""
@property
@abstractmethod
def n_sres(self) -> int:
"""Return number or residual sensitivity evaluations."""
@property
@abstractmethod
def start_time(self) -> float:
"""Return start time."""
[docs] @abstractmethod
def get_x_trace(
self,
ix: Union[int, Sequence[int], None] = None,
trim: bool = False,
) -> Union[Sequence[np.ndarray], np.ndarray]:
"""
Return parameters.
Takes as parameter an index or indices and returns corresponding trace
values. If only a single value is requested, the list is flattened.
"""
[docs] @abstractmethod
def get_fval_trace(
self,
ix: Union[int, Sequence[int], None] = None,
trim: bool = False,
) -> Union[Sequence[float], float]:
"""
Return function values.
Takes as parameter an index or indices and returns corresponding trace
values. If only a single value is requested, the list is flattened.
"""
[docs] @abstractmethod
def get_grad_trace(
self,
ix: Union[int, Sequence[int], None] = None,
trim: bool = False,
) -> Union[Sequence[MaybeArray], MaybeArray]:
"""
Return gradients.
Takes as parameter an index or indices and returns corresponding trace
values. If only a single value is requested, the list is flattened.
"""
[docs] @abstractmethod
def get_hess_trace(
self,
ix: Union[int, Sequence[int], None] = None,
trim: bool = False,
) -> Union[Sequence[MaybeArray], MaybeArray]:
"""
Return hessians.
Takes as parameter an index or indices and returns corresponding trace
values. If only a single value is requested, the list is flattened.
"""
[docs] @abstractmethod
def get_res_trace(
self,
ix: Union[int, Sequence[int], None] = None,
trim: bool = False,
) -> Union[Sequence[MaybeArray], MaybeArray]:
"""
Residuals.
Takes as parameter an index or indices and returns corresponding trace
values. If only a single value is requested, the list is flattened.
"""
[docs] @abstractmethod
def get_sres_trace(
self,
ix: Union[int, Sequence[int], None] = None,
trim: bool = False,
) -> Union[Sequence[MaybeArray], MaybeArray]:
"""
Residual sensitivities.
Takes as parameter an index or indices and returns corresponding trace
values. If only a single value is requested, the list is flattened.
"""
[docs] def get_chi2_trace(
self,
ix: Union[int, Sequence[int], None] = None,
trim: bool = False,
) -> Union[Sequence[float], float]:
"""
Chi2 values.
Takes as parameter an index or indices and returns corresponding trace
values. If only a single value is requested, the list is flattened.
"""
fval_trace = self.get_fval_trace(ix, trim)
reduce = isinstance(ix, numbers.Integral)
if reduce:
return fval_to_chi2(fval_trace)
else:
return [fval_to_chi2(fval) for fval in fval_trace]
[docs] def get_schi2_trace(
self,
ix: Union[int, Sequence[int], None] = None,
trim: bool = False,
) -> Union[Sequence[MaybeArray], MaybeArray]:
"""
Chi2 sensitivities.
Takes as parameter an index or indices and returns corresponding trace
values. If only a single value is requested, the list is flattened.
"""
grad_trace = self.get_grad_trace(ix, trim)
reduce = isinstance(ix, numbers.Integral)
if reduce:
return grad_to_schi2(grad_trace)
else:
return [grad_to_schi2(grad) for grad in grad_trace]
[docs] @abstractmethod
def get_time_trace(
self,
ix: Union[int, Sequence[int], None] = None,
trim: bool = False,
) -> Union[Sequence[float], float]:
"""
Cumulative execution times.
Takes as parameter an index or indices and returns corresponding trace
values. If only a single value is requested, the list is flattened.
"""
raise NotImplementedError()
[docs] def get_trimmed_indices(self) -> np.ndarray:
"""Get indices for a monotonically decreasing history."""
fval_trace = self.get_fval_trace()
return np.where(fval_trace <= np.fmin.accumulate(fval_trace))[0]
[docs] def implements_trace(self) -> bool:
"""Check whether the history has a trace that can be queried."""
try:
self.get_fval_trace()
except NotImplementedError:
return False
return True
[docs]class NoHistory(HistoryBase):
"""Dummy history that does not do anything.
Can be used whenever a history object is needed, but no history is desired.
Can be created, but not queried.
"""
[docs] def update( # noqa: D102
self,
x: np.ndarray,
sensi_orders: Tuple[int, ...],
mode: ModeType,
result: ResultDict,
) -> None:
pass
def __len__(self) -> int: # noqa: D102
raise NotImplementedError()
@property
def n_fval(self) -> int: # noqa: D102
raise NotImplementedError()
@property
def n_grad(self) -> int: # noqa: D102
raise NotImplementedError()
@property
def n_hess(self) -> int: # noqa: D102
raise NotImplementedError()
@property
def n_res(self) -> int: # noqa: D102
raise NotImplementedError()
@property
def n_sres(self) -> int: # noqa: D102
raise NotImplementedError()
@property
def start_time(self) -> float: # noqa: D102
raise NotImplementedError()
[docs] def get_x_trace( # noqa: D102
self, ix: Union[int, Sequence[int], None] = None, trim: bool = False
) -> Union[Sequence[np.ndarray], np.ndarray]:
raise NotImplementedError()
[docs] def get_fval_trace( # noqa: D102
self, ix: Union[int, Sequence[int], None] = None, trim: bool = False
) -> Union[Sequence[float], float]:
raise NotImplementedError()
[docs] def get_grad_trace( # noqa: D102
self, ix: Union[int, Sequence[int], None] = None, trim: bool = False
) -> Union[Sequence[MaybeArray], MaybeArray]:
raise NotImplementedError()
[docs] def get_hess_trace( # noqa: D102
self, ix: Union[int, Sequence[int], None] = None, trim: bool = False
) -> Union[Sequence[MaybeArray], MaybeArray]:
raise NotImplementedError()
[docs] def get_res_trace( # noqa: D102
self, ix: Union[int, Sequence[int], None] = None, trim: bool = False
) -> Union[Sequence[MaybeArray], MaybeArray]:
raise NotImplementedError()
[docs] def get_sres_trace( # noqa: D102
self, ix: Union[int, Sequence[int], None] = None, trim: bool = False
) -> Union[Sequence[MaybeArray], MaybeArray]:
raise NotImplementedError()
[docs] def get_time_trace( # noqa: D102
self, ix: Union[int, Sequence[int], None] = None, trim: bool = False
) -> Union[Sequence[float], float]:
raise NotImplementedError()
[docs]class CountHistoryBase(HistoryBase):
"""Abstract class tracking counts of function evaluations.
