Unimodality Tests for Global Optimization of Single Variable Functions Using Statistical Methods

Similar documents
Econometric Methods. Review of Estimation

Functions of Random Variables

Summary of the lecture in Biostatistics

UNIVERSITY OF OSLO DEPARTMENT OF ECONOMICS

CHAPTER VI Statistical Analysis of Experimental Data

Estimation of Stress- Strength Reliability model using finite mixture of exponential distributions

On Modified Interval Symmetric Single-Step Procedure ISS2-5D for the Simultaneous Inclusion of Polynomial Zeros

Chapter 5 Properties of a Random Sample

Chapter 8: Statistical Analysis of Simulated Data

Derivation of 3-Point Block Method Formula for Solving First Order Stiff Ordinary Differential Equations

Bayes Estimator for Exponential Distribution with Extension of Jeffery Prior Information

Lecture 7. Confidence Intervals and Hypothesis Tests in the Simple CLR Model

Chapter 13 Student Lecture Notes 13-1

STK4011 and STK9011 Autumn 2016

Special Instructions / Useful Data

Solving Constrained Flow-Shop Scheduling. Problems with Three Machines

Multivariate Transformation of Variables and Maximum Likelihood Estimation

X ε ) = 0, or equivalently, lim

LINEARLY CONSTRAINED MINIMIZATION BY USING NEWTON S METHOD

Simulation Output Analysis

LINEAR REGRESSION ANALYSIS

Maximum Likelihood Estimation

Chapter 13, Part A Analysis of Variance and Experimental Design. Introduction to Analysis of Variance. Introduction to Analysis of Variance

Chapter 14 Logistic Regression Models

UNIVERSITY OF OSLO DEPARTMENT OF ECONOMICS

Multiple Linear Regression Analysis

Non-uniform Turán-type problems

best estimate (mean) for X uncertainty or error in the measurement (systematic, random or statistical) best

On the Interval Zoro Symmetric Single Step. Procedure IZSS1-5D for the Simultaneous. Bounding of Real Polynomial Zeros

Lecture 3. Sampling, sampling distributions, and parameter estimation

12.2 Estimating Model parameters Assumptions: ox and y are related according to the simple linear regression model

4. Standard Regression Model and Spatial Dependence Tests

Comparison of Dual to Ratio-Cum-Product Estimators of Population Mean

Dr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur

Objectives of Multiple Regression

Some Statistical Inferences on the Records Weibull Distribution Using Shannon Entropy and Renyi Entropy

ECONOMETRIC THEORY. MODULE VIII Lecture - 26 Heteroskedasticity

Analysis of Lagrange Interpolation Formula

THE ROYAL STATISTICAL SOCIETY GRADUATE DIPLOMA

ESS Line Fitting

Research Article A New Iterative Method for Common Fixed Points of a Finite Family of Nonexpansive Mappings

{ }{ ( )} (, ) = ( ) ( ) ( ) Chapter 14 Exercises in Sampling Theory. Exercise 1 (Simple random sampling): Solution:

Chapter 8. Inferences about More Than Two Population Central Values

= 1. UCLA STAT 13 Introduction to Statistical Methods for the Life and Health Sciences. Parameters and Statistics. Measures of Centrality

UNIT 2 SOLUTION OF ALGEBRAIC AND TRANSCENDENTAL EQUATIONS

A Remark on the Uniform Convergence of Some Sequences of Functions

Simple Linear Regression

A New Family of Transformations for Lifetime Data

Lecture 1 Review of Fundamental Statistical Concepts

Point Estimation: definition of estimators

Confidence Intervals for Double Exponential Distribution: A Simulation Approach

ENGI 4421 Joint Probability Distributions Page Joint Probability Distributions [Navidi sections 2.5 and 2.6; Devore sections

STA302/1001-Fall 2008 Midterm Test October 21, 2008

The Mathematical Appendix

Lecture 8: Linear Regression

X X X E[ ] E X E X. is the ()m n where the ( i,)th. j element is the mean of the ( i,)th., then

6.867 Machine Learning

STATISTICAL PROPERTIES OF LEAST SQUARES ESTIMATORS. x, where. = y - ˆ " 1

( ) = ( ) ( ) Chapter 13 Asymptotic Theory and Stochastic Regressors. Stochastic regressors model

