Simple root finding and one dimensional integrals algorithms were implemented in previous posts.

These algorithms can be used to estimate the cumulative probabilities and quantiles.

Here, take normal distribution as an example.

Normal distribution is defined as:

#probability density function
y.dnorm <- function(x, mean=0, sd=1) exp(-(x-mean)^2/(2*sd^2))/sqrt(2*pi*sd^2)

The cumulative probabilities can be estimated by integrating the PDF function. Here, using function *simpson_v2*, which implemented in previous post, for integral calculation.

y.cdf <- function(pdf=y.dnorm, q) simpson_v2(pdf, -Inf, q)

The quantile function is mathematically the inverse of the cumulative distribution function. Here, the quantile was estimated by using newton-raphson algorithm to find the root of function CDF(q) - p = 0.

1
2
3
4
5
6
7
8
9
10
11

y.qf <- function(pdf=y.dnorm, p, tol=1e-7, niter=100) {
	x0 = 0
	for (i in 1:niter) {
		fx <- y.cdf(pdf, x0) - p
		x = x0 - fx/pdf(x0)
		if (abs(fx) < tol)
			return(x)
		x0 = x
	}
	stop("exceeded allowed number of iterations")	
}

> y.dnorm(1)
[1] 0.2419707
> x=y.cdf(pdf=y.dnorm, q=1.64)
> y.qf(pdf=y.dnorm, p=x)
[1] 1.64
> qnorm(p=x)
[1] 1.636615

Related Posts

• January 1, 2011 -- Single variable optimization (0)
• December 25, 2010 -- one-dimensional integrals (2)
• December 23, 2010 -- Project Euler — Problem 187 (2)
• December 4, 2010 -- Root finding (4)
• December 1, 2010 -- bubble chart by using ggplot2 (0)
• December 1, 2010 -- The avalanche of publications mentioning GO (0)
• November 30, 2010 -- GOSemSim redesign in terms of S4 classes (1)
• October 20, 2010 -- upgrade R – F77 cause compilation error (0)
• October 19, 2010 -- Listing gene IDs from hyperGTest (0)
• October 15, 2010 -- The S3 OOP system (0)