Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Chapter 4 Continuous Random Variables and their Probability Distributions The Theoretical Continuous Distributions starring The Rectangular The Normal The Exponential and The Weibull Chapter 4B Continuous Uniform Distribution A continuous RV X with probability density function X Rect(a, b) 1 f ( x) , a xb ba has a continuous uniform distribution or rectangular distribution 1 x' xa F ( x) dx ' ba ba a ba a x x E( X ) b a 2 x x dx ba 2(b a ) b a b b a a V ( X ) x 2 f ( x)dx 2 ab 2 a b 2 x a b ( b a ) dx ba 12 2 2 2 4-5 Continuous Uniform Random Variable Mean and Variance Using Continuous PDF’s b Given a pdf, f(x), a <= x <= b and f ( x)dx 1 a and a <= m < n <= b n n P(m <= x <= n) = f ( x)dx F x F (n) F (m) m m If f ( x) 0.05, 0 x 20 10 P(5 x 10) 0.05dx 0.05 x 5 0.05(10 5) 0.25 10 5 20 P(10 x 30) 0.05dx 0.05 x 10 0.05(20 10) 0.50 10 20 Problem 4-33 X Rect(1,1) b a 1 1 0 2 2 b a 2 1 1 2 1 1 , 0.577 12 12 3 3 x x 1 P( x X x) 0.90 f (t )dt dt x x b a x 1 1 x 1 = dt t x ( x x) x x 11 2 2 x 0.90 2 x (1) x 1 F ( x) 1 (1) 2 Let’s get Normal Most widely used distribution; bell shaped curve Histograms often resemble this shape Often seen in experimental results if a process is reasonably stable & deviations result from a very large number of small effects – central limit theorem. Variables that are defined as sums of other random variables also tend to be normally distributed – again, central limit theorem. If the experimental process is not stable, some systematic trend is likely present (e.g., machine tool has worn excessively) a normal distribution will not result. 4-6 Normal Distribution Definition X n( , 2 ) 4-6 Normal Distribution The Normal PDF http://www.stat.ucla.edu/~dinov/courses_students.dir/Applets.dir/NormalCurveInteractive.html Normal IQs 4-6 Normal Distribution Some useful results concerning the normal distribution Normal Distributions Standard Normal Distribution 0 and is 1called a standard normal A normal RV with RV and is denoted as Z. Appendix A Table III provides probabilities of the form P(Z < z) where ( z ) P( Z z ) 2 You cannot integrate the normal density function in closed form. Fig 4-13. Standard Normal Probability Function Examples – standard normal P(Z > 1.26) = 1 – P(Z 1.26) = 1 - .89616 = .10384 P(Z < -0.86) = .19490 P(Z > -1.37) = P(z < 1.37) = .91465 P(-1.25< Z<0.37) = P(Z<.0.37) – P(Z<-1.25) = .64431 - .10565 = .53866 P(Z < -4.6) = not found in table; prob calculator = .0000021 P(Z > z) = 0.05; P(Z < z) =.95; from tables z 1.65; from prob calc = 1.6449 P(-z < Z < z) = 0.99; P(Z<z) =.995; z = 2.58 Converting Normal RV’s to Standard Normal Variates (so you can use the tables!) Any arbitrary normal RV can be converted to a standard normal RV using the following formula: After this transformation, Z ~ N(0, 1) Z X the number of standard deviations from the mean X E[ X ] E[ Z ] E 0 2 X 1 V [Z ] V 2 V [ X ] V 2 1 4-6 Normal Distribution To Calculate Probability Converting Normal RV’s to Standard Normal Variates (an example) Z X For example, if X ~ N(10, 4) To determine P(X > 13): X 13 10 P X 13 P P z 1.5 2 1 P z 1.5 1 0.93319 .06681 from Table III Converting Normal RV’s A scaling and a shift are involved. More Normal vs. Std Normal RV X ~ N(10,4) 9 10 X 11 10 P 9 X 11 P P .5 z .5 2 2 P z .5 P z .5 0.69146 0.30854 0.38292 Example 4-14 Continued X ~ N(10,4) (sometimes you need to work backward Determine the value of x such that P(X x) = 0.98 x 10 X 10 x 10 P( X x) P P Z 0.98 2 2 2 Table II: P(Z z ) 0.98 P(Z 2.05) 0.97982 x 10 ==> = 2.05 2 x = 14.1 That is, there is a 98% probability that a current measurement is less than 14.1 Check out this website http://www.ms.uky.edu/~mai/java/stat/GaltonMachine.html An Illustration of Basic Probability: The Normal Distribution See the normal curve generated right in front of your very own eyes 4-8 Exponential Distribution Definition X Exp( ) The Shape of Things Exponential Probability Distribution 1.2 1 f(X) 0.8 0.6 0.4 0.2 0 0 1 2 3 4 X lambda = .1 lambda = .5 lambda = 1.0 5 The Mean, Variance, and CDF xe