PTAS for Bin-Packing

Random Variables and Probability Distributions

ENGI 3423 Simple Linear Regression Page 12-01

Chapter 4 Multiple Random Variables

ρ < 1 be five real numbers. The

1 0, x? x x. 1 Root finding. 1.1 Introduction. Solve[x^2-1 0,x] {{x -1},{x 1}} Plot[x^2-1,{x,-2,2}] 3

STK3100 and STK4100 Autumn 2017

Goodness of Fit Test for The Skew-T Distribution

Complete Convergence and Some Maximal Inequalities for Weighted Sums of Random Variables

Comparison of Parameters of Lognormal Distribution Based On the Classical and Posterior Estimates

Lecture Notes Types of economic variables

Lecture 3 Probability review (cont d)

BERNSTEIN COLLOCATION METHOD FOR SOLVING NONLINEAR DIFFERENTIAL EQUATIONS. Aysegul Akyuz Dascioglu and Nese Isler

Bayes Interval Estimation for binomial proportion and difference of two binomial proportions with Simulation Study

The number of observed cases The number of parameters. ith case of the dichotomous dependent variable. the ith case of the jth parameter

GENERALIZED METHOD OF MOMENTS CHARACTERISTICS AND ITS APPLICATION ON PANELDATA

Multiple Choice Test. Chapter Adequacy of Models for Regression

Module 7. Lecture 7: Statistical parameter estimation

f f... f 1 n n (ii) Median : It is the value of the middle-most observation(s).

STK3100 and STK4100 Autumn 2018

Ordinary Least Squares Regression. Simple Regression. Algebra and Assumptions.

THE ROYAL STATISTICAL SOCIETY 2009 EXAMINATIONS SOLUTIONS GRADUATE DIPLOMA MODULAR FORMAT MODULE 2 STATISTICAL INFERENCE

MEASURES OF DISPERSION

Discrete Mathematics and Probability Theory Fall 2016 Seshia and Walrand DIS 10b

Chapter 4 (Part 1): Non-Parametric Classification (Sections ) Pattern Classification 4.3) Announcements

Analysis of Variance with Weibull Data

Confidence Interval Estimations of the Parameter for One Parameter Exponential Distribution

Homework 1: Solutions Sid Banerjee Problem 1: (Practice with Asymptotic Notation) ORIE 4520: Stochastics at Scale Fall 2015

2.28 The Wall Street Journal is probably referring to the average number of cubes used per glass measured for some population that they have chosen.

Feature Selection: Part 2. 1 Greedy Algorithms (continued from the last lecture)

Midterm Exam 1, section 2 (Solution) Thursday, February hour, 15 minutes

COMPROMISE HYPERSPHERE FOR STOCHASTIC DOMINANCE MODEL

Parameter, Statistic and Random Samples

Lecture Note to Rice Chapter 8

Chapter Business Statistics: A First Course Fifth Edition. Learning Objectives. Correlation vs. Regression. In this chapter, you learn:

Estimation of the Loss and Risk Functions of Parameter of Maxwell Distribution

Mu Sequences/Series Solutions National Convention 2014

THE ROYAL STATISTICAL SOCIETY 2010 EXAMINATIONS SOLUTIONS GRADUATE DIPLOMA MODULE 2 STATISTICAL INFERENCE

Introduction to local (nonparametric) density estimation. methods

Fourth Order Four-Stage Diagonally Implicit Runge-Kutta Method for Linear Ordinary Differential Equations ABSTRACT INTRODUCTION

CHAPTER 3 POSTERIOR DISTRIBUTIONS

Transcription:

