Analysis and Selection of Myriad Estimate Tuning Parameter For SαS Distributions Roenko A.A., Lukin V.V., Djurović I.

Slides:



Advertisements
Similar presentations
DATA & STATISTICS 101 Presented by Stu Nagourney NJDEP, OQA.
Advertisements

NORMAL OR GAUSSIAN DISTRIBUTION Chapter 5. General Normal Distribution Two parameter distribution with a pdf given by:
Estimation of Means and Proportions
Statistical Techniques I EXST7005 Start here Measures of Dispersion.
Sampling: Final and Initial Sample Size Determination
POINT ESTIMATION AND INTERVAL ESTIMATION
Statistics 1: Introduction to Probability and Statistics Section 3-3.
Normal Distribution * Numerous continuous variables have distribution closely resemble the normal distribution. * The normal distribution can be used to.
Maximum likelihood Conditional distribution and likelihood Maximum likelihood estimations Information in the data and likelihood Observed and Fisher’s.
Chapter 6 Continuous Random Variables and Probability Distributions
2.3. Measures of Dispersion (Variation):
Maximum likelihood (ML)
Continuous Probability Distribution  A continuous random variables (RV) has infinitely many possible outcomes  Probability is conveyed for a range of.
Lecture 4 Dustin Lueker.  The population distribution for a continuous variable is usually represented by a smooth curve ◦ Like a histogram that gets.
Continuous Probability Distributions
Common Probability Distributions in Finance. The Normal Distribution The normal distribution is a continuous, bell-shaped distribution that is completely.
© Copyright McGraw-Hill CHAPTER 6 The Normal Distribution.
Chapter 7 Estimation: Single Population
1 More about the Sampling Distribution of the Sample Mean and introduction to the t-distribution Presentation 3.
Short Resume of Statistical Terms Fall 2013 By Yaohang Li, Ph.D.
National Aerospace University of Ukraine ANALYSIS OF MERIDIAN ESTIMATOR PERFORMANCE FOR NON-GAUSSIAN PDF DATA SAMPLES Dmitriy Kurkin, Alexey Roenko, Vladimir.
HAWKES LEARNING SYSTEMS math courseware specialists Copyright © 2010 by Hawkes Learning Systems/Quant Systems, Inc. All rights reserved. Chapter 8 Continuous.
1 Normal Random Variables In the class of continuous random variables, we are primarily interested in NORMAL random variables. In the class of continuous.
Lecture 3 A Brief Review of Some Important Statistical Concepts.
Measures of Dispersion CUMULATIVE FREQUENCIES INTER-QUARTILE RANGE RANGE MEAN DEVIATION VARIANCE and STANDARD DEVIATION STATISTICS: DESCRIBING VARIABILITY.
CONTINUOUS RANDOM VARIABLES AND THE NORMAL DISTRIBUTION.
Slide 1 © 2002 McGraw-Hill Australia, PPTs t/a Introductory Mathematics & Statistics for Business 4e by John S. Croucher 1 n Learning Objectives –Identify.
Geo597 Geostatistics Ch9 Random Function Models.
Lecture 2 Forestry 3218 Lecture 2 Statistical Methods Avery and Burkhart, Chapter 2 Forest Mensuration II Avery and Burkhart, Chapter 2.
1 Chapter 7 Sampling Distributions. 2 Chapter Outline  Selecting A Sample  Point Estimation  Introduction to Sampling Distributions  Sampling Distribution.
