signlogsumexp: sum over signed numbers, with some logsumexp applied to...

In zhenkewu/doubletree: Nested Latent Class Models for Domain-Adaptive and Semisupervised Learning of Cause-of-Deaths using Verbal Autopsy

sum over signed numbers, with some logsumexp applied to positive and negative values separately

Description

sum over signed numbers, with some logsumexp applied to positive and negative values separately

Usage

signlogsumexp(logabsv, sign)

Arguments

logabsv	a vector; positive or negative; log of abs(v)
sign	sign of v, a vector; +1 or -1.

Value

image

Examples

set.seed(123)
x <- rnorm(n = 100, mean = 1000, sd = 10)
comp <- c(log(sum(exp(x))),signlogsumexp(x,rep(1,length(x)))$res,logsumexp(x))
print(comp)

sign_indicators <- c(rep(1,length(x)-50),rep(-1,length(x)-50))
comp2 <- c(log(sum(sign_indicators*exp(x))),signlogsumexp(x,sign_indicators)$res)
print(comp2)

x = c(.5,-.4,-.1)
sum(x) # not zero

signlogsumexp(c(log(abs(x))),sign(x)) # not zero, but much closer to zero.

zhenkewu/doubletree documentation built on Oct. 21, 2023, 7:04 a.m.