反復回数を0にするのではなくAPIにしたほうが良い。

A new perspective for optimization is proposed based on the mathematics of heat transfer

Refference

1. Vivek K. Patel, Vimal J. Savsani, "Heat transfer search (HTS): a novel optimization algorithm"

http://www.sciencedirect.com/science/article/pii/S0020025515004764

Namespace Optimization
    ...
    Public MustInherit Class absOptimization
        ...
        Protected m_rand As System.Random = New clsRandomXorshift(CUInt(Environment.TickCount))

Environment.TickCount can actually return negative number (it does at my machine).

There should be
Protected m_rand As System.Random = New clsRandomXorshift(BitConverter.ToUInt32(BitConverter.GetBytes(Environment.TickCount), 0))

VCS simulates diffusion and infection strategies for the host cells adopted by virus to survive and propagate in the cell environment.

Refference

1. Mu Dong Li, , Hui Zhao , Xing Wei Weng , Tong Han, "A novel nature-inspired algorithm for optimization: Virus colony search"

http://www.sciencedirect.com/science/article/pii/S0965997815001702

Today, when initializing DE with upper and lower bounds, points are evaluated twice. This is inefficient and noticeable when evaluation is expensive.

Today's process:

1. clsOptDE.Init(): generate a random point using initial value ranges
2. Create and evaluate the clsPoint
3. Limit the solution space to upper and lower bounds
4. Reevaluate the point

You can see the point is evaluated twice, in steps #2 and #4

Ideal process:

1. Generate a random point using the upper and lower bounds if they exist (don't use initial value ranges)
2. Create the point

In this flow the point is only evaluated once.

There is a secondary bug with the limit function where the random point is not uniformly sampled from the upper and lower bounds. Instead it is uniformly sampled from the initial value ranges, so if the upper and lower bounds are wider, you miss parts of those bounds, and if the bounds are narrower, you are more likely to get "random" points from the edge of the bounds.

Roach Infestation Optimization

ワモンゴキブリなどの一般的なゴキブリの行動に大まかに基づく数値最適化アルゴリズム

• Refference

https://msdn.microsoft.com/ja-jp/magazine/mt632275.aspx

対角成分だけになっているのを修正

CMA-ES
共分散行列適応進化戦略、Covariance Matrix Adaptation Evolution Strategy

• Refference

https://ja.wikipedia.org/wiki/CMA-ES

git-flow
https://nvie.com/posts/a-successful-git-branching-model/

デフォルトブランチをdevelopへ

TODO:NSGA
NSGA for Multi objective optimization.
*NSGA(Non-dominated Sorting Genetic Algorithm)

assembly.infoが更新されていないためバージョンが古く見える。

数値微分
https://jp.mathworks.com/matlabcentral/fileexchange/13490-adaptive-robust-numerical-differentiation

Hello,

I would like to use Liboptimization for a 2 variables function in my vb.net code.
I've coded my function f(x,y), but then, the syntax
Dim optimization As New clsOptDEJADE(New f( ))
is not correct, as the two parameters x and y are needed, and have to be declared. Could you give an example with a customize function please?

Kazu

初期値のプロパティがあった場合、値を足していた。

If MyBase.InitialPosition IsNot Nothing AndAlso MyBase.InitialPosition.Length = Me.m_func.NumberOfVariable Then
    value += Me.InitialPosition(j)
End If

集団の中で1つだけ設定しても効果が薄いと思うので、集団の1/10を初期値に書き換えるようにする。

値を出力するようにする

多次元配列を入れて初期化できるようにする。

var mat = new DenseMatrix( array[3,3] );

TODO:NSGAII
NSGAII for Multi objective optimization. NSGAII that improved NSGA.

• Refference
  Deb. K ,Amrit Pratap, Sameer Agarwal and T. Meyarivan (2000). A Fast and Elitist Multi-Objective Genetic Algorithm-NSGA-II. KanGAL Report Number 2000001.
  http://www.iitk.ac.in/kangal/reports.shtml

