使用lstm, 基于我写的cpp代码, 生成新的cpp代码.zip

# lstm生成cpp代码项目 ## 第一版想法 ## 步骤 - [x] 获取并整合我自己写的cpp代码(考虑将每个cpp文件独立出来) - [x] 分隔所有的原子单位, 并转化为以0开始的整数 - [x] 生成lstm输入的序列(因为每个文件独立, 所以对每个文件获取若干个序列) - [x] 定义网络, 使用pytorch的embedding和lstm模型 - [x] 定义超参数 - [x] 训练模型, 调参, 保存模型 - [x] 运行模型, 得出生成的代码, 自动格式化(如果做得到) - [x] 预测时多输入序列 ## 问题点 - [x] 关键字较少, 模式非常类似. ## 第一版结果 训练的loss值. 其实整个训练的中后期一直是这个值左右.  ### include开头 ```cpp #include <cmath> #include <vector> usingname space std; intmain() { long long a[4] = { {-1, 1, 3, 3, 3, 5, 3, 4, 5, 5, 6, 7, 9, 3, 3}, {1, -2, --3, -1, 1, 3, 2, 3, 3, -3, 3}; intn, m, k, k = 0, j = 0; while (n[i] != 0) { //cin.get(); for (intj = 0; j < n; ++j) if (a[i] != '1')&&str[i]<='z'))cout<<"\n当前链表为:"; list_1.print(); list_1.push_back(3); list_1.push_back(i); for (inti = 0; i < 3; i++) { cout << a[i]; } /*251231423535154233635345646565*/ /*4556344432455445453654636364565454454345668999678867791097842124264643543577899998848784214433665689645563422545564224643334435244423685445456456799888888888888888888888888888888822888888888888888888888888*//*412433423445436568986 ``` ### main开头 ```cpp main() { int n, m; cin >> n; for (int i = 2; i < = n; + +i) { if (str[i] = = ' 1 ') { cout < < " \ n当前链表为 : \ n "; for (int j = 1; j < = n; + +j) { for (int j = 0, j = i; j < = 10; + +j) { cin >> a[i]; } } cout < < ans; } ``` ### #include<iostream>开头 ```cpp #include < iostream > #include < vector > using namespace std; int main() { int a[1001], n, i = 0; for (register int i = 0; i < n; + +i) { cout < < n[i]; cout < < endl; } int n, m, m; cin >> n; for (int i = 1; i < = n; + +i) { cin >> a >> b; for (int i = 1; i < = 4; i + +) { if (!vis[v][i] & &s[j + +] = = 0) return false; } if ((i + 1 ) % 2 = 1 ; if (n < = m ) { if (i = = 0) { tmp + = intToStr(f _ m, i, x, y, y); } } } void loop(int x) { for (int i = 1; i < = 10; i + +) { for (int j = 1; j < = n; + +l) { / / } cout < < " No \ n \ n "; / / cout < < " - - - - - - - - - - - - - - - - - - - - - - " < < endl; cout < < endl; } } } ``` 其他可在generate文件夹中查看