LibOptimization is numerical optimization algorithm library for .NET Framework. / .NET用の数値計算、最適化ライブラリ
appveyor | nuget |
---|---|
LibOptimization is a numerical optimization library that simplifies optimization using C#, VisualBasic.Net and other .NET Framework languages. This library is used by people who need optimization such as science (eg physics), engineering, sound, finance, statistics, medical care, structural design etc.
LibOptimizationは制約条件の無い数値最適化を行う.NET Frameworkのライブラリです。科学 (物理学など)、エンジニアリング、音響、金融、統計、医療、構造設計などの最適化を必要とする人に使用されているようです。
実装しているアルゴリズムは最急降下法、ニュートン法、HookeJeevesのパターンサーチ法、Nelder-Mead法(オリジナルの実装、Wikipediaの実装)、適応パラメータ Nelder-Mead法、実数値遺伝的アルゴリズム(BLX-α、UNDX、SPX(シンプレクス)、REX、世代交代はJGG、PCX(世代交代はG3))、進化戦略(Evolution Strategy、1+1 ES)、粒子群最適化(Basic PSO, LDIW-PSO, CDIW-PSO, CRIW-PSO, AIW-PSO)、Differential Evolution(差分進化? DE/rand/1/bin, DE/rand/2/bin, DE/best/1/bin, DE/best/2/bin)、JADE(自己適応型DE)ホタルアルゴリズム、Cuckoo Search(Matlabコードの移植版)、焼きなまし法、山登り法です。
I may miss your Issues. When a reply is slow, please give me e-mail.
tomi.nori+github atmark gmail.com
If you use LibOptimization in your publication, please cite the following
@misc{LibOptimization, author = "N.Tomita", title = "LibOptimization", howpublished = "\url{https://github.com/tomitomi3/LibOptimization}", }
Changed absOptimiazation.NumberOfVariable from propety to function in ver1.9.0. Refactoring LibOptimization code with development branch. In the future, I will add new function to the new branch.
LibOptimization has several optimization algorithms implemented. You design the objective function, you can use all the optimized algorithms implemented.
URL:https://www.nuget.org/packages/LibOptimization/
PM> Install-Package LibOptimization
Typical Use
See this link for details
for VB.NET
'Instantiation optimization class and set objective function.
Dim optimization As New clsOptSteepestDescent(New clsBenchSphere(1))
'Initialize starting value
optimization.Init()
'Do calc
optimization.DoIteration()
'Get result. Check recent error.
If optimization.IsRecentError() = True Then
Return
Else
clsUtil.DebugValue(optimization)
End If
for C#
//Instantiation objective Function
var func = new RosenBlock();
//Instantiation optimization class and set objective function.
var opt = new clsOptNelderMead(func);
opt.Init();
//Do calc
opt.DoIteration();
//Check Error
if (opt.IsRecentError() == true)
{
return;
}
else
{
//Get Result
clsUtil.DebugValue(opt);
}
objective function : clsBenchTest2(x1,x2) = x1^4 - 20x1^2 + 20x1 + x2^4 - 20x2^2 + 20x2
x1 -> 0.0 to 5.0
x2 -> 1.0 to 4.0
for VB.NET
'Set boundary variable
opt.LowerBounds = New Double() {0, 1.0}
opt.UpperBounds = New Double() {5, 4.0}
'Init
opt.Init()
for C#
//Set boundary variable
opt.LowerBounds = new double[] {0, 1.0};
opt.UpperBounds = new double[] {5, 4.0};
//Init
opt.Init();
When using a typical code, internal criteria are enabled. For details, see EPS property, clsUtil.IsCriterion.
for C#
var opt = new LibOptimization.Optimization.clsOptDEJADE(new RosenBrock(10));
//Disable Internal criterion
opt.IsUseCriterion = false;
//Init
opt.Init();
clsUtil.DebugValue(opt);
//do optimization!
while (opt.DoIteration(100) == false)
{
var eval = opt.Result.Eval;
//my criterion
if (eval < 0.01)
{
break;
}
else
{
clsUtil.DebugValue(opt, ai_isOutValue: false);
}
}
clsUtil.DebugValue(opt);
Generate initial positions around x1=10 and x2=10.
for VB.NET
optimization.InitialPosition = {10, 10}
optimization.Init()
optimization.DoIteration()
clsUtil.DebugValue(optimization)
for C#
opt.InitialPosition = new double[] { 10, 10 };
optimization.Init()
optimization.DoIteration()
clsUtil.DebugValue(optimization)
Generate the initial position in the range of 7 to 10.
