Related
Given a Laplace Distribution proposal:
g(x) = 1/2*e^(-|x|)
and sample size n = 1000, I want to Conduct the Monte Carlo (MC) integration for estimating θ:
via importance sampling. Eventually I want to calculate the mean and standard deviation of this MC estimate in R once I get there.
Edit (arrived late after the answer below)
This is what I have for my R code so far:
library(VGAM)
n = 1000
x = rexp(n,0.5)
hx = mean(2*exp(-sqrt(x))*(sin(x))^2)
gx = rlaplace(n, location = 0, scale = 1)
Now we can write a simple R function to sample from Laplace distribution:
## `n` is sample size
rlaplace <- function (n) {
u <- runif(n, 0, 1)
ifelse(u < 0.5, log(2 * u), -log(2* (1 - u)))
}
Also write a function for density of Laplace distribution:
g <- function (x) ifelse(x < 0, 0.5 * exp(x), 0.5 * exp(-x))
Now, your integrand is:
f <- function (x) {
ifelse(x > 0, exp(-sqrt(x) - 0.5 * x) * sin(x) ^ 2, 0)
}
Now we estimate the integral using 1000 samples (set.seed for reproducibility):
set.seed(0)
x <- rlaplace(1000)
mean(f(x) / g(x))
# [1] 0.2648853
Also compare with numerical integration using quadrature:
integrate(f, lower = 0, upper = Inf)
# 0.2617744 with absolute error < 1.6e-05
Given the following function:
f(x) = (1/2*pi)(1/(1+x^2/4))
How do I identify it's distribution and write this distribution function in R?
So this is your function right now (hopefully you know how to write an R function; if not, check writing your own function):
f <- function (x) (pi / 2) * (1 / (1 + 0.25 * x ^ 2))
f is defined on (-Inf, Inf) so integration on this range gives an indefinite integral. Fortunately, it approaches to Inf at the speed of x ^ (-2), so the integral is well defined, and can be computed:
C <- integrate(f, -Inf, Inf)
# 9.869604 with absolute error < 1e-09
C <- C$value ## extract integral value
# [1] 9.869604
Then you want to normalize f, as we know that a probability density should integrate to 1:
f <- function (x) (pi / 2) * (1 / (1 + 0.25 * x ^ 2)) / C
You can draw its density by:
curve(f, from = -10, to = 10)
Now that I have the probably distribution function I was wondering how to create a random sample of say n = 1000 using this new distribution function?
An off-topic question, but OK to answer without your making a new thread. Useful as it turns out subtle.
Compare
set.seed(0); range(simf(1000, 1e-2))
#[1] -56.37246 63.21080
set.seed(0); range(simf(1000, 1e-3))
#[1] -275.3465 595.3771
set.seed(0); range(simf(1000, 1e-4))
#[1] -450.0979 3758.2528
set.seed(0); range(simf(1000, 1e-5))
#[1] -480.5991 8017.3802
So I think e = 1e-2 is reasonable. We could draw samples, make a (scaled) histogram and overlay density curve:
set.seed(0); x <- simf(1000)
hist(x, prob = TRUE, breaks = 50, ylim = c(0, 0.16))
curve(f, add = TRUE, col = 2, lwd = 2, n = 201)
suppose that we have a pde that describes the evolution of a variable y(t,x) over time t and space x, and I would like to plot its evolution on a three dimensional diagram (t,x,y). With deSolve I can solve the pde, but I have no idea about how to obtain this kind of diagram.
The example in the deSolve package instruction is the following, where y is aphids, t=0,...,200 and x=1,...,60:
library(deSolve)
Aphid <- function(t, APHIDS, parameters) {
deltax <- c (0.5, rep(1, numboxes - 1), 0.5)
Flux <- -D * diff(c(0, APHIDS, 0)) / deltax
dAPHIDS <- -diff(Flux) / delx + APHIDS * r
list(dAPHIDS )
}
D <- 0.3 # m2/day diffusion rate
r <- 0.01 # /day net growth rate
delx <- 1 # m thickness of boxes
numboxes <- 60
Distance <- seq(from = 0.5, by = delx, length.out = numboxes)
APHIDS <- rep(0, times = numboxes)
APHIDS[30:31] <- 1
state <- c(APHIDS = APHIDS) # initialise state variables
times <-seq(0, 200, by = 1)
out <- ode.1D(state, times, Aphid, parms = 0, nspec = 1, names = "Aphid")
"out" produces a matrix containing all the data that we need, t, y(x1), y(x2), ... y(x60). How can I produce a surface plot to show the evolution and variability of y in (t,x)?