x 0 dx xe 2 x 0 x F ( x) e 0 x 0 x e 2 u 1 1 dx 2 2 2 1 1 2 1 dx 3 2 u x e x x du e 1 1 e 0 table of definite integrals What about the median? F ( x) 1 e x .5 x e .5 x ln .5 1 x ln .5 ln .5 .6931472 Next Example Let X = a continuous random variable, the time to failure in operating hours of an electronic circuit f(x) = (1/25) e-x/25 X Exp( 1/ 25 hr) F(x) = 1 - e-x/25 = 1/ = E[X] = 25 hours median = .6931472 (25) = 17.3287 hours 2 = V[X] = 252 = 25 Example X Exp( 1/ 25 hr) What is the probability there are no failures for 6 hours? P( X 6) 6 6 1 25x e dx e 25 0.7866 25 What is the probability that the time until the next failure is between 3 and 6 hours? x 25 F ( x) 1 e P(3 X 6) F (6) F (3) .2134 .1131 .1003 Exponential & Lack of Memory Property: If X ~ exponential P( X t1 t2 X t1 ) P ( X t2 ) This implies that knowledge of previous results (past history) does not affect future events. An exponential RV is the continuous analog of a geometric RV & they both share this lack of memory property. Example: The probability that no customer arrives in the next ten minutes at a checkout counter is not affected by the time since the last customer arrival. Essentially, it does not become more likely (as time goes by without a customer) that a customer is going to arrive. Chapter Two stuff! P(A | B) P(A B) / P(B) Proof of Memoryless Property Pr t1 X t1 t2 Pr X t1 t2 | X t1 Pr Pr X t1 t1 ( t1 t2 ) 1 e 1 e F (t1 t2 ) F (t1 ) 1 F (t1 ) e t1 e t1 t1 t2 e e e t1 e t1 1 e t2 e t1 F t2 Pr X t2 A – the event that X < t1 + t2 and B – the event that X > t1 Exponential as the Flip Side of the Poisson If time between events is exponentially distributed, then the number of events in any interval has a Poisson distribution. NT events till time T Time 0 Time between events has exponential distribution Time T Exponential and Poisson Let X(t) = the number of events that occur in time t; assume X(t) ~ Pois(t) then E[X(t)] = t Pr X (t ) n t e t n n! , for n 0,1, 2,...; Let T = the time until the next event; assume T ~ Exp() then E[T] = 1/ Pr T t 1 F (t ) et Pr X (t ) 0 4-10 Weibull Distribution Definition X W ( , ) The PDF in Graphical Splendor 1.0 f(t) 0.9 Beta 0.8 0.5 0.7 1.5 2.0 0.6 4.0 Delta = 2 0.5 0.4 0.3 0.2 0.1 0.0 0.0 -0.1 1.0 2.0 3.0 4.0 5.0 6.0 t More Splendor 1.6 f(t) 1.4 Delta 1.2 0.5 1.0 1 2.0 0.8 0.6 Beta = 1.5 0.4 0.2 0 0.0 1.0 2.0 3.0 4.0 5.0 6.0 t 4-10 Weibull Distribution The Gamma Function (x) = the gamma function = zy 0 x-1 -y (x) = (x - 1) (x - 1) fine print: easier method is to use the prob calculator e dy 4-10 Weibull Distribution Example 4-25 The Mode of a Distribution a measure of central tendency f(t) 0.06 0.05 0.04 0.03 0.02 0.01 0 0 10 20 30 40 50 60 The Mode of a Distribution a measure of central tendency f(t) 0.06 0.05 0.04 0.03 0.02 0.01 0 0 10 20 30 40 50 60 The Mode of a Distribution MAX f(x) = x0 df(x) ( - 1) x = 2 dx e x - x -2 -2 2 x ( - 1) = 0 e -1 x - x e x - x - 2 2 2 -2 e x - 0 x ( -1) =0 1 -1 Mode = for 1 A Weibull Example X W (80, 2.4) The design life of the members used in constructing the roof of the Weibull Building, a engineering marvel, has a Weibull distribution with = 80 years and = 2.4. Pr{ X 100} 1 F (100) 1 1 e 1 80 1 80 1.42 70.92 yr. 2.4 100 80 1 2.4 2.4 - 1 Mode = 80 63.91yr. 2.4 2.4 .1812 Other Continuous Distributions Worth Knowing Gamma Erlang is a special case of the gamma Beta Like the triangular – used in the absence of data Used to model random proportions Lognormal Used in queuing analysis used to model repair times (maintainability) quantities that are a product of other quantities (central limit theorem) Pearson Type V and Type VI like lognormal – models task times Picking a Distribution We now have some distributions at our disposal. Selecting one as an appropriate model is a combination of understanding the physical situation and data-fitting Some situations imply a distribution, e.g. arrivals Poisson process is a good guess. Collected data can be tested statistically for a ‘fit’ to distributions. Next Week – Chapter 5 Double our pleasure by considering joint distributions.