core.bound.AbstractBound 

 1from abc import ABC, abstractmethod
 2
 3from torch import Tensor
 4
 5
 6class AbstractBound(ABC):
 7    """
 8    Abstract PAC bound class for evaluating risk certificates.
 9
10    Args:
11        bound_delta (float): Confidence level over random data samples.
12            It represents the probability that the upper bound of the PAC bound holds.
13        loss_delta (float): Confidence level over random weight samples.
14            It represents the probability that the upper bound of empirical loss holds.
15
16    Overall probability is (1 - loss_bound) - bound_delta.
17
18    Attributes:
19        _bound_delta (float): Confidence level over random data samples.
20        _loss_delta (float): Confidence level over random weight samples.
21    """
22
23    def __init__(self, bound_delta: float, loss_delta: float):
24        self._bound_delta = bound_delta
25        self._loss_delta = loss_delta
26
27    @abstractmethod
28    def calculate(self, *args, **kwargs) -> tuple[Tensor | float, Tensor | float]:
29        """
30        Calculates the PAC Bayes bound.
31
32        Args:
33            args: Variable length argument list.
34            kwargs: Arbitrary keyword arguments.
35
36        Returns:
37            Tuple[Union[Tensor, float], Union[Tensor, float]]:
38                A tuple containing the calculated PAC bound and the upper bound of empirical risk.
39        """
40        pass
class AbstractBound(abc.ABC): 
 7class AbstractBound(ABC):
 8    """
 9    Abstract PAC bound class for evaluating risk certificates.
10
11    Args:
12        bound_delta (float): Confidence level over random data samples.
13            It represents the probability that the upper bound of the PAC bound holds.
14        loss_delta (float): Confidence level over random weight samples.
15            It represents the probability that the upper bound of empirical loss holds.
16
17    Overall probability is (1 - loss_bound) - bound_delta.
18
19    Attributes:
20        _bound_delta (float): Confidence level over random data samples.
21        _loss_delta (float): Confidence level over random weight samples.
22    """
23
24    def __init__(self, bound_delta: float, loss_delta: float):
25        self._bound_delta = bound_delta
26        self._loss_delta = loss_delta
27
28    @abstractmethod
29    def calculate(self, *args, **kwargs) -> tuple[Tensor | float, Tensor | float]:
30        """
31        Calculates the PAC Bayes bound.
32
33        Args:
34            args: Variable length argument list.
35            kwargs: Arbitrary keyword arguments.
36
37        Returns:
38            Tuple[Union[Tensor, float], Union[Tensor, float]]:
39                A tuple containing the calculated PAC bound and the upper bound of empirical risk.
40        """
41        pass

Abstract PAC bound class for evaluating risk certificates.

Arguments:
  • bound_delta (float): Confidence level over random data samples. It represents the probability that the upper bound of the PAC bound holds.
  • loss_delta (float): Confidence level over random weight samples. It represents the probability that the upper bound of empirical loss holds.

Overall probability is (1 - loss_bound) - bound_delta.

Attributes:
  • _bound_delta (float): Confidence level over random data samples.
  • _loss_delta (float): Confidence level over random weight samples.
@abstractmethod
def calculate( self, *args, **kwargs) -> tuple[torch.Tensor | float, torch.Tensor | float]: 
28    @abstractmethod
29    def calculate(self, *args, **kwargs) -> tuple[Tensor | float, Tensor | float]:
30        """
31        Calculates the PAC Bayes bound.
32
33        Args:
34            args: Variable length argument list.
35            kwargs: Arbitrary keyword arguments.
36
37        Returns:
38            Tuple[Union[Tensor, float], Union[Tensor, float]]:
39                A tuple containing the calculated PAC bound and the upper bound of empirical risk.
40        """
41        pass

Calculates the PAC Bayes bound.

Arguments:
  • args: Variable length argument list.
  • kwargs: Arbitrary keyword arguments.
Returns:

Tuple[Union[Tensor, float], Union[Tensor, float]]: A tuple containing the calculated PAC bound and the upper bound of empirical risk.