The ways change a bit depending on using package. But you can do it with little cost because out[,-1] is an ideal matrix form to draw surface. I showed two examples using rgl and plot3D package.
out2 <- out[,-1]
AphID <- 1:ncol(out2)
library(rgl)
persp3d(times, AphID, out2, col="gray50", zlab="y")
# If you want to change color with value of Z-axis
# persp3d(times, AphID, out2, zlab="y", col=topo.colors(256)[cut(c(out2), 256)])
library(plot3D)
mat <- mesh(times, AphID)
surf3D(mat$x, mat$y, out2, bty="f", ticktype="detailed", xlab="times", ylab="AphID", zlab="y")
I am trying to solve for the parameters of a gamma distribution that is convolved with both normal and lognormal distributions. I can experimentally derive parameters for both the normal and lognormal components, hence, I just want to solve for the gamma params.
I have attempted 3 approaches to this problem:
1) generating convolved random datasets (i.e. rnorm()+rlnorm()+rgamma()) and using least-squares regression on the linear- or log-binned histograms of the data (not shown, but was very biased by RNG and didn't optimize well at all.)
2) "brute-force" numerical integration of the convolving functions (example code #1)
3) numerical integration approaches w/ the distr package. (example code #2)
I have had limited success with all three approaches. Importantly, these approaches seem to work well for "nominal" values for the gamma parameters, but they all begin to fail when k(shape) is low and theta(scale) is high—which is where my experimental data resides. please find the examples below.
Straight-up numerical Integration
# make the functions
f.N <- function(n) dnorm(n, N[1], N[2])
f.L <- function(l) dlnorm(l, L[1], L[2])
f.G <- function(g) dgamma(g, G[1], scale=G[2])
# make convolved functions
f.Z <- function(z) integrate(function(x,z) f.L(z-x)*f.N(x), -Inf, Inf, z)$value # L+N
f.Z <- Vectorize(f.Z)
f.Z1 <- function(z) integrate(function(x,z) f.G(z-x)*f.Z(x), -Inf, Inf, z)$value # G+(L+N)
f.Z1 <- Vectorize(f.Z1)
# params of Norm, Lnorm, and Gamma
N <- c(0,5)
L <- c(2.5,.5)
G <- c(2,7) # this distribution is the one we ultimately want to solve for.
# G <- c(.5,10) # 0<k<1
# G <- c(.25,5e4) # ballpark params of experimental data
# generate some data
set.seed(1)
rN <- rnorm(1e4, N[1], N[2])
rL <- rlnorm(1e4, L[1], L[2])
rG <- rgamma(1e4, G[1], scale=G[2])
Z <- rN + rL
Z1 <- rN + rL + rG
# check the fit
hist(Z,freq=F,breaks=100, xlim=c(-10,50), col=rgb(0,0,1,.25))
hist(Z1,freq=F,breaks=100, xlim=c(-10,50), col=rgb(1,0,0,.25), add=T)
z <- seq(-10,50,1)
lines(z,f.Z(z),lty=2,col="blue", lwd=2) # looks great... convolution performs as expected.
lines(z,f.Z1(z),lty=2,col="red", lwd=2) # this works perfectly so long as k(shape)>=1
# I'm guessing the failure to compute when shape 0 < k < 1 is due to
# numerical integration problems, but I don't know how to fix it.
integrate(dgamma, -Inf, Inf, shape=1, scale=1) # ==1
integrate(dgamma, 0, Inf, shape=1, scale=1) # ==1
integrate(dgamma, -Inf, Inf, shape=.5, scale=1) # !=1
integrate(dgamma, 0, Inf, shape=.5, scale=1) # != 1
# Let's try to estimate gamma anyway, supposing k>=1
optimFUN <- function(par, N, L) {
print(par)
-sum(log(f.Z1(Z1[1:4e2])))
}
f.G <- function(g) dgamma(g, par[1], scale=par[2])
fitresult <- optim(c(1.6,5), optimFUN, N=N, L=L)
par <- fitresult$par
lines(z,f.Z1(z),lty=2,col="green3", lwd=2) # not so great... likely better w/ more data,
# but it is SUPER slow and I observe large step sizes.
Attempting convolving via distr package
# params of Norm, Lnorm, and Gamma
N <- c(0,5)
L <- c(2.5,.5)
G <- c(2,7) # this distribution is the one we ultimately want to solve for.
