Last updated: 2018-05-05

workflowr checks: (Click a bullet for more information)

  • ✔ R Markdown file: up-to-date

    Great! Since the R Markdown file has been committed to the Git repository, you know the exact version of the code that produced these results.

  • ✔ Environment: empty

    Great job! The global environment was empty. Objects defined in the global environment can affect the analysis in your R Markdown file in unknown ways. For reproduciblity it’s best to always run the code in an empty environment.

  • ✔ Seed: set.seed(20180501)

    The command set.seed(20180501) was run prior to running the code in the R Markdown file. Setting a seed ensures that any results that rely on randomness, e.g. subsampling or permutations, are reproducible.

  • ✔ Session information: recorded

    Great job! Recording the operating system, R version, and package versions is critical for reproducibility.

  • ✔ Repository version: 3bd6a61

    Great! You are using Git for version control. Tracking code development and connecting the code version to the results is critical for reproducibility. The version displayed above was the version of the Git repository at the time these results were generated.

    Note that you need to be careful to ensure that all relevant files for the analysis have been committed to Git prior to generating the results (you can use wflow_publish or wflow_git_commit). workflowr only checks the R Markdown file, but you know if there are other scripts or data files that it depends on. Below is the status of the Git repository when the results were generated: 
    
Ignored files:
    Ignored:    .Rhistory
    Ignored:    .Rproj.user/
    Ignored:    docs/figure/
    Ignored:    log/
    Note that any generated files, e.g. HTML, png, CSS, etc., are not included in this status report because it is ok for generated content to have uncommitted changes.
Expand here to see past versions:
    File Version Author Date Message
    Rmd 3bd6a61 Dongyue 2018-05-05 first commit
    html ca322e8 Dongyue 2018-05-05 first commit

Examine the performance of smash-gen(known \(\sigma\)) under different simulation settings.

Algorithm

Let \(X_t\) be a Poisson observation, \(t=1,2,\dots,T\).

  1. Input \(\sigma\) and initialize \(m_t^{(0)}=\frac{\Sigma_{t=1}^T X_t}{T}\), \(Y_t^{(0)}=\log(m_t^{(0)})+\frac{X_t-m_t^{(0)}}{m_t^{(0)}}\) and \(s_t^{2(0)}=\frac{1}{m_t^{(0)}}\) for \(t=1,2,\dots,T\).
  2. For \(i=1,2,...\), iterate until convergence:
  1. Fit \(Y_t=\mu_t+N(0,\sigma^2)+N(0,s_t^2)\) using smash.gaus and obtain \(\hat\mu_t\).
  2. Update \(m_t^{(i)}=\exp(\hat\mu_t)\), \(Y_t^{(i)}=\log(m_t^{(i)})+\frac{X_t-m_t^{(i)}}{m_t^{(i)}}\), and \(s_t^{2(i)}=\frac{1}{m_t^{(i)}}\)

Convergence criteria: \(||\mu_t^{(i)}-\mu_t^{(i-1)}||_2\leq \epsilon\).

#' smash generaliation function

#' This function is for $Y_t=\mu_t+N(0,s_t^2)+N(0,\sigma^2)$ with known $s_t^2$ and $\sigma^2$.

#' @param x: a vector of observations
#' @param sigma: standard deviations, scalar.
#' @param family: choice of wavelet basis to be used, as in wavethresh.
#' @param niter: number of iterations for IRLS
#' @param tol: criterion to stop the iterations

smash.gen=function(x,sigma,family='DaubExPhase',niter=100,tol=1e-2){
  mu=c()
  s=c()
  mu=rbind(mu,rep(mean(x),length(x)))
  s=rbind(s,rep(1/mu[1],length(x)))
  y=log(mean(x))+(x-mean(x))/mean(x)
  for(i in 1:niter){
    mu.hat=smash.gaus(y,sigma=sigma+s[i,])
    mu=rbind(mu,mu.hat)
    #update m and s_t
    s=rbind(s,1/mu.hat)
    #update y
    mt=exp(mu.hat)
    y=log(mt)+(x-mt)/mt
    #y=log(mu.hat)+(x-mu.hat)/mu.hat
    if(norm(mu.hat-mu[i,],'2')<tol){
      break
    }
  }
  return(list(mu.hat=mu.hat,mu=mu,s=s))
}

Data generation by Poisson glm:

\(\lambda_t=\exp(m_t+\epsilon_t)\), where \(\epsilon_t\sim N(0,\sigma^2)\).