MS-PLからMITへ変更する。
ライセンス変更に伴うデメリットは特にない。

制約条件の追加。
強い要望がある。

.NET Core対応に向けて全てC#化

シリアライズして途中結果を保存できるようにする。

• overview
  Bayesian Adaptive Direct Search

• Refference

https://arxiv.org/abs/1705.04405

https://github.com/lacerbi/bads

Refference:
https://journal.r-project.org/archive/2013/RJ-2013-002/RJ-2013-002.pdf
https://scipy.github.io/devdocs/generated/scipy.optimize.dual_annealing.html

hi, tomi, your project is really awesome

i'm trying to simulate this R code in vb.net

dp <- garch_test$X__1
llh <- function(par) { mu <- par[1]; omega <- par[2]; alpha <- par[3]; beta <- par[4]; e2 <- (dp-mu)^2; e2t <- omega + alphac(mean(e2), e2[-length(dp)]); s2 <- filter(e2t, beta, "recursive", init = mean(e2)); 0.5sum( log(2pi) + log(s2) + e2/s2) }
mu <- mean(dp); omega <- 0.1var(dp); alpha <- 0.1; beta <- 0.8
par <- c(mu, omega, alpha, beta)
small <- 1e-6
low <- c(-10abs(mu), small, small, small)
up <- c(10abs(mu), 100*abs(mu), 1-small, 1-small)
fit <- nlminb(start=par, objective=llh, lower=low, upper=up, control=list(x.tol = 1e-8,trace=0))
names(fit$par) <- c("mu", "omega", "alpha", "beta")
round(fit$par, 6)

nlminb optimization method allows initial parameter values and low/up bound of parameters
it converges really fast

any of your method allows this options?

i tested clsOptDE and gets a really accurate values, but does not converge so fast

thanks in advance

Good afternoon.
The question of how to maximize the function?
Example: z = 3 * Math.Pow((1 - x), 2) * Math.Exp(-(Math.Pow(x , 2)) - Math.Pow((y + 1) , 2)) - 10 * (x / 5 - Math.Pow(x , 3) - Math.Pow(y , 5)) * Math.Exp(-Math.Pow(x , 2) - Math.Pow(y , 2)) - 1 / 3 * Math.Exp(-Math.Pow((x + 1) , 2) - Math.Pow(y , 2));

エラー関係を見直し。
例外を投げるようにした方が良さそう。

It is the same simple as the greedy Hill-Climbing, but much more powerful.

Refference

http://www.yuribykov.com/LAHC/

In the SampleVB project there is this code for using clsOptDE, with two Console write-line statements I added:

   'LeastSquaresMethod 最小二乗法
    With Nothing
        Console.WriteLine("------------------------------ BEGIN: DATA.CSV --------------------")

        '評価関数
        Dim objectiveFunction = New clsLeastSquaresMethod()
        If objectiveFunction.Init("..\..\..\_sampledata\data.csv") = False Then
            Return
        End If

        '最小化
        Dim opt As New LibOptimization.Optimization.clsOptDE(objectiveFunction)
        opt.Init()
        LibOptimization.Util.clsUtil.DebugValue(opt)
        While (opt.DoIteration(50) = False)
            LibOptimization.Util.clsUtil.DebugValue(opt, ai_isOutValue:=False)
        End While
        LibOptimization.Util.clsUtil.DebugValue(opt)
        Console.WriteLine("------------------------------ END: DATA.CSV --------------------")
    End With

When I run the program I get:

------------------------------ BEGIN: DATA.CSV --------------------
TargetFunction:clsLeastSquaresMethod Dimension:5
OptimizeMethod:clsOptDE
Eval :59438373.1751817
IterationCount:0
Result :
-0.0158704898822757
-0.381333913044383
4.93756533831767
-4.28593948466842
3.24185321532233

Eval :1429.78513592157
Eval :0.956465983765879
Eval :0.000497619640101209
Eval :3.08647227393116E-07
Eval :1.38365563205351E-10
Eval :1.84394063162285E-13
Eval :1.06744757544432E-16
Eval :4.16369815393116E-20
Eval :2.33202429712711E-23
TargetFunction:clsLeastSquaresMethod Dimension:5
OptimizeMethod:clsOptDE
Eval :3.04300885378471E-24
IterationCount:460
Result :
0.0499999999999999
0.999999999999999
-9.99999999999997
-99.9999999999999
99.9999999999983

------------------------------ END: DATA.CSV --------------------

Can someone tell me how this relates to the data found in data.csv, the data file used with the object function?:

x,y
-20,-1900
-19.8,-1918.02392
-18.8,-1953.08032
-7.2,62.72128
3.8,-359.10232
7.2,-630.78272

I'm trying to figure out how to use clsOptDE with my app, where I need to call a particular function (the "fitness" function in GA terms) that analyzes a particular test vector against my fitness function. My fitness function uses the field values in the test vector to run a test against a data set.

ベクトルと行列のテストコードを追加したい。

.NET Core対応