神经网络优化计算C语言代码例子

以下是一个简单的神经网络优化计算的 C 语言代码例子：

#include <stdio.h>
#include <stdlib.h>
#include <math.h>

// 定义神经网络结构体
typedef struct {
    int num_inputs;   // 输入层节点数
    int num_hidden;   // 隐藏层节点数
    int num_outputs;  // 输出层节点数
    double *inputs;   // 输入层节点值
    double *hidden;   // 隐藏层节点值
    double *outputs;  // 输出层节点值
    double *weights1; // 输入层到隐藏层的权重矩阵
    double *weights2; // 隐藏层到输出层的权重矩阵
} neural_network;

// 定义激活函数
double sigmoid(double x) {
    return 1.0 / (1.0 + exp(-x));
}

// 初始化神经网络
void init_network(neural_network *nn) {
    int i, j;

    // 分配内存空间
    nn->inputs = (double*)malloc(nn->num_inputs * sizeof(double));
    nn->hidden = (double*)malloc(nn->num_hidden * sizeof(double));
    nn->outputs = (double*)malloc(nn->num_outputs * sizeof(double));
    nn->weights1 = (double*)malloc(nn->num_inputs * nn->num_hidden * sizeof(double));
    nn->weights2 = (double*)malloc(nn->num_hidden * nn->num_outputs * sizeof(double));

    // 初始化权重矩阵
    for (i = 0; i < nn->num_inputs; i++) {
        for (j = 0; j < nn->num_hidden; j++) {
            nn->weights1[i * nn->num_hidden + j] = (double)rand() / RAND_MAX - 0.5;
        }
    }

    for (i = 0; i < nn->num_hidden; i++) {
        for (j = 0; j < nn->num_outputs; j++) {
            nn->weights2[i * nn->num_outputs + j] = (double)rand() / RAND_MAX - 0.5;
        }
    }
}

// 计算神经网络输出
void forward(neural_network *nn) {
    int i, j;

    // 计算隐藏层节点值
    for (i = 0; i < nn->num_hidden; i++) {
        nn->hidden[i] = 0.0;
        for (j = 0; j < nn->num_inputs; j++) {
            nn->hidden[i] += nn->inputs[j] * nn->weights1[j * nn->num_hidden + i];
        }
        nn->hidden[i] = sigmoid(nn->hidden[i]);
    }

    // 计算输出层节点值
    for (i = 0; i < nn->num_outputs; i++) {
        nn->outputs[i] = 0.0;
        for (j = 0; j < nn->num_hidden; j++) {
            nn->outputs[i] += nn->hidden[j] * nn->weights2[j * nn->num_outputs + i];
        }
        nn->outputs[i] = sigmoid(nn->outputs[i]);
    }
}

// 反向传播算法
void backpropagation(neural_network *nn, double *targets, double learning_rate) {
    int i, j;

    // 计算输出层误差
    double *output_deltas = (double*)malloc(nn->num_outputs * sizeof(double));
    for (i = 0; i < nn->num_outputs; i++) {
        double error = targets[i] - nn->outputs[i];
        output_deltas[i] = error * nn->outputs[i] * (1.0 - nn->outputs[i]);
    }

    // 计算隐藏层误差
    double *hidden_deltas = (double*)malloc(nn->num_hidden * sizeof(double));
    for (i = 0; i < nn->num_hidden; i++) {
        double error = 0.0;
        for (j = 0; j < nn->num_outputs; j++) {
            error += output_deltas[j] * nn->weights2[i * nn->num_outputs + j];
        }
        hidden_deltas[i] = error * nn->hidden[i] * (1.0 - nn->hidden[i]);
    }

    // 更新权重矩阵
    for (i = 0; i < nn->num_hidden; i++) {
        for (j = 0; j < nn->num_outputs; j++) {
            nn->weights2[i * nn->num_outputs + j] += learning_rate * output_deltas[j] * nn->hidden[i];
        }
    }

    for (i = 0; i < nn->num_inputs; i++) {
        for (j = 0; j < nn->num_hidden; j++) {
            nn->weights1[i * nn->num_hidden + j] += learning_rate * hidden_deltas[j] * nn->inputs[i];
        }
    }

    // 释放内存空间
    free(output_deltas);
    free(hidden_deltas);
}

int main() {
    // 初始化随机数种子
    srand(1);

    // 初始化神经网络
    neural_network nn;
    nn.num_inputs = 2;
    nn.num_hidden = 3;
    nn.num_outputs = 1;
    init_network(&nn);

    // 定义训练数据集
    double training_data[][3] = {{0, 0, 0}, {0, 1, 1}, {1, 0, 1}, {1, 1, 0}};

    // 开始训练神经网络
    int i, j, k;
    for (i = 0; i < 10000; i++) {
        for (j = 0; j < 4; j++) {
            nn.inputs[0] = training_data[j][0];
            nn.inputs[1] = training_data[j][1];
            forward(&nn);
            double targets[] = {training_data[j][2]};
            backpropagation(&nn, targets, 0.1);
        }
    }

    // 测试神经网络
    for (i = 0; i < 4; i++) {
        nn.inputs[0] = training_data[i][0];
        nn.inputs[1] = training_data[i][1];
        forward(&nn);
        printf("%f XOR %f = %f\n", nn.inputs[0], nn.inputs[1], nn.outputs[0]);
    }

    // 释放内存空间
    free(nn.inputs);
    free(nn.hidden);
    free(nn.outputs);
    free(nn.weights1);
    free(nn.weights2);

    return 0;
}

该代码实现了一个简单的 XOR 问题的神经网络，并使用反向传播算法进行权重矩阵的优化。具体实现细节请参考代码注释。