基于Python实现BP神经网络四分类混淆矩阵代码

以下是一个基于Python实现BP神经网络四分类混淆矩阵的示例代码：

import numpy as np

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

# 定义BP神经网络类
class BPNeuralNetwork:
    def __init__(self, input_num, hidden_num, output_num):
        # 初始化权重
        self.weights_ih = np.random.rand(hidden_num, input_num)
        self.weights_ho = np.random.rand(output_num, hidden_num)
        # 初始化偏置
        self.bias_h = np.random.rand(hidden_num, 1)
        self.bias_o = np.random.rand(output_num, 1)

    def forward(self, inputs):
        # 计算隐藏层输出
        hidden_inputs = np.dot(self.weights_ih, inputs) + self.bias_h
        hidden_outputs = sigmoid(hidden_inputs)
        # 计算输出层输出
        output_inputs = np.dot(self.weights_ho, hidden_outputs) + self.bias_o
        output_outputs = sigmoid(output_inputs)
        return output_outputs

    def train(self, inputs, targets, learning_rate):
        # 前向传播
        hidden_inputs = np.dot(self.weights_ih, inputs) + self.bias_h
        hidden_outputs = sigmoid(hidden_inputs)
        output_inputs = np.dot(self.weights_ho, hidden_outputs) + self.bias_o
        output_outputs = sigmoid(output_inputs)

        # 计算输出层误差
        output_errors = targets - output_outputs
        # 计算隐藏层误差
        hidden_errors = np.dot(self.weights_ho.T, output_errors) * hidden_outputs * (1 - hidden_outputs)

        # 更新权重和偏置
        self.weights_ho += learning_rate * np.dot(output_errors * output_outputs * (1 - output_outputs), hidden_outputs.T)
        self.weights_ih += learning_rate * np.dot(hidden_errors * hidden_outputs * (1 - hidden_outputs), inputs.T)
        self.bias_o += learning_rate * output_errors * output_outputs * (1 - output_outputs)
        self.bias_h += learning_rate * hidden_errors * hidden_outputs * (1 - hidden_outputs)

# 加载数据集
data = np.genfromtxt('data.csv', delimiter=',')
inputs = data[:, :4]
targets = data[:, 4:]

# 数据归一化
inputs = inputs / np.max(inputs, axis=0)
targets = targets / np.max(targets, axis=0)

# 划分训练集和测试集
train_inputs = inputs[:120, :].T
train_targets = targets[:120, :].T
test_inputs = inputs[120:, :].T
test_targets = targets[120:, :].T

# 创建BP神经网络
network = BPNeuralNetwork(4, 8, 4)

# 训练BP神经网络
for i in range(1000):
    network.train(train_inputs, train_targets, 0.1)

# 在测试集上进行预测
test_outputs = network.forward(test_inputs)

# 计算混淆矩阵
conf_matrix = np.zeros((4, 4))
for i in range(test_outputs.shape[1]):
    predict_label = np.argmax(test_outputs[:, i])
    true_label = np.argmax(test_targets[:, i])
    conf_matrix[true_label, predict_label] += 1

# 输出混淆矩阵
print('Confusion matrix:')
print(conf_matrix)

其中，`sigmoid`函数用于计算神经网络的输出，`BPNeuralNetwork`类定义了BP神经网络的结构和训练方法，`train`方法用于训练神经网络，`forward`方法用于进行前向传播，`data`变量存储了数据集，`inputs`和`targets`分别存储了输入和输出数据，`train_inputs`、`train_targets`、`test_inputs`和`test_targets`分别存储了训练集和测试集的输入和输出数据，`network`变量创建了一个BP神经网络，`test_outputs`用于存储在测试集上的预测结果，`conf_matrix`变量用于存储混淆矩阵。