Задача 09B. Activations ≡

Входной файл:	Стандартный вход		Ограничение времени:	1 сек
Выходной файл:	Стандартный выход		Ограничение памяти:	512 Мб
Максимальный балл:	18

Условие

Требуется реализовать на языке Python классы, описывающие различные функции активации:

import numpy as np


class Activation:
    """Base activation class"""

    def __init__(self):
        self._input = None

    @property
    def input(self):
        """Returns the last input received by the activation"""
        return self._input

    def __call__(self, x: np.ndarray) -> np.ndarray:
        """Computes activation output
        
        Arguments:
            x: Input array of shape (`batch_size`, ...)

        Returns:
            An array of the same shape as `x`"""
        raise NotImplementedError()

    def grad(self, gradOutput: np.ndarray) -> np.ndarray:
        """Computes loss gradient with respect to the activation input.
        
        Arguments:
            gradOutput: Gradient of loss function with recpect to the activation output.
                An array of the same shape as the array received in `__call__` method.

        Returns:
            An array of the same shape as `gradOutput`"""
        raise NotImplementedError()
    

class ReLU(Activation):
    """Implements ReLU activation layer"""

    def __call__(self, x: np.ndarray) -> np.ndarray:
        pass

    def grad(self, gradOutput: np.ndarray) -> np.ndarray:
        pass


class LeakyReLU(Activation):
    """Implements LeakyReLU activation layer"""

    def __init__(self, slope: float = 0.03):
        """Initializes LeakyReLU layer.

        Arguments:
            slope: the slope coeffitient of the activation."""
        pass

    def __call__(self, x: np.ndarray) -> np.ndarray:
        pass

    def grad(self, gradOutput: np.ndarray) -> np.ndarray:
        pass


class GeLU(Activation):
    """Implements GeLU activation layer"""

    def __call__(self, x: np.ndarray) -> np.ndarray:
        pass

    def grad(self, gradOutput: np.ndarray) -> np.ndarray:
        pass


class SiLU(Activation):
    """Implements SiLU (swish) activation layer"""

    def __call__(self, x: np.ndarray) -> np.ndarray:
        pass

    def grad(self, gradOutput: np.ndarray) -> np.ndarray:
        pass


class Softplus(Activation):
    """Implements Softplus (SmoothReLU) activation layer"""

    def __call__(self, x: np.ndarray) -> np.ndarray:
        pass

    def grad(self, gradOutput: np.ndarray) -> np.ndarray:
        pass


class ELU(Activation):
    """Implements ELU activation layer"""

    def __init__(self, alpha: float = 1):
        """Initializes ELU layer.

        Arguments:
            alpha: the alpha coeffitient of the activation."""
        pass

    def __call__(self, x: np.ndarray) -> np.ndarray:
        pass

    def grad(self, gradOutput: np.ndarray) -> np.ndarray:
        pass


class Sigmoid(Activation):
    """Implements Sigmoid activation layer"""

    def __call__(self, x: np.ndarray) -> np.ndarray:
        pass

    def grad(self, gradOutput: np.ndarray) -> np.ndarray:
        pass


class Tanh(Activation):
    """Implements Tanh activation layer"""

    def __call__(self, x: np.ndarray) -> np.ndarray:
        pass

    def grad(self, gradOutput: np.ndarray) -> np.ndarray:
        pass


class Softmax(Activation):
    """Implements Softmax activation layer"""

    def __call__(self, x: np.ndarray) -> np.ndarray:
        """Computes Softmax activation output
        
        Arguments:
            x: Input array of shape (`batch_size`, `n_features`)

        Returns:
            An array of the same shape as `x`"""
        pass

    def grad(self, gradOutput: np.ndarray) -> np.ndarray:
        pass

Функции активации тестируются в порядке их определения в задаче циклично, то есть activation_idx = (test_idx − 1) mod 9.

Формат выходных данных

Код решения должен содержать только импортируемые модули, определение и реализацию классов.

---