Malaysa Umodalty Joural Tests of Mathematcal for Global Optmzato Sceces (): of 05 Sgle - 5 Varable (007) Fuctos Usg Statstcal Methods Umodalty Tests for Global Optmzato of Sgle Varable Fuctos Usg Statstcal Methods Ismal b Mohd, Sska Cadra Ngsh & Yosza Dasrl 3 Departmet of Mathematcs, Faculty of Scece ad Techology, Uverst Malaysa Tereggau (UMT), Megabag Telpot, 030 Kuala Tereggau, Malaysa E-mal: smalmd@kustem.edu.my, sskagsh@yahoo.com, 3 yosza@kustem.edu.my ABSTRACT Ths paper dscusses the radomess ad ormalty tests to the set of pots whch are obtaed by splttg a terval uder cosderato to several subtervals for verfyg that the optmzato problem costtutes a Weer process. The, by usg the optmzato probablstc algorthm, a best subterval ca be chose for umodalty test ad wll be cotued wth a smple local search method, startg at ay pot the such chose best terval. Furthermore, umercal results are gve to llustrate the tests. Keywords: Global optmzato, probablty, radomess test, ormalty test, ad umodalty test. INTRODUCTION The task of global optmzato of the optmzato problem z = f ( x) (.) subect to a x b (.) where f : R R, s to fd a soluto ts feasble soluto set for whch the obectve fucto acheved ts smallest value, the global mmum. Therefore, the global optmzato ams determg ot ust a local mmum but the smallest local mmum wth respect to ts feasble soluto set whch s defed by { } F = x a x b, x R. (.3) I (Ismal, 005) ad (Sska, 006) we have show how to use the radomess ad ormalty tests to aalyze the data obtaed by splttg the terval [ ab, ] X Rto several subtervals = [ x, x ]( = 0,, ) wth x0 = a ad x = b for verfyg + that the optmzato problem costtutes a Weer process. I ths paper, we wll descrbe how to use the umodalty test to the chose subterval whch s obtaed by usg the optmzato probablstc algorthm (Ismal, 005), (Sska, 006) before the (smple) local search procedure s employed. Malaysa Joural of Mathematcal Sceces 05

Ismal b Mohd, Sska Cadra Ngsh & Yosza Dasrl. STOCHASTIC PROCESS We shall cosder the Weer process (Archett, 978) ad Gve f = f x = µ 0 0 ( σ ) f x f y N 0, x y for x, y X. f = f x =,,, the dstrbuto of f ( x ), codtoed by (,,,, ) z = x f x f = 0, wth f x for x X [ x x] (.) s ormal wth some expected value E f ( x) z = µ x ad varace ( var f x z ) = σ ( x) gve by ad respectvely. x+ x x x 0,, ( 0,, ) f + f+ f = µ x = µ x = x+ x x+ x f ( x) = [ x, x],ad x, = x x x+ x ( 0,, ( 0,, ) ) σ f = µ x = σ ( x) = x+ x σ ( x x) ( = [ x, x],ad x, = ) Theorem. (Archett, 980) If f ( x), x [ a, b], s a Weer process such that f ( a) = f a, ad by cosderg the dstrbuto the f ( z) = P{m f () t z f ( b) = h}, a t b ( ) z m fa, f f ( z) = ( fa z)( fb z) exp < m, z fa f σ ( b a) ( ( b) ) b. 06 Malaysa Joural of Mathematcal Sceces

Umodalty Tests for Global Optmzato of Sgle Varable Fuctos Usg Statstcal Methods 3. STATISTICAL TESTS Gve f : S R R ad the pots x ( =,, ) splt the search terval X S equal parts. Let = ( = ) (3.) z f f,, The goodess of ft of the Weer process for f ca be checked by testg that whether z ( =,, ) s a radom sample draw from a ormal dstrbuto where the detals are as follows. Radomess Test The radomess of z ( =,, ) ca be checked by the smple correlato coefcet where R = z = SS z zk (3.) z = = + z = z, S = ( z z), K =, ad S = If z ( =,, ) are depedet o, the the radom varable T = R ( p) ( ) R (3.3) follows a t-dstrbuto wth ν = p degrees of freedom. The tabulated t-dstrbuto ca be used to costruct the statstcal tests for dfferet sgfcace levels α ad sample szes. The observatos z ( =,, ) are depedet f T < T αν,. Normalty Test I order to exame the ormalty of z ( =,, ), we frstly compute: { max F0 z() ; z; Sz,, F0( z )} () ; z; Sz σ = (3.4) where z() ( =,, ) are values z (,, ) F = rearraged creasg order, ad 0 x; x; s s the stadard ormal dstrbuto fucto gve by ( x x) / s u / F0 ( x; x; s) = e du, π (3.5) Malaysa Joural of Mathematcal Sceces 07