1 Functions of a Random Variable Let X be a r.v defined on the model and suppose g(x) is a function of the variable x. Define Is Y necessarily a r.v? If.
1 6. Mean, Variance, Moments and Characteristic Functions For a r.v X, its p.d.f represents complete information about it, and for any Borel set B on the.
Random Variables (1) A random variable (also known as a stochastic variable), x, is a quantity such as strength, size, or weight, that depends upon a.
PROBABILITY AND STATISTICS FOR ENGINEERING Hossein Sameti Department of Computer Engineering Sharif University of Technology Mean, Variance, Moments and.
CHAPTER FIVE SOME CONTINUOUS PROBABILITY DISTRIBUTIONS.
Lecture 4 Dustin Lueker.  The population distribution for a continuous variable is usually represented by a smooth curve ◦ Like a histogram that gets.
Normal Distribution * Numerous continuous variables have distribution closely resemble the normal distribution. * The normal distribution can be used to.
1 6. Mean, Variance, Moments and Characteristic Functions For a r.v X, its p.d.f represents complete information about it, and for any Borel set B on the.
CHAPTER 2: Basic Summary Statistics
CHAPTER 2.3 PROBABILITY DISTRIBUTIONS. 2.3 GAUSSIAN OR NORMAL ERROR DISTRIBUTION  The Gaussian distribution is an approximation to the binomial distribution.
CHAPTER – 1 UNCERTAINTIES IN MEASUREMENTS. 1.3 PARENT AND SAMPLE DISTRIBUTIONS  If we make a measurement x i in of a quantity x, we expect our observation.
© 1999 Prentice-Hall, Inc. Chap Statistics for Managers Using Microsoft Excel Chapter 6 The Normal Distribution And Other Continuous Distributions.
Confidence Intervals. Point Estimate u A specific numerical value estimate of a parameter. u The best point estimate for the population mean is the sample.
This represents the most probable value of the measured variable. The more readings you take, the more accurate result you will get.
THE NORMAL DISTRIBUTION
Chapter 8 Confidence Intervals Copyright © 2014 by The McGraw-Hill Companies, Inc. All rights reserved.McGraw-Hill/Irwin.
Chapter 6 The Normal Distribution and Other Continuous Distributions
Chapter 3 Probability Distribution
Confidence Intervals and Sample Size
Sampling Distributions
ESTIMATION.
Chapter 6 Confidence Intervals.
Alexey Roenko1, Vladimir Lukin1, Sergey Abramov1, Igor Djurovic2
SOME IMPORTANT PROBABILITY DISTRIBUTIONS
Chapter 7 Sampling Distributions.
Econ 3790: Business and Economics Statistics
Chapter 6 Confidence Intervals.
Arithmetic Mean This represents the most probable value of the measured variable. The more readings you take, the more accurate result you will get.
Chapter 5 Continuous Random Variables and Probability Distributions
5. Functions of a Random Variable
Normal Distribution “Bell Curve”.
C19: Unbiased Estimators
6.3 Sampling Distributions
CHAPTER 2: Basic Summary Statistics
CHAPTER – 1.2 UNCERTAINTIES IN MEASUREMENTS.
Sampling Distributions
2.3. Measures of Dispersion (Variation):
C19: Unbiased Estimators
MBA 510 Lecture 4 Spring 2013 Dr. Tonya Balan 10/30/2019.
CHAPTER – 1.2 UNCERTAINTIES IN MEASUREMENTS.
Presentation transcript:

