import numpy as np
from typing import Callable, Dict
[docs]class Scalarizer:
"""Implements a class for scalarizing vector valued functions with a
given scalarization function.
"""
def __init__(
self, evaluator: Callable, scalarizer: Callable, evaluator_args: Dict = None, scalarizer_args: Dict = None
):
"""
Args:
evaluator (Callable): A Callable object returning a numpy array.
scalarizer (Callable): A function which should accepts as its
arguments the output of evaluator and return a single value.
evaluator_args (Any, optional): Optional arguments to be passed to
evaluator. Defaults to None.
scalarizer_args (Any, optional): Optional arguments to be passed to
scalarizer. Defaults to None.
"""
self._evaluator = evaluator
self._scalarizer = scalarizer
self._evaluator_args = evaluator_args
self._scalarizer_args = scalarizer_args
[docs] def evaluate(self, xs: np.ndarray) -> np.ndarray:
"""Evaluates the scalarized function with the given arguments and
returns a scalar value for each vector of variables given in a numpy
array.
Args:
xs (np.ndarray): A 2D numpy array containing vectors of variables
on each of its rows.
Returns:
np.ndarray: A 1D numpy array with the values returned by the
scalarizer for each row in xs.
"""
if self._evaluator_args is not None:
res_eval = self._evaluator(xs, **self._evaluator_args)
else:
res_eval = self._evaluator(xs)
if self._scalarizer_args is not None:
res_scal = self._scalarizer(res_eval, **self._scalarizer_args)
else:
res_scal = self._scalarizer(res_eval)
return res_scal
[docs] def __call__(self, xs: np.ndarray) -> np.ndarray:
"""Wrapper to the evaluate method.
"""
return self.evaluate(xs)
[docs]class DiscreteScalarizer:
"""Implements a class to scalarize discrete vectors given a scalarizing function.
"""
def __init__(self, scalarizer: Callable, scalarizer_args: Dict = None):
self._scalarizer = scalarizer
self._scalarizer_args = scalarizer_args
[docs] def evaluate(self, vectors: np.ndarray) -> np.ndarray:
# guarantee two dimensions, makes sure works with single vectors as well.
vectors = np.atleast_2d(vectors)
if self._scalarizer_args is not None:
res_scal = self._scalarizer(vectors, **self._scalarizer_args)
else:
res_scal = self._scalarizer(vectors)
return res_scal
[docs] def __call__(self, vectors: np.ndarray):
return self.evaluate(vectors)
if __name__ == "__main__":
[docs] vectors = np.array([[1, 1, 1], [2, 2, 2], [4, 5, 6.0]])
vector = np.array([1, 2, 3])
dscalarizer = DiscreteScalarizer(lambda x, a=1: a * np.sum(x, axis=1), scalarizer_args={"a": 2})
res = dscalarizer(vectors)
res_1d = dscalarizer(vector)
print(res)
print(res_1d)
