C#,数值计算——Dynpro的计算方法与源程序

给定向量nstate,其整数值是每个状态中的状态数阶段(第一和最后阶段为1),并给定函数成本(j,k,i)返回在阶段i的状态j和的状态k之间移动的成本阶段i+1,此例程返回与nstate长度相同的向量包含最低成本路径的状态号。状态数从0开始,并且返回向量的第一个和最后一个分量将因此总是0。

Given the vector nstate whose integer values are the number of states in each stage(1 for the first and last stages), and given a function cost(j, k, i)that returns the cost of moving between state j of stage i and state k ofstage i+1, this routine returns a vector of the same length as nstatecontaining the state numbers of the lowest cost path. States number from 0,and the first and last components of the returned vector will thus always be 0.

using System;

namespace Legalsoft.Truffer

{

/// <summary>

/// Given the vector nstate whose integer values are the number of states in each

/// stage(1 for the first and last stages), and given a function cost(j, k, i)

/// that returns the cost of moving between state j of stage i and state k of

/// stage i+1, this routine returns a vector of the same length as nstate

/// containing the state numbers of the lowest cost path. States number from 0,

/// and the first and last components of the returned vector will thus always be 0.

/// </summary>

public abstract class Dynpro

{

public Dynpro()

{

}

public abstract double cost(int jj, int kk, int ii);

public int[] dynpro(int[] nstate)

{

const double BIG = 1.0e99;

const double EPS = float.Epsilon; //numeric_limits<double>.epsilon();

int nstage = nstate.Length - 1;

int[] answer = new int[nstage + 1];

if (nstate[0] != 1 || nstate[nstage] != 1)

{

throw new Exception("One state allowed in first and last stages.");

}

double[,] best = new double[nstage + 1, nstate[0]];

best[0, 0] = 0.0;

for (int i = 1; i <= nstage; i++)

{

for (int k = 0; k < nstate[i]; k++)

{

double b = BIG;

for (int j = 0; j < nstate[i - 1]; j++)

{

double a = best[i - 1, j] + cost(j, k, i - 1);

if ((a) < b)

{

b = a;

}

}

best[i, k] = b;

}

}

answer[nstage] = answer[0] = 0;

for (int i = nstage - 1; i > 0; i--)

{

int k = answer[i + 1];

double b = best[i + 1, k];

int j = 0;

for (; j < nstate[i]; j++)

{

double temp = best[i, j] + cost(j, k, i);

if (Math.Abs(b - temp) <= EPS * Math.Abs(temp))

{

break;

}

}

answer[i] = j;

}

return answer;

}

}

}

相关推荐
颜酱3 小时前
图的数据结构:从「多叉树」到存储与遍历
javascript·后端·算法
zone77398 小时前
006:RAG 入门-面试官问你,RAG 为什么要切块?
后端·算法·面试
CoovallyAIHub11 小时前
OpenClaw 近 2000 个 Skills,为什么没有一个好用的视觉检测工具?
深度学习·算法·计算机视觉
CoovallyAIHub11 小时前
CVPR 2026 | 用一句话告诉 AI 分割什么——MedCLIPSeg 让医学图像分割不再需要海量标注
深度学习·算法·计算机视觉
CoovallyAIHub11 小时前
Claude Code 突然变成了 66 个专家?这个 5.8k Star 的开源项目,让我重新理解了什么叫"会用 AI"
深度学习·算法·计算机视觉
兆子龙11 小时前
前端哨兵模式(Sentinel Pattern):优雅实现无限滚动加载
前端·javascript·算法
CoovallyAIHub15 小时前
9个视觉语言模型工厂实测:Qwen 87.9%碾压全场,你的显卡能跑哪个?
算法
程序设计实验室15 小时前
C# 扩展方法只会写 this 吗?C# 14 新语法直接把扩展方法玩出了花
c#
SparkX开源AI知识库15 小时前
手摸手带你安装OpenClaw并对接飞书
算法·架构
一语071615 小时前
3分钟搞懂深度学习AI:实操篇:卷积层
人工智能·算法