\(X_t\sim Poi(\lambda_t)\).

#' Simulation study comparing smash and smashgen

simu_study=function(m,sigma,seed=1234,
                    niter=100,family='DaubExPhase',tol=1e-2,
                    reflect=FALSE){
  set.seed(seed)
  lamda=exp(m+rnorm(length(m),0,sigma))
  x=rpois(length(m),lamda)
  #fit data
  smash.out=smash.poiss(x,reflect=FALSE)
  smash.gen.out=smash.gen(x,sigma=sigma,niter=niter,family = family,tol=tol)
  return(list(smash.out=smash.out,smash.gen.out=exp(smash.gen.out$mu.hat),smash.gen.est=smash.gen.out,x=x))
}

Simulation 1: Constant trend Poisson nugget

\(\sigma=0.01\)

library(smashr)
m=rep(3,256)
simu.out=simu_study(m,0.01)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))
legend("topright", # places a legend at the appropriate place
       c("truth","smash-gen"), # puts text in the legend
       lty=c(1,1), # gives the legend appropriate symbols (lines)
       lwd=c(1,1),
       cex = 1,
       col=c("black","red", "blue"))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth", "smash"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black", "blue"))

\(\sigma=0.1\)

simu.out=simu_study(m,0.1)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth","smash-gen"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black","red", "blue"))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth", "smash"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black", "blue"))

\(\sigma=0.5\)

simu.out=simu_study(m,0.5)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m),col='gray80')
legend("topright", 
       c("truth","smash-gen"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black","red", "blue"))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m),col='gray80')
legend("topright", 
       c("truth", "smash"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black", "blue"))

\(\sigma=1\)

simu.out=simu_study(m,1)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m),col='black')
legend("topleft", 
       c("truth","smash-gen"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black","red", "blue"))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m),col='black')
legend("topleft", 
       c("truth", "smash"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black", "blue"))

Simulation 2: Step trend

\(\sigma=0.01\)

m=c(rep(3,128), rep(5, 128), rep(6, 128), rep(4, 128))
simu.out=simu_study(m,0.01)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth","smash-gen"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black","red", "blue"))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth", "smash"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black", "blue"))

\(\sigma=0.1\)

simu.out=simu_study(m,0.1)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth","smash-gen"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black","red", "blue"))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth", "smash"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black", "blue"))

\(\sigma=0.5\)

simu.out=simu_study(m,0.5)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth","smash-gen"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black","red", "blue"))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth", "smash"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black", "blue"))

\(\sigma=1\)

simu.out=simu_study(m,1)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))
legend("topleft", 
       c("truth","smash-gen"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black","red", "blue"))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))
legend("topleft", 
       c("truth", "smash"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black", "blue"))

Simulation 3: Oscillating Poisson nugget

Low Oscillating Poisson nugget

\(\sigma=0.01\)

m=c()
for(k in 1:8){
    m=c(m, rep(1,16), rep(5, 16))
}
simu.out=simu_study(m,0.01)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '',main='smash-gen')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))

plot(simu.out$x,col = "gray80" ,ylab = '',main='smash')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))

\(\sigma=0.1\)

simu.out=simu_study(m,0.1)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '',main='smash-gen')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))

plot(simu.out$x,col = "gray80" ,ylab = '',main='smash')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))

\(\sigma=0.5\)

simu.out=simu_study(m,0.5)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '',main='smash-gen')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))

plot(simu.out$x,col = "gray80" ,ylab = '',main='smash')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))

\(\sigma=1\)

simu.out=simu_study(m,1)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '',main='smash-gen')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))

plot(simu.out$x,col = "gray80" ,ylab = '',main='smash')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))

