Sampling from the generalized logarithmic series distribution (Q1362834): Difference between revisions
From MaRDI portal
Created a new Item |
Added link to MaRDI item. |
||
links / mardi / name | links / mardi / name | ||
Revision as of 15:40, 31 January 2024
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Sampling from the generalized logarithmic series distribution |
scientific article |
Statements
Sampling from the generalized logarithmic series distribution (English)
0 references
3 February 1998
0 references
The generalized logarithmic series distribution (GLSD) of \textit{G. C. Jain} and \textit{R. P. Gupta} [Trabajos. Estadist. Invest. Oper. 24, No. 1/2, 99-105 (1973; Zbl 0267.60012)] has the probability function \( P(X=x)={1 \over \beta x} {\beta x \choose x} \theta^x (1-\theta)^{\beta x-x} / (-log(1-\theta)) \) for \(x=1,2,\dots\) and zero otherwise where \(0< \theta \leq \theta\beta < 1\). The ordinary logarithmic distribution is the special case \(\beta=1\). The author presents four different algorithms for generating pseudo-random variates from the GLSD. Some of the theoretical bases for the four algorithms are given in the article. Others can be found in the references given. The first method described is the `table-look-up' or inversion method, which uses a recurrence relation between the GLSD probabilities for a sequential search. A rejection algorithm based on the inequality \( P(X=x) \leq (1+\sqrt{2})({1 \over \sqrt{x}}-{1 \over \sqrt{x+1}})(-log(1-\theta) \sqrt{\pi\beta (\beta-1)})^{-1}\) for \(\beta>1\), and a rejection algorithm based on the monotone property of the probability vector with \(P(X=x) \leq 1/x\) for \(x=1,2,\dots\) are also provided. The fourth method given is a branching process algorithm for generating generalized logarithmic series variates which does not need to compute the GLSD probabilities. The algorithms are compared by their execution times. Based on the result of the comparison the author recommends a modified algorithm which is a combination of the inversion method and the branching process method depending on the size of the two parameters of the generalized logarithmic series distribution. For \(\theta\beta\leq 0.45\), the invervsion method is recommended and for \(\theta\beta > 0.45\) the author recommends the use of the branching method.
0 references
generalized logarithmic series distribution
0 references
random number generation
0 references
inversion method
0 references
rejection method
0 references
branching method
0 references
algorithms
0 references
pseudo-random variates
0 references