bp算法 python 两个隐藏层

下是一个使用Python实现BP算法的示例，包含两个隐藏层```python
import numpy as np

# 定义sigmoid函数
def sigmoid(x):
return 1 / (1 + np.exp(-x))

# 定义sigmoid函数的导数
def sigmoid_derivative(x):
return x * (1 - x)

# 定义BP算法
def backpropagation(X, y, hidden_neurons_1, hidden_neurons_2, epochs, learning_rate):
input_neurons = X.shape[1]
output_neurons = y.shape[1]

# 初始化权值和偏置
weight_hidden_1 = np.random.uniform(size=(input_neurons, hidden_neurons_1))
bias_hidden_1 = np.random.uniform(size=(1, hidden_neurons_1))
weight_hidden_2 = np.random.uniform(size=(hidden_neurons_1, hidden_neurons_2))
bias_hidden_2 = np.random.uniform(size=(1, hidden_neurons_2))
weight_output = np.random.uniform(size=(hidden_neurons_2, output_neurons))
bias_output = np.random.uniform(size=(1, output_neurons))

# 训练模型
for i in range(epochs):
# 前向传播
hidden_layer_1_input = np.dot(X, weight_hidden_1) + bias_hidden_1
hidden_layer_1_output = sigmoid(hidden_layer_1_input)
hidden_layer_2_input = np.dot(hidden_layer_1_output, weight_hidden_2) + bias_hidden_2
hidden_layer_2_output = sigmoid(hidden_layer_2_input)
output_layer_input = np.dot(hidden_layer_2_output, weight_output) + bias_output
output_layer_output = sigmoid(output_layer_input)

# 反向传播
error = y - output_layer_output
d_output = error * sigmoid_derivative(output_layer_output)
error_hidden_2 = d_output.dot(weight_output.T)
d_hidden_2 = error_hidden_2 * sigmoid_derivative(hidden_layer_2_output)
error_hidden_1 = d_hidden_2.dot(weight_hidden_2.T)
d_hidden_1 = error_hidden_1 * sigmoid_derivative(hidden_layer_1_output)

# 更新权值和偏置
weight_output += hidden_layer_2_output.T.dot(d_output) * learning_rate
bias_output += np.sum(d_output, axis=0, keepdims=True) * learning_rate
weight_hidden_2 += hidden_layer_1_output.T.dot(d_hidden_2) * learning_rate
bias_hidden_2 += np.sum(d_hidden_2, axis=0, keepdims=True) * learning_rate
weight_hidden_1 += X.T.dot(d_hidden_1) * learning_rate
bias_hidden_1 += np.sum(d_hidden_1, axis=0, keepdims=True) * learning_rate

return weight_hidden_1, bias_hidden_1, weight_hidden_2, bias_hidden_2, weight_output, bias_output

其中，`X`是输入数据，`y`是输出数据，`hidden_neurons_1`和`hidden_neurons_2`是两个隐藏层中的神经元数量，`epochs`是迭代次数，`learning_rate`是学习率。函数返回训练好的模型的权值和偏置。