# G <- c(.5,10) # 0<k<1
# G <- c(.25,5e4) # ballpark params of experimental data
# make the distributions and "convolvings'
dN <- Norm(N[1], N[2])
dL <- Lnorm(L[1], L[2])
dG <- Gammad(G[1], G[2])
d.NL <- d(convpow(dN+dL,1))
d.NLG <- d(convpow(dN+dL+dG,1)) # for large values of theta, no matter how I change
# getdistrOption("DefaultNrFFTGridPointsExponent"), grid size is always wrong.
# Generate some data
set.seed(1)
rN <- r(dN)(1e4)
rL <- r(dL)(1e4)
rG <- r(dG)(1e4)
r.NL <- rN + rL
r.NLG <- rN + rL + rG
# check the fit
hist(r.NL, freq=F, breaks=100, xlim=c(-10,50), col=rgb(0,0,1,.25))
hist(r.NLG, freq=F, breaks=100, xlim=c(-10,50), col=rgb(1,0,0,.25), add=T)
z <- seq(-10,50,1)
lines(z,d.NL(z), lty=2, col="blue", lwd=2) # looks great... convolution performs as expected.
lines(z,d.NLG(z), lty=2, col="red", lwd=2) # this appears to work perfectly
# for most values of K and low values of theta
# this is looking a lot more promising... how about estimating gamma params?
optimFUN <- function(par, dN, dL) {
tG <- Gammad(par[1],par[2])
d.NLG <- d(convpow(dN+dL+tG,1))
p <- d.NLG(r.NLG)
p[p==0] <- 1e-15 # because sometimes very low probabilities evaluate to 0...
# ...and logs don't like that.
-sum(log(p))
}
fitresult <- optim(c(1,1e4), optimFUN, dN=dN, dL=dL)
fdG <- Gammad(fitresult$par[1], fitresult$par[2])
fd.NLG <- d(convpow(dN+dL+fdG,1))
lines(z,fd.NLG(z), lty=2, col="green3", lwd=2) ## this works perfectly when ~k>1 & ~theta<100... but throws
## "Error in validityMethod(object) : shape has to be positive" when k decreases and/or theta increases
## (boundary subject to RNG).
Can i speed up the integration in example 1? can I increase the grid size in example 2 (distr package)? how can I address the k<1 problem? can I rescale the data in a way that will better facilitate evaluation at high theta values?
Is there a better way all-together?
Help!
Well, convolution of function with gaussian kernel calls for use of Gauss–Hermite quadrature. In R it is implemented in special package: https://cran.r-project.org/web/packages/gaussquad/gaussquad.pdf
UPDATE
For convolution with Gamma distribution this package might be useful as well via Gauss-Laguerre quadrature
UPDATE II
Here is quick code to convolute gaussian with lognormal,
hopefully not a lot of bugs and and prints some reasonable looking graph
library(gaussquad)
n.quad <- 170 # integration order
# get the particular weights/abscissas as data frame with 2 observables and n.quad observations
rule <- ghermite.h.quadrature.rules(n.quad, mu = 0.0)[[n.quad]]
# test function - integrate 1 over exp(-x^2) from -Inf to Inf
# should get sqrt(pi) as an answer
f <- function(x) {
1.0
}
q <- ghermite.h.quadrature(f, rule)
print(q - sqrt(pi))
# convolution of lognormal with gaussian
# because of the G-H rules, we have to make our own function
# for simplicity, sigmas are one and mus are zero
sqrt2 <- sqrt(2.0)
c.LG <- function(z) {
#print(z)
f.LG <- function(x) {
t <- (z - x*sqrt2)
q <- 0.0
if (t > 0.0) {
l <- log(t)
q <- exp( - 0.5*l*l ) / t
}
q
}
ghermite.h.quadrature(Vectorize(f.LG), rule) / (pi*sqrt2)
}
library(ggplot2)
p <- ggplot(data = data.frame(x = 0), mapping = aes(x = x))
p <- p + stat_function(fun = Vectorize(c.LG))
p <- p + xlim(-1.0, 5.0)
print(p)
I am interested to build a R function that I can use to test the limits of the Taylor series approximation. I am aware that there is limits to what I am doing, but it's exactly those limits I wish to investigate.