Ismal b Mohd, Sska Cadra Ngsh & Yosza Dasrl ad secodly, we compute D = max =,, ( σ ) (3.6) The varables are draw from a ormal dstrbuto f D <,. D α 4. THE PARAMETERS The followg parameters are eeded our algorthm. x x µ = f ( x0), = 0, x = x0 + ( =,, ) (4.) ad σˆ, the maxmum lkelhood estmate of σ, s computed as σˆ ( f ) f = x x = (4.) 5. STOPPING CRITERION The stoppg crtero for choose the best subterval s provded by Theorem.. Let L be the sum of the tervals of local adequacy of the model, after fucto evaluatos, performed x ( = 0,..., ) wth x x + ( = 0,..., ). At the ext step, f(x) s assumed umodal ay of the tervals of L ad a sample path of the Weer process for x X L. Defe the followg probablty We have P m, {. [, ] } = P f x f x x ε f f + x+ (5.) ( f ε f+ )( f ε f) exp, P = ( + ) σ x x 0, ([ x x+ ] L) ([ x x+ ] L) (5.) where f = m { f, = 0,, } the probablty ad ε s the prefxed accuracy level. Fally, we compute { PF = P m f x > f ε z} = ( P). (5.3) x X = 0 08 Malaysa Joural of Mathematcal Sceces

Umodalty Tests for Global Optmzato of Sgle Varable Fuctos Usg Statstcal Methods The algorthm wll termate whe PF > PT where PT s a gve probablty level, ad as example PT = 0.99. If PF PT, the chage the value of. 6. THE CONTROL OF PARAMETER The { γ } sequece of parameters for cotrollg the covergecy of the algorthm, ca be determed by Theorem. gve by the followg probablty. { Pγ = P m f x > f }. γ z (6.) x X If The parameter γ s kept costat f Pγ P T where P T s a prefxed probablty level. Pγ > P, the compute T γ + = / γ. (6.) A subterval 7. THE CHOSEN SUBINTERVAL p s chose such that the probablty P m f x < f { γ z = max P m f x < f } γ z x p x = max P m f x < f γ f, f { } { } + x ( p ) ( f γ f )( f γ f ) + = max exp σ ( x+ x) = sze (7.) where f = m ( f0, f,, f ), ad γ s some postve value. Now, dvde sze( X ) by sze( p ) to obta subterval = x, x+ ( = 0,, m) where x x0 m =. sze( p ) Suppose that we have (7.) f = f x = m f, f,, f. (7.3) t t 0 m Therefore, the (best) chose subterval s a terval whch cotas x t ad ca be expressed by [ ] I = x, x. t t+ (7.4) Malaysa Joural of Mathematcal Sceces 09

Ismal b Mohd, Sska Cadra Ngsh & Yosza Dasrl 8. UNIMODALITY TEST Defto 8. If x ( a, b) s a mmzer of f over ( ab, ), f( x) f( x ) ( x< x ), ad f ( x ) f ( x) ( x < x) for x [ a, b], the f s sad to be umodal [ ab, ]. Suppose that f umodal I = [ xt, xt+ ] where the subterval I = [ xt, xt+ ] s chose as descrbed Secto 7. Deote x ( 0) = ad x ( 0) = x + ad compute ( 0) ( 0) ( 0) ( 0) 3 = 3 ad f4 f x4 ( 0) ( 0) ( 0) ( 0) 3 4 f f x xt = for ay par x ( 0), x ( 0) x < x < x < x. Based o the values of [, ] s s+ 3 4 ( 0) f ad 3 I = x x ca be reduced accordg to the followg algorthm. Algorthm (Umodalty Test) ( 0) ( 0) ( 0) ( 0) ( 0) ( 0) ( 0) ( 0) 0 0 Data: x, x, x < x3 < x4 < x, f3 = f x, ad f4 = f ( x4 ). ( 3 ) s ( 0, 0 ) x x whch satsfy ( 0) f (Ismal, 989), the wdth of 4. case true of ( 0) ( 0) () () ( 0) ( 0) ( 0) ( 0).. f3 < f4 : x, x = x, x 4 // The mmzer, x x, x 4 ( 0) ( 0) () () ( 0) ( 0) ( 0) ( 0).. f3 > f4 : x, x = x3, x // The mmzer, x x3, x ( 0) ( 0) ( 0) ( 0) ( 0) ( 0).3. f3 = f4 : x, x = x3, x 4 // The mmzer, x x3, x 4.4. default: {}! A error. retur. However, we do ot kow exactly about the umodalty property of f the ( 0) ( 0), = s, s+ subterval x x x x eve though ths chose terval produced by our algorthm s assumed has fullflled the umodalty property. Therefore, order to obta more evdece, durg mplemetato, f for the sequeces of pots ( 0) () ( ) x, x, x, or ( 0 ) () ( x, x, x ), we ca get () ( + ) f f > or () ( + ) > f respectvely, the t ca we assumed that f s umodal that subterval where () () f = f x = 0,,, ; =,. Statstcally, ths crtera ca be acheved sce 0 0 the subterval x, x = xt, x t+ has bee tested. Ths statstcal tests have bee doe before the umodalty test. f 9. TEST EXAMPLES The statstcal ad umodalty tests have bee tested o fve fuctos as follows: 0 Malaysa Joural of Mathematcal Sceces

