Skip to content
TheChatArchiveThe Chat Archive
gpt-4-0314· openaiHugging Face Suzhen/CodeChat

请写出一个2-2-1全连接无偏置的神经网络代码,用来解决异或问题,使得训练结果远大于0时输出1,远小于1时输出0

  1. User

    请写出一个2-2-1全连接无偏置的神经网络代码,用来解决异或问题,使得训练结果远大于0时输出1,远小于1时输出0

  2. Assistant

    我们可以使用Python中的NumPy库来实现一个全连接无偏置的神经网络。这里我们使用一个2-2-1结构来解决异或问题。以下是代码实现:

    import numpy as np
    
    # 激活函数和对应的梯度
    def sigmoid(x):
        return 1 / (1 + np.exp(-x))
    
    def sigmoid_gradient(x):
        return x * (1 - x)
    
    # 训练数据
    X = np.array([[0, 0],
                  [0, 1],
                  [1, 0],
                  [1, 1]])
    
    # 标签数据
    y = np.array([[0],
                  [1],
                  [1],
                  [0]])
    
    # 随机初始化权重
    np.random.seed(1)
    weights_0_1 = np.random.random((2, 2)) # 从输入层到隐藏层的权重
    weights_1_2 = np.random.random((2, 1)) # 从隐藏层到输出层的权重
    
    # 学习率
    alpha = 0.1
    
    # 训练次数
    epochs = 10000
    
    for epoch in range(epochs):
        # 正向传播
        layer_0 = X
        layer_1 = sigmoid(np.dot(layer_0, weights_0_1))
        layer_2 = sigmoid(np.dot(layer_1, weights_1_2))
    
        # 计算输出层误差和梯度
        layer_2_error = y - layer_2
        layer_2_gradient = sigmoid_gradient(layer_2) * layer_2_error
    
        # 计算隐藏层误差和梯度
        layer_1_error = np.dot(layer_2_gradient, weights_1_2.T)
        layer_1_gradient = sigmoid_gradient(layer_1) * layer_1_error
    
        # 更新权重
        weights_1_2 += alpha * np.dot(layer_1.T, layer_2_gradient)
        weights_0_1 += alpha * np.dot(layer_0.T, layer_1_gradient)
    
        # 输出误差
        if (epoch % 1000) == 0:
            print("Error:", str(np.mean(np.abs(layer_2_error))))
    
    # 测试训练结果
    for x in X:
        layer_0 = x
        layer_1 = sigmoid(np.dot(layer_0, weights_0_1))
        layer_2 = sigmoid(np.dot(layer_1, weights_1_2))
        output = 1 if layer_2 > 0.5 else 0
        print("Input:", x, "Output:", output)
    

    这个代码实现了一个2-2-1全连接无偏置的神经网络,使用梯度下降训练网络。训练结果在远大于0时输出1,远小于1时输出0。

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.