I have two normally distributed random variables x and y. x has a mean of 7 and a standard deviation (sd) of 1. y has a mean of 5 and a sd of 4.
me.x <- 4; sd.x <- 1
me.y <- 5; sd.y <- 4
I know how to estimate the mean ratio of y/x, like this
# E(y/x) = E(y)/E(x) - Cov(y,x)/E(x)^2 + Var(x)*E(y)/E(x)^3
me.y/me.x - 0/me.x^2 + sd.x*me.y/me.x^3
[1] 1.328125
I am however stuck on how to estimate the Standard Deviation of the ratio? I realize I have to use a Taylor expansion, but not how to use it.
Doing a simple simulation I get
x <- rnorm(10^4, mean = 4, sd = 1); y <- rnorm(10^4, mean = 5, sd = 4)
sd(y/x)
[1] 2.027593
mean(y/x)[1]
1.362142
There is an analytical expression for the PDF of the ratio of two gaussians, done
by David Hinkley (e.g. see Wikipedia). So we could compute all momentums, means etc. I typed it and apparently it clearly doesn't have finite second momentum, thus it doesn't have finite standard deviation. Note, I've denoted your Y gaussian as my X, and your X as my Y (formulas assume X/Y). I've got mean value of ratio pretty close to the what you've got from simulation, but last integral is infinite, sorry. You could sample more and more values, but from sampling std.dev is growing as well, as noted by #G.Grothendieck
library(ggplot2)
m.x <- 5; s.x <- 4
m.y <- 4; s.y <- 1
a <- function(x) {
sqrt( (x/s.x)^2 + (1.0/s.y)^2 )
}
b <- function(x) {
(m.x*x)/s.x^2 + m.y/s.y^2
}
c <- (m.x/s.x)^2 + (m.y/s.y)^2
d <- function(x) {
u <- b(x)^2 - c*a(x)^2
l <- 2.0*a(x)^2
exp( u / l )
}
# PDF for the ratio of the two different gaussians
PDF <- function(x) {
r <- b(x)/a(x)
q <- pnorm(r) - pnorm(-r)
(r*d(x)/a(x)^2) * (1.0/(sqrt(2.0*pi)*s.x*s.y)) * q + exp(-0.5*c)/(pi*s.x*s.y*a(x)^2)
}
# normalization
nn <- integrate(PDF, -Inf, Inf)
nn <- nn[["value"]]
# plot PDF
p <- ggplot(data = data.frame(x = 0), mapping = aes(x = x))
p <- p + stat_function(fun = function(x) PDF(x)/nn) + xlim(-2.0, 6.0)
print(p)
# first momentum
m1 <- integrate(function(x) x*PDF(x), -Inf, Inf)
m1 <- m1[["value"]]
# mean
print(m1/nn)
# some sampling
set.seed(32345)
n <- 10^7L
x <- rnorm(n, mean = m.x, sd = s.x); y <- rnorm(n, mean = m.y, sd = s.y)
print(mean(x/y))
print(sd(x/y))
# second momentum - Infinite!
m2 <- integrate(function(x) x*x*PDF(x), -Inf, Inf)
Thus, it is impossible to test any Taylor expansion for std.dev.