Analysis and Selection of Myriad Estimate Tuning Parameter For SαS Distributions Roenko A.A., Lukin V.V., Djurović I.

Symmetric α-stable distributions and their properties Characteristic function of random variable Х with SαS distribution: where α denotes the characteristic exponent or index and γ is dispersion; α=2 corresponds to Gaussian distribution with zero mean and variance 2γ: α=1 describes Cauchy distribution with location parameter equal to 0: Fig.1. SαS distributions for various α values 1 (1) (2) (3) Fig.2. Realizations of processes with SαS distributions for α=0.5 and different γ values equal to 0.5, 1 and 4, correspondingly γ 1/α denotes parameter characterizing the scale of the data sample with SαS pdf.

Sample myriad based location parameter estimator The sample myriad relates to the class of М-estimators and is the efficient (optimal) estimator for Cauchy distribution. For the data sample х 1, х 2,…,х N and parameter К>0 it is defined as (5) An infinite values of K converts the myriad to sample mean: (4) When K tends to 0 the sample myriad precisely defines the pdf mode: (6)(6) The behavior of myriad estimator in case of fixed values of tuning parameter depends upon the values of processing realization: (7)(7) where с>0. Fig.3. The dependence of myriad estimator properties upon the K values 2

Analysis of known dependencies of optimal К values upon SαS distribution parameters 3 Fig.4. The dependence K opt Arce ( α ) for γ=1 proposed in the works of G. Arce and J. Gonzalez The number of experiments was fixed and equal to The α value was varied with step size 0.1 and the increment of the parameter K value in case of fixed α and γ was equal to The approximation formula for calculation of optimal К values proposed in the works of G.Arce and J.Gonzalez: (8)(8) Fig.5. Dependence σ 2 (K) for N=128 and 256; α=1.4, γ=1 Since the minima of the dependencies of σ 2 min (K opt ) for fixed α and γ are commonly not very obvious (see Fig.5) and because of the fact that small variations of К values (accordingly to Fig.5) dont considerably influence the estimator accuracy, we determined not only the optimal value К opt (α,γ) but also the values К opt min (α, γ) and К opt max (α, γ). Then σ 2 (К min, α, γ)= σ 2 (К max, α, γ)=1,1σ 2 min (К opt, α, γ) (9)

Analysis of obtained dependencies Irrespectively to γ, the functions К opt min (α) and К opt max (α) are monotonically increase if α grows. The following conditions are valid: for α0, К opt (α,γ) 0; for α1,К opt (α,γ) γ; for α2, К opt (α,γ) (note that the curve К opt (α) goes in between the curves К opt min (α) and К opt max (α) for any given γ). Arces approximations К opt Arce (α,γ) are not absolutely correct: -for 0<α<1 the values К opt Arce (α) are larger than К opt (α); -for 1α<1,8 the values К opt Arce (α) are slightly smaller than К opt (α); -for γ>1 and 0<α<1 the values К opt Arce (α) start to quickly increase if α reduces; this fact is explained by the behavior of the factor γ 1/α used in Arces approximation formula. Thus, one needs more accurate approximation К opt (α,γ), especially for γ>1. 4 Fig.6. The plots of obtained optimal К values and К opt Arce (α,γ) curves depending upon α for fixed values of γ equal to 0.5, 1 and 4

The proposed approximation of obtained dependencies К opt (α,γ) and its analysis 5 We propose to use the following approximation formula: (10) Fig.7. The plots К opt min (α, γ), К opt max (α, γ) and К optAp (α, γ) obtained by (10) for fixed γ values equal to 0.5, 1 and 4

Adaptation parameters to the distribution tail heaviness and data scale 6 The percentile coefficient of kurtosis (PCK): (11) where Q=(Q 3 -Q 1 )/2 is the half of interquantile range; P 90, P 10 denotes the 90th and 10th percentiles. Absolute median deviation from median (MAD): (12) where median denotes the sample median; x 1, x 2,…, x N are the order statistics data sample with size N. Fig.8. The dependence of average PCK values upon the parameter α The MSE estimate for the case of Gaussian noise (α=2) is equal to (13) For the SαS pdf with α=2 and γ we can define: (14) Then (15)

Analysis of PCK values for the cases of Cauchy and Gaussian distributions 7 Cauchy distribution with zero location parameter and scale parameter σ 2 С =γ 2 can be defined as: (16) Then, for Р 90 -Р 10 : (17) Taking into account that the distribution is the symmetric one obtains: (18) Since Р 50 coincides with the distribution location equal to 0, we can write: (19) Thus,Similarly, Q = Q 3 - Q 1 = Р 75 - Р 25 = σ C. Consequently, Similarly for Gaussian distribution with zero-mean and variance σ 2 G one obtains PCK=

Proposed adaptive procedure for calculation of tuning parameter value 8 Proposed adaptive procedure for determination of the parameter K value in case of a priori unknown α and γ: (20) Fig.9. The plots К opt min (α, γ), К opt max (α, γ) and К ADAPT (α, γ) for fixed values of parameter γ equal to 0.5, 1 и 4