bmdl: Bayesian MDL (or MDL) for multiple changepoint detection

In yingboli/BayesMDL: Bayesian MDL for Multiple Changepoint Detection

Description

Changes in both mean and linear trend, which permitting a global seasonal mean and AR(p) errors.

Usage

bmdl(x, dates, iter = 10000, thin = max(1, iter/1000), weights = NULL,
  p = 2, time_unit = "month", seasonal_means = "harmonic", k = 3,
  scale_trend_design = 0.05, fit = "marlik", penalty = "bmdl", nu = 5,
  kappa = 3, a = 1, b = 1, start_eta = NULL, start_xi = NULL,
  track_time = TRUE)

Arguments

x	The time series data, a numeric vector of length n.
dates	The dates records are observed, a POSIXlt vector. It should have the mon component for months, and the year component for years. See also strptime.
iter	Total number of MCMC iterations.
thin	Thinning; save one MCMC iteration for every thin number of iterations.
weights	A numeric vector of observation weights, defined the same as the weights argument in the function lm.
p	The order of the AR process.
time_unit	Default is 'month' ('week' may be added in the future).
seasonal_means	The seasonal means variables in the linear model. Either 'fixed_effects' such that each season has a coefficient (with no intercept), or 'harmonic' for harmonic regression (with an intercept).
k	The highest degree of harmonic regression. It is only used if the argument seasonal_means == 'harmonic'.
scale_trend_design	The factor multiplied to the design matrix of trend. Default is 1/50.
fit	For likelihood calculation, 'marlik' for marginal likelihood, or 'lik' for likelihood. Note that the 'lik' option already includes the two-part MDL of mu.
penalty	For penalty function calculation, 'bmdl' for Beta-Binomial prior, or 'mdl' for MDL.
nu	Prior variance scale of mu; only used if fit == 'marlik'.
kappa	Prior variance scale of outliers.
a	The first and second parameters in the Beta-Binomial prior; only used if penalty == 'bmdl'.
b	The first and second parameters in the Beta-Binomial prior; only used if penalty == 'bmdl'.
start_eta	A vector of 0/1 indicators for the initial model, or NULL to randomly sample an initial model.
start_xi	A vector of 0/1 indicators for the initial outliers, or NULL to randomly sample an initial model.
track_time	Logical, whether to show runtime on screen.

Value

best	The optimal model which minimizes BMDL or MDL; a list representing the model, which contains components: eta, inference (output of the fit_eta function), and change_eta.
eta_mcmc	A matrix to save MCMC iterations of eta. Each row is about an iteration, in the format of a vector of length n + p + 2, containing eta, bmdl, phi, sigmasq.
A	The design matrix for the nuisance coefficients in the linear model. It is usually the matrix of seasonal indicators, if the argument seasonal_means = 'fixed_effects', or the design matrix for harmonic regression with a column of all 1 for intercept, if the argument seasonal_means = 'harmonic').
runtime	Runtime, in second.
input_parameters	A list of input parameters.

yingboli/BayesMDL documentation built on May 29, 2019, 12:18 p.m.