Umodalty Tests for Global Optmzato of Sgle Varable Fuctos Usg Statstcal Methods Example : Mmze f ( x) = s ( x) e x subect to 0 x 0 where ts graph s gve by Example : Mmze f ( x) = /cos( x) + /3s( x) subect to 0 x 5 where ts graph s gve by Example 3: Mmze ( 0.75) π f x = x + s 8π x graph s gve by subect to 0 x where ts Malaysa Joural of Mathematcal Sceces

Ismal b Mohd, Sska Cadra Ngsh & Yosza Dasrl Example 4: Mmze f ( x) = s( x) 4xcos( x) subect to 0 x 5 where ts graph s gve by Example 5: Mmze f x = xcos x x s ( x) subect to 0 x 5 where ts graph s gve by 0. NUMERICAL RESULTS The results of the statstcal ad umodalty tests for the cosdered examples are gve the Tables.,.,.,.,, 5.. I Tables.,., 3.,, ad 5., deotes the umber of samples, T ad D are computed usg equatos (3.3) ad (3.6) respectvely. The values of T0.05, T0.0, D 0.05,, ad D 0.0, are obtaed from the t-dstrbuto table ad the table of Crtcal Values for the Llefors test for Normalty (Shesk, 000). Tables., TABLE. Results of radomess ad ormalty tests for Example T T 0.05 T 0.0 D D 0.05, D 0.0, 0 0.365.306 3.355 0.85 0.58 0.94 0 0.68.0.878 0.5 0.90 0.3 30 0..048.763 0.6 0.6 0.87 40 0.94.04.7 0.0 0.40 0.63 50 0.74.0.68 0.30 0.5 0.46 00 0.4.980.68 0.5 0.089 0.03 Malaysa Joural of Mathematcal Sceces

Umodalty Tests for Global Optmzato of Sgle Varable Fuctos Usg Statstcal Methods., 3.,..., 5. cota the results of the umodalty test up to four teratos for each example. TABLE. Results of umodalty test for Example Iterato 0 3 () () () () (3) (3) 4.378-0.9396 4.4977-0.988 4.677 -.0054 4.677 -.0054 4.8576-0.9973 4.8576-0.9973 4.8576-0.9973 4.7776 -.0063 3 4.4977-0.988 4.677 -.0054 4.6976 -.0090 4.670 -.0085 4 4.6777 -.0087 4.7376 -.0084 4.7776 -.0063 4.743 -.0088 TABLE. Results of radomess ad ormalty tests for Example T T 0.05 T 0.0 D D 0.05, D 0.0, 0 0.38.306 3.355 0.39 0.58 0.94 0 0.8.0.878 0.8 0.90 0.3 30 0.88.048.763 0.0 0.6 0.87 40 0.64.04.7 0.03 0.40 0.63 50 0.47.0.68 0.04 0.5 0.46 00 0.05.980.68 0.0 0.089 0.03 TABLE. Results of umodalty test for Example Iterato 0 3 () () () () (3) (3).3970-0.6999.3970-0.6999.3970-0.6999.4985-0.700 3.089-0.5388.8536-0.660.704-0.7093.704-0.7093 3.653-0.70.549-0.736.4985-0.700.566-0.739 4.8536-0.660.704-0.7093.5999-0.79.6338-0.70 TABLE 3. Results of radomess ad ormalty tests for Example 3 T T 0.05 T 0.0 D D 0.05, D 0.0, 0 0.74.306 3.355 0.88 0.58 0.94 0 0.04.0.878 0.67 0.90 0.3 30 0.083.048.763 0.4 0.6 0.87 40 0.74.04.7 0.53 0.40 0.63 50 0.064.0.68 0.3 0.5 0.46 00 0.045.980.68 0.03 0.089 0.03 Malaysa Joural of Mathematcal Sceces 3

