import numpy as np

from neural_net.activation_layers.activation_layer import ActivationLayer
from neural_net.functions.activation import sigmoid_derivative_activation


class SigmoidLayer(ActivationLayer):
    def __init__(self, input_dim, output_dim, weights=None, biases=None):
        super().__init__(input_dim, output_dim, weights, biases)
        self.subtype = 'Sigmoid'

    def initialize_weights(self):
        # Xavier initialization for sigmoid activation
        limit = np.sqrt(6 / (self.input_dim + self.output_dim))
        self.weights = np.random.uniform(-limit, limit, (self.input_dim, self.output_dim))

    def initialize_biases(self):
        self.biases = np.zeros((1, self.output_dim))  # Biases initialized to zero

    def activation(self, outputs: np.array):
        return sigmoid_derivative_activation(outputs)

    def activation_derivative(self, outputs: np.array):
        return sigmoid_derivative_activation(outputs)