Python的单个神经元神经网络

神经网络是深度学习的核心，深度学习是一个在许多不同领域都有实际应用的领域。如今，神经网络已用于图像分类，语音识别，对象检测等。现在，让我们尝试了解所有这些最新技术背后的基本单位。

单个神经元将给定的输入转换为某些输出。根据给定的输入和分配给每个输入的权重，确定神经元是否触发。假设神经元具有3个输入连接和1个输出。

在给定的示例中，我们将使用tanh激活函数。

最终目标是为该神经元找到最佳的权重集，以产生正确的结果。通过使用几个不同的训练示例训练神经元来做到这一点。在每个步骤中，计算神经元输出中的误差，然后反向传播梯度。计算神经元输出的步骤称为正向传播，而计算梯度的步骤称为反向传播。

下面是实现：

# Python program to implement a 
# single neuron neural network
  
# import all necessery libraries
from numpy import exp, array, random, dot, tanh
  
# Class to create a neural 
# network with single neuron
class NeuralNetwork():
      
    def __init__(self):
          
        # Using seed to make sure it'll  
        # generate same weights in every run
        random.seed(1)
          
        # 3x1 Weight matrix
        self.weight_matrix = 2 * random.random((3, 1)) - 1
  
    # tanh as activation function
    def tanh(self, x):
        return tanh(x)
  
    # derivative of tanh function.
    # Needed to calculate the gradients.
    def tanh_derivative(self, x):
        return 1.0 - tanh(x) ** 2
  
    # forward propagation
    def forward_propagation(self, inputs):
        return self.tanh(dot(inputs, self.weight_matrix))
      
    # training the neural network.
    def train(self, train_inputs, train_outputs,
                            num_train_iterations):
                                  
        # Number of iterations we want to
        # perform for this set of input.
        for iteration in range(num_train_iterations):
            output = self.forward_propagation(train_inputs)
  
            # Calculate the error in the output.
            error = train_outputs - output
  
            # multiply the error by input and then 
            # by gradient of tanh funtion to calculate
            # the adjustment needs to be made in weights
            adjustment = dot(train_inputs.T, error *
                             self.tanh_derivative(output))
                               
            # Adjust the weight matrix
            self.weight_matrix += adjustment
  
# Driver Code
if __name__ == "__main__":
      
    neural_network = NeuralNetwork()
      
    print ('Random weights at the start of training')
    print (neural_network.weight_matrix)
  
    train_inputs = array([[0, 0, 1], [1, 1, 1], [1, 0, 1], [0, 1, 1]])
    train_outputs = array([[0, 1, 1, 0]]).T
  
    neural_network.train(train_inputs, train_outputs, 10000)
  
    print ('New weights after training')
    print (neural_network.weight_matrix)
  
    # Test the neural network with a new situation.
    print ("Testing network on new examples ->")
    print (neural_network.forward_propagation(array([1, 0, 0])))

输出 ：

Random weights at the start of training
[[-0.16595599]
 [ 0.44064899]
 [-0.99977125]]

New weights after training
[[5.39428067]
 [0.19482422]
 [0.34317086]]

Testing network on new examples ->
[0.99995873]