import typing
from collections.abc import Callable, Iterable
from typing import Any
import numpy as np
from .fitness_normalizer import FitnessNormalizer
T = typing.TypeVar("T")
[docs]
class Power(FitnessNormalizer[T]):
"""
Raises fitness values to some power.
Examples:
*Raising Fitness Values to a Power*
Sometimes you might calculate a property for a molecule, where
that property indicates a low fitness value. You can use
:class:`.Power` to raise it to the power of -1 to get your
final fitness value
.. testcode:: raising-fitness-values-to-a-power
import stk
building_block = stk.BuildingBlock('BrCCBr', stk.BromoFactory())
record1 = stk.MoleculeRecord(
topology_graph=stk.polymer.Linear(
building_blocks=[building_block],
repeating_unit='A',
num_repeating_units=2,
),
)
record2 = stk.MoleculeRecord(
topology_graph=stk.polymer.Linear(
building_blocks=[building_block],
repeating_unit='A',
num_repeating_units=2,
),
)
fitness_values = {
record1: 1,
record2: 2,
}
normalizer = stk.Power(-1)
normalized_fitness_values = normalizer.normalize(fitness_values)
assert normalized_fitness_values[record1] == 1
assert normalized_fitness_values[record2] == 0.5
*Raising Fitness Values by a Set of Powers*
In this example, assume that each fitness value consists of a
:class:`tuple` of numbers, each representing a different property
of the molecule, and each contributing to the final fitness value.
The properties can be anything, such as energy, number of atoms
or diameter.
If your final fitness value depends on the combination of these
properties, you will probably first want to scale them with
:class:`.DivideByMean`. Once this is done, you may want to
raise each property by some power. For example
if you raise the value of one property by ``1`` and another
by ``-1``, it means that when the value of property 1 is big,
the fitness value should also be big, but if the value of property
2 is big, the fitness value should be small.
Giving a concrete example
.. testcode:: raising-fitness-values-by-a-set-of-powers
import stk
import numpy as np
building_block = stk.BuildingBlock('BrCCBr', stk.BromoFactory())
record = stk.MoleculeRecord(
topology_graph=stk.polymer.Linear(
building_blocks=[building_block],
repeating_unit='A',
num_repeating_units=2,
),
)
fitness_values = {
record: (2, 2, 2),
}
normalizer = stk.Power((1, -1, 2))
normalized_fitness_values = normalizer.normalize(fitness_values)
assert np.all(np.equal(
normalized_fitness_values[record],
(2, 0.5, 4),
))
*Selectively Normalizing Fitness Values*
Sometimes, you only want to normalize some members of a population,
for example if some do not have an assigned fitness value,
because the fitness calculation failed for whatever reason.
You can use the `filter` parameter to exclude records from the
normalization
.. testcode:: selectively-normalizing-fitness-values
import stk
import numpy as np
building_block = stk.BuildingBlock('BrCCBr', stk.BromoFactory())
record1 = stk.MoleculeRecord(
topology_graph=stk.polymer.Linear(
building_blocks=[building_block],
repeating_unit='A',
num_repeating_units=2,
),
)
record2 = stk.MoleculeRecord(
topology_graph=stk.polymer.Linear(
building_blocks=[building_block],
repeating_unit='A',
num_repeating_units=2,
),
)
fitness_values = {
record1: (2, 2, 2),
record2: None,
}
normalizer = stk.Power(
power=(1, -1, 2),
# Only normalize values which are not None.
filter=lambda fitness_values, record:
fitness_values[record] is not None,
)
normalized_fitness_values = normalizer.normalize(fitness_values)
assert np.all(np.equal(
normalized_fitness_values[record1],
(2, 0.5, 4),
))
assert normalized_fitness_values[record2] is None
"""
def __init__(
self,
power: float | Iterable[float],
filter: Callable[[dict[T, Any], T], bool] = lambda fitness_values,
record: True,
) -> None:
"""
Parameters:
power (float | list[float]):
The power each fitness value is raised to. Can
be a single number or multiple numbers, depending on the
form of the fitness value.
filter:
A function which returns ``True`` or ``False``. Only
molecules which return ``True`` will have fitness values
normalized. By default, all molecules will have fitness
values normalized.
The instance passed to the `fitness_values` argument of
:meth:`.normalize` is passed as the first argument, while
the second argument will be passed every
:class:`.MoleculeRecord` in it, one at a time.
"""
if not isinstance(power, int | float):
power = tuple(power)
self._power = power
self._filter = filter
[docs]
def normalize(self, fitness_values: dict[T, Any]) -> dict[T, Any]:
return {
record: (
np.float_power(fitness_value, self._power)
if self._filter(fitness_values, record)
else fitness_value
)
for record, fitness_value in fitness_values.items()
}