Ismal b Mohd, Sska Cadra Ngsh & Yosza Dasrl TABLE 3. Results of umodalty test for Example 3 Iterato 0 3 () () () () (3) (3) 0.7390-0.969 0.7390-0.969 0.7455-0.9934 0.7455-0.9934 0.7680-0.8988 0.7583-0.9780 0.7583-0.9780 0.7540-0.9948 3 0.7487-0.9994 0.7455-0.9934 0.7497 -.000 0.7483-0.999 4 0.7583-0.9780 0.759-0.9989 0.7540-0.9948 0.75-0.9996 TABLE 4. Results of radomess ad ormalty tests for Example 4 T T 0.05 T 0.0 D D 0.05, D 0.0, 0 0.354.306 3.355 0.53 0.58 0.94 0 0.54.0.878 0.35 0.90 0.3 30 0.09.048.763 0.9 0.6 0.87 40 0.8.04.7 0.0 0.40 0.63 50 0.64.0.68 0. 0.5 0.46 00 0.7.980.68 0.0 0.089 0.03 TABLE 4. Results of umodalty test for Example 4 Iterato 0 3 () () () () (3) (3) 3.49-7.877 3.4734-8.3364 3.4743-8.3364 3.5007-8.4403 3.5963-8.305 3.5963-8.305 3.5553-8.4588 3.5553-8.4588 3 3.4743-8.3364 3.543-8.4687 3.5007-8.4403 3.589-8.4747 4 3.5348-8.486 3.5553-8.4588 3.580-8.483 3.537-8.4809 TABLE 5. Results of radomess ad ormalty tests for Example 5 T T 0.05 T 0.0 D D 0.05, D 0.0, 0 0.99.306 3.355 0.98 0.58 0.94 0 0.0.0.878 0.95 0.90 0.3 30 0.83.048.763 0.08 0.6 0.87 40 0.59.04.7 0.9 0.40 0.63 50 0.43.0.68 0.8 0.5 0.46 00 0.0.980.68 0.8 0.089 0.03 4 Malaysa Joural of Mathematcal Sceces

Umodalty Tests for Global Optmzato of Sgle Varable Fuctos Usg Statstcal Methods TABLE 5. Results of umodalty test for Example 5 Iterato 0 3 () () () () (3) (3) 8.056-65.98 8.056-65.98 8.480-65.904 8.480-65.904 8.4694-63.4606 8.336-65.4767 8.336-65.4767 8.704-65.8994 3 8.939-66.0303 8.480-65.904 8.09-66.0389 8.888-66.037 4 8.336-65.4767 8.398-66.004 8.704-65.8994 8.96-66.035. CONCLUSION From the Tables.,., 3.,..., 5., for the gve fuctos, we ca say that varables are radom varables sce T < T αν,, ad the values of D < D α, suggests that the varables are draw from a ormal dstrbuto. The Tables.,., 3.,, 5. ad the gve fuctos cofrm that the best subterval determed by the suggested probablstc algorthm, s umodal.. ACKNOWLEDGEMENT The authors wsh to express ther apprecato to Uverst Malaysa Tereggau ad the Govermet of Malaysa for ther facal support ths proect. REFERENCES ARCHETTI, F. ad B. BETRO. 978. A Probablstc Algorthm For Global Optmzato. ARCHETTI, F. 980. Aalyss of Stochastc Strateges For The Numercal Soluto of The Global Optmzato Problem. Workg paper Numercal Techques For Stochastc System North- Hollad. ISMAIL, B.M., D. YOSZA, ad C.N. SISKA. 005. Probablstc Algorthm for Optmzato Usg Newto s Method, Refereed Coferece Proceedgs of Australa Operatos Research Socety ad th. Australa Optmzato Day, 9-34. ISMAIL, B.M., ad H.H. MALIK. 989. Itroducto to Optmzato Methods, Uverst Pertaa Malaysa, Serdag. SISKA, C.N., D. YOSZA, ad B.M. ISMAIL. 006. Probablstc Algorthm Optmzato, MATEMATIKA, (). SHESKIN, J. D. 000. Hadbook of Parametrc ad Noparametrc Statstcal Procedures. Lodo. Malaysa Joural of Mathematcal Sceces 5