With the cautions suggested by #G.Grothendieck in mind: a useful mnemonic for products and quotients of independent X and Y variables is
CV^2(X/Y) = CV^2(X*Y) = CV^2(X) + CV^2(Y)
where CV is the coefficient of variation (sd(X)/mean(X)), so CV^2 is Var/mean^2. In other words
Var(Y/X)/(m(Y/X))^2 = Var(X)/m(X)^2 + Var(Y)/m(Y)^2
or rearranging
sd(Y/X) = sqrt[ Var(X)*m(Y/X)^2/m(X)^2 + Var(Y)*m(Y/X)^2/m(Y)^2 ]
For random variables with the mean well away from zero, this is a reasonable approximation.
set.seed(101)
y <- rnorm(1000,mean=5)
x <- rnorm(1000,mean=10)
myx <- mean(y/x)
sqrt(var(x)*myx^2/mean(x)^2 + var(y)*myx^2/mean(y)^2) ## 0.110412
sd(y/x) ## 0.1122373
Using your example is considerably worse because the CV of Y is close to 1 -- I initially thought it looked OK, but now I see that it's biased as well as not capturing the variability very well (I'm also plugging in the expected values of the mean and SD rather than their simulated values, but for such a large sample that should be a minor part of the error.)
me.x <- 4; sd.x <- 1
me.y <- 5; sd.y <- 4
myx <- me.y/me.x - 0/me.x^2 + sd.x*me.y/me.x^3
x <- rnorm(1e4,me.x,sd.x); y <- rnorm(1e4,me.y,sd.y)
c(myx,mean(y/x))
sdyx <- sqrt(sd.x^2*myx^2/me.x^2 + sd.y^2*myx^2/me.y^2)
c(sdyx,sd(y/x))
## 1.113172 1.197855
rvals <- replicate(1000,
sd(rnorm(1e4,me.y,sd.y)/rnorm(1e4,me.x,sd.x)))
hist(log(rvals),col="gray",breaks=100)
abline(v=log(sdyx),col="red",lwd=2)
min(rvals) ## 1.182698
All the canned delta-method approaches to computing the variance of Y/X use the point estimate for Y/X (i.e. m(Y/X) = mY/mX), rather than the second-order approximation you used above. Constructing higher-order forms for both the mean and the variance should be straightforward if possibly tedious (a computer algebra system might help ...)
mvec <- c(x = me.x, y = me.y)
V <- diag(c(sd.x, sd.y)^2)
car::deltaMethod(mvec, "y/x", V)
## Estimate SE
## y/x 1.25 1.047691
library(emdbook)
sqrt(deltavar(y/x,meanval=mvec,Sigma=V)) ## 1.047691
sqrt(sd.x^2*(me.y/me.x)^2/me.x^2 + sd.y^2*(me.y/me.x)^2/me.y^2) ## 1.047691
For what it's worth, I took the code in #SeverinPappadeux's answer and made it into a function gratio(mx,my,sx,sy). For the Cauchy case (gratio(0,0,1,1)) it gets confused and reports a mean of 0 (which should be NA/divergent) but correctly reports the variance/std dev as divergent. For the parameters specified by the OP (gratio(5,4,4,1)) it gives mean=1.352176, sd=NA as above. For the first parameters I tried above (gratio(10,5,1,1)) it gives mean=0.5051581, sd=0.1141726.
These numerical experiments strongly suggest to me that the ratio of Gaussians sometimes has a well-defined variance, but I don't know when (time for another question on Math StackOverflow or CrossValidated?)
Such approximations are unlikely to be useful since the distribution may not have a finite standard deviation. Look at how unstable it is:
set.seed(123)
n <- 10^6
X <- rnorm(n, me.x, sd.x)
Y <- rnorm(n, me.y, sd.y)
sd(head(Y/X, 10^3))
## [1] 1.151261
sd(head(Y/X, 10^4))
## [1] 1.298028
sd(head(Y/X, 10^5))
## [1] 1.527188
sd(Y/X)
## [1] 1.863168
Contrast that with what happens when we try the same thing with a normal random variable:
sd(head(Y, 10^3))
## [1] 3.928038
sd(head(Y, 10^4))
## [1] 3.986802
sd(head(Y, 10^5))
## [1] 3.984113
sd(Y)
## [1] 3.999024
Note: If you were in a different situation, e.g. the denominator has compact support, then you could do this:
library(car)
m <- c(x = me.x, y = me.y)
v <- diag(c(sd.x, sd.y)^2)
deltaMethod(m, "y/x", v)