for VB.NET
Dim optimization As New Optimization.clsOptDE(New clsBenchSphere(2))
'set initialposition
optimization.InitialPosition = New Double() {10, 10}
'Initial value is generated in the range of -3 to 3.
optimization.InitialValueRangeLower = -3
optimization.InitialValueRangeLower = 3
'init and do optimization
optimization.Init()
optimization.DoIteration()
clsUtil.DebugValue(optimization)
for C#
var func = new RosenBrock(2);
var opt = new LibOptimization.Optimization.clsOptPSO(func);
opt.InitialPosition = new double[] { -10, -10 };
opt.InitialValueRangeLower = -3;
opt.InitialValueRangeUpper = 3;
opt.Init();
opt.DoIteration();
for VB.NET
Dim optimization As New clsOptSteepestDescent(New clsBenchSphere(2))
optimization.Init()
//per 5 iteration
While (optimization.DoIteration(5) = False)
clsUtil.DebugValue(optimization)
End While
clsUtil.DebugValue(optimization, ai_isOnlyIterationCount:=True)
End With
for C#
//per 100 iteration
while (opt.DoIteration(100)==false)
{
clsUtil.DebugValue(opt, ai_isOutValue: false);
}
clsUtil.DebugValue(opt);
for VB.NET
//fix RND for random sequence
Util.clsRandomXorshiftSingleton.GetInstance.SetDefaultSeed()
Dim optimization As New Optimization.clsOptDE(New clsBenchSphere(2))
//fix RND for generate position
optimization.Random = New Util.clsRandomXorshift()
'init
optimization.Init()
for C#
//fix RND for random sequence
LibOptimization.Util.clsRandomXorshiftSingleton.GetInstance().SetDefaultSeed();
var func = new RosenBrock(2);
var opt = new LibOptimization.Optimization.clsOptPSO(func);
//fix RND for generate position
opt.Random = new LibOptimization.Util.clsRandomXorshift();
//init
opt.Init();
for VB.NET
Dim optimization As New Optimization.clsOptRealGAREX(New clsBenchDeJongFunction3())
'1st try
optimization.Init()
While (optimization.DoIteration(100) = False)
clsUtil.DebugValue(optimization, ai_isOutValue:=False)
End While
clsUtil.DebugValue(optimization)
'2nd try reuse
optimization.InitialPosition = optimization.Result().ToArray()
optimization.Init()
While (optimization.DoIteration(100) = False)
clsUtil.DebugValue(optimization, ai_isOutValue:=False)
End While
clsUtil.DebugValue(optimization)
Dim optimization As New clsOptRealGASPX(New clsBenchRastriginFunction(20))
optimization.Init()
clsUtil.DebugValue(optimization)
While True
If optimization.DoIteration(10) = True Then
Exit While
End If
clsUtil.DebugValue(optimization, ai_isOutValue:=False)
End While
If optimization.IsRecentError() = True Then
Return
End If
clsUtil.DebugValue(optimization)
'prepare many optimization class.
Dim multipointNumber As Integer = 30
Dim listOptimization As New List(Of absOptimization)
For i As Integer = 0 To multipointNumber - 1
Dim tempOpt As New clsOptNelderMead(New clsBenchAckley(20))
tempOpt.Init()
listOptimization.Add(tempOpt)
Next
'using Parallel.ForEach
Dim lockObj As New Object()
Dim best As LibOptimization.absOptimization = Nothing
Threading.Tasks.Parallel.ForEach(listOptimization, Sub(opt As absOptimization)
opt.DoIteration()
'Swap best result
SyncLock lockObj
If best Is Nothing Then
best = opt
ElseIf best.Result.Eval > opt.Result.Eval Then
best = opt
End If
End SyncLock
End Sub)
'Check Error
If best.IsRecentError() = True Then
Return
Else
clsUtil.DebugValue(best)
End If
You design the evaluation function to minimize residual sum of squares. The following example estimate a parameter of the multinomial expression.
Public Overrides Function F(x As List(Of Double)) As Double
Dim sumDiffSquare As Double = 0
For Each temp In Me.datas
'e.g a * x^4 + b * x^3 + c * x^2 + d * x^4 + e
Dim predict = x(0) * temp(0) ^ 4 + x(1) * temp(0) ^ 3 + x(2) * temp(0) ^ 2 + x(3) * temp(0) + x(4)
Dim diffSquare = (temp(1) - predict) ^ 2
sumDiffSquare += diffSquare
Next
Return sumDiffSquare
End Function
This Library's license was MS-PL until this commit.