Fast Oscillating Poisson nugget

\(\sigma=0.01\)

m=c()
for(k in 1:32){
    m=c(m, c(1,5))
}
simu.out=simu_study(m,0.01)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '',main='smash-gen')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))

plot(simu.out$x,col = "gray80" ,ylab = '',main='smash')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))

\(\sigma=0.1\)

simu.out=simu_study(m,0.1)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '',main='smash-gen')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))

plot(simu.out$x,col = "gray80" ,ylab = '',main='smash')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))

\(\sigma=0.5\)

simu.out=simu_study(m,0.5)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '',main='smash-gen')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))

plot(simu.out$x,col = "gray80" ,ylab = '',main='smash')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))

\(\sigma=1\)

simu.out=simu_study(m,1)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '',main='smash-gen')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))

plot(simu.out$x,col = "gray80" ,ylab = '',main='smash')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))

Simulation 4: Polynomial curve Poisson nugget

\(\sigma=0.01\)

m = seq(-1,1,length.out = 256)
m = m^3-2*m+1
simu.out=simu_study(m,0.01)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth","smash-gen"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black","red", "blue"))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth", "smash"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black", "blue"))

\(\sigma=0.1\)

m = seq(-1,1,length.out = 256)
m = m^3-2*m+1
simu.out=simu_study(m,0.1)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth","smash-gen"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black","red", "blue"))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth", "smash"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black", "blue"))

\(\sigma=0.5\)

m = seq(-1,1,length.out = 256)
m = m^3-2*m+1
simu.out=simu_study(m,0.5)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth","smash-gen"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black","red", "blue"))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth", "smash"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black", "blue"))

\(\sigma=1\)

m = seq(-1,1,length.out = 256)
m = m^3-2*m+1
simu.out=simu_study(m,1)

#par(mfrow = c(1,2))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.gen.out, col = "red", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth","smash-gen"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black","red", "blue"))

plot(simu.out$x,col = "gray80" ,ylab = '')
lines(simu.out$smash.out, col = "blue", lwd = 2)
lines(exp(m))
legend("topright", 
       c("truth", "smash"), 
       lty=c(1,1), 
       lwd=c(1,1),
       cex = 1,
       col=c("black", "blue"))

Summary

  1. Generally, smash-gen gives more smooth fit under large nugget effect. But sometimes it seems that the iterative algorithm does not converge.
  2. When the nugget effect is small, smash-gen may not perform well especially under oscillating Poisson nugget and polynomial curve.

Session information

sessionInfo()
R version 3.4.0 (2017-04-21)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 16299)

Matrix products: default

locale:
[1] LC_COLLATE=English_United States.1252 
[2] LC_CTYPE=English_United States.1252   
[3] LC_MONETARY=English_United States.1252
[4] LC_NUMERIC=C                          
[5] LC_TIME=English_United States.1252    

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] smashr_1.1-1

loaded via a namespace (and not attached):
 [1] Rcpp_0.12.16        knitr_1.20          whisker_0.3-2      
 [4] magrittr_1.5        workflowr_1.0.1     REBayes_1.3        
 [7] MASS_7.3-47         pscl_1.4.9          doParallel_1.0.11  
[10] SQUAREM_2017.10-1   lattice_0.20-35     foreach_1.4.3      
[13] ashr_2.2-7          stringr_1.3.0       caTools_1.17.1     
[16] tools_3.4.0         parallel_3.4.0      grid_3.4.0         
[19] data.table_1.10.4-3 R.oo_1.21.0         git2r_0.21.0       
[22] iterators_1.0.8     htmltools_0.3.5     assertthat_0.2.0   
[25] yaml_2.1.19         rprojroot_1.3-2     digest_0.6.13      
[28] Matrix_1.2-9        bitops_1.0-6        codetools_0.2-15   
[31] R.utils_2.6.0       evaluate_0.10       rmarkdown_1.8      
[34] wavethresh_4.6.8    stringi_1.1.6       compiler_3.4.0     
[37] Rmosek_8.0.69       backports_1.0.5     R.methodsS3_1.7.1  
[40] truncnorm_1.0-7

This reproducible R Markdown analysis was created with workflowr 1.0.1