tubfs: Bayesian Frequency Selection of time-uncertain data sets

In PeterFranke/tuts: Time Uncertain Time Series Analysis

Description

tubfs spectral analysis of time-uncertain time series using the Bayesian Frequency Selection method described in the paper "Frequency selection in palaeoclimate time series: a model-based approach incorporating possible time uncertainty" by P. Franke, Prof B. Huntley, Dr A. Parnell.

Usage

tubfs(y, ti.mu, ti.sd, n.sim, CV = FALSE, ...)

Arguments

y	A vector of observations.
ti.mu	A vector of estimates/observed timings of observations.
ti.sd	A vector of standard deviations of timings.
n.sim	A number of simulations.
CV	cross-validation indicator.
...	optional arguments. n.chains: number of MCMC chains, default is set to 2. Thin: Thinning factor, default is set to 4.m: maximum number of significant frequencies in the data, the default is set to 5. polyorder: order of the polynomial regression component, the default is set to 3. freqs: set to a positive integer k returns a vector of k equally spaced frequencies in the Nyquist range. freqs can be provided as a vector of custom frequencies of interest. Set to 'internal' generates an equally spaced vector of frequencies in the Nyquist range

Examples

# Import or simulate the data (simulation is chosen for illustrative purposes):
DATA=simtuts(N=50,Harmonics=c(10,30,0), sin.ampl=c(10,10, 0), cos.ampl=c(0,0,0),
trend=0,y.sd=3, ti.sd=1)
y=DATA$observed$y.obs
ti.mu=DATA$observed$ti.obs.tnorm
ti.sd= rep(1, length(ti.mu))

# Run the Bayesian Frequency Selection model with cross validation of the model:
n.sim=1000
BFS=tubfs(y=y,ti.mu=ti.mu,ti.sd=ti.sd,freqs='internal',n.sim=n.sim,n.chains=2, CV=TRUE)

# Generate summary results (optional parameters are listed in brackets):
summary(BFS)                               # Summary results (CI, burn).
summary(BFS,burn=0.2)                      # Results after 20% of burn-in (CI.

# Plots and diagnostics (optional parameters are listed in brackets):
plot(BFS,type='periodogram')               # spectral analysis (CI, burn).
plot(BFS,type='predTUTS', CI=0.99)         # One step predictions of the model (CI, burn).
plot(BFS,type='cv')                        # 5 fold cross validation plot (CI, burn).
plot(BFS,type='GR')                        # Gelman-Rubin diagnostics (CI, burn).
plot(BFS,type='mcmc')                      # mcmc diagnostics.
plot(BFS,type='volatility')                # Volatility realizaitons.

PeterFranke/tuts documentation built on May 30, 2019, 6:24 a.m.