Needs a separate implementation of trace.
"""
[docs] def __init__(self, options: Union[HistoryOptions, Dict] = None):
super().__init__(options)
self._n_fval: int = 0
self._n_grad: int = 0
self._n_hess: int = 0
self._n_res: int = 0
self._n_sres: int = 0
self._start_time: float = time.time()
[docs] def update( # noqa: D102
self,
x: np.ndarray,
sensi_orders: Tuple[int, ...],
mode: ModeType,
result: ResultDict,
) -> None:
self._update_counts(sensi_orders, mode)
def _update_counts(
self,
sensi_orders: Tuple[int, ...],
mode: ModeType,
):
"""Update the counters."""
if mode == MODE_FUN:
if 0 in sensi_orders:
self._n_fval += 1
if 1 in sensi_orders:
self._n_grad += 1
if 2 in sensi_orders:
self._n_hess += 1
elif mode == MODE_RES:
if 0 in sensi_orders:
self._n_res += 1
if 1 in sensi_orders:
self._n_sres += 1
@property
def n_fval(self) -> int: # noqa: D102
return self._n_fval
@property
def n_grad(self) -> int: # noqa: D102
return self._n_grad
@property
def n_hess(self) -> int: # noqa: D102
return self._n_hess
@property
def n_res(self) -> int: # noqa: D102
return self._n_res
@property
def n_sres(self) -> int: # noqa: D102
return self._n_sres
@property
def start_time(self) -> float: # noqa: D102
return self._start_time
[docs]class CountHistory(CountHistoryBase):
"""History that can only count, other functions cannot be invoked."""
def __len__(self) -> int: # noqa: D102
raise NotImplementedError()
[docs] def get_x_trace( # noqa: D102
self, ix: Union[int, Sequence[int], None] = None, trim: bool = False
) -> Union[Sequence[np.ndarray], np.ndarray]:
raise NotImplementedError()
[docs] def get_fval_trace( # noqa: D102
self, ix: Union[int, Sequence[int], None] = None, trim: bool = False
) -> Union[Sequence[float], float]:
raise NotImplementedError()
[docs] def get_grad_trace( # noqa: D102
self, ix: Union[int, Sequence[int], None] = None, trim: bool = False
) -> Union[Sequence[MaybeArray], MaybeArray]:
raise NotImplementedError()
[docs] def get_hess_trace( # noqa: D102
self, ix: Union[int, Sequence[int], None] = None, trim: bool = False
) -> Union[Sequence[MaybeArray], MaybeArray]:
raise NotImplementedError()
[docs] def get_res_trace( # noqa: D102
self, ix: Union[int, Sequence[int], None] = None, trim: bool = False
) -> Union[Sequence[MaybeArray], MaybeArray]:
raise NotImplementedError()
[docs] def get_sres_trace( # noqa: D102
self, ix: Union[int, Sequence[int], None] = None, trim: bool = False
) -> Union[Sequence[MaybeArray], MaybeArray]:
raise NotImplementedError()
[docs] def get_time_trace( # noqa: D102
self, ix: Union[int, Sequence[int], None] = None, trim: bool = False
) -> Union[Sequence[float], float]:
raise NotImplementedError()
def add_fun_from_res(result: ResultDict) -> ResultDict:
"""Calculate function values from residual values.
Copies the result, but apart performs calculations only if entries
are not present yet in the result object
(thus can be called repeatedly).
Parameters
----------
result: Result dictionary from the objective function.
Returns
-------
full_result:
Result dicionary, adding whatever is possible to calculate.
"""
result = result.copy()
# calculate function values from residuals
if result.get(FVAL) is None:
result[FVAL] = res_to_fval(result.get(RES))
if result.get(GRAD) is None:
result[GRAD] = sres_to_grad(result.get(RES), result.get(SRES))
if result.get(HESS) is None:
result[HESS] = sres_to_fim(result.get(SRES))
return result
def reduce_result_via_options(
result: ResultDict, options: HistoryOptions
) -> ResultDict:
"""Set values not to be stored in history or missing to NaN.
Parameters
----------
result:
Result dictionary with all fields present.
options:
History options.
Returns
-------
result:
Result reduced to what is intended to be stored in history.
"""
result = result.copy()
# apply options to result
for key in HistoryBase.RESULT_KEYS:
if result.get(key) is None or not options.get(
f'trace_record_{key}', True
):
result[key] = np.nan
return result