bsts (version 0.9.5)

compare.bsts.models: Compare bsts models

Description

Produce a set of line plots showing the cumulative absolute one step ahead prediction errors for different models. This plot not only shows which model is doing the best job predicting the data, it highlights regions of the data where the predictions are particularly good or bad.

Usage

CompareBstsModels(model.list,
                  burn = SuggestBurn(.1, model.list[[1]]),
                  filename = "",
                  colors = NULL,
                  lwd = 2,
                  xlab = "Time",
                  main = "",
                  grid = TRUE,
                  cutpoint = NULL)

Arguments

model.list

A list of bsts models.

burn

The number of initial MCMC iterations to remove from each model as burn-in.

filename

A string. If non-empty string then a pdf of the plot will be saved in the specified file.

colors

A vector of colors to use for the different lines in the plot. If NULL then the rainbow pallette will be used.

lwd

The width of the lines to be drawn.

xlab

Label for the horizontal axis.

main

Main title for the plot.

grid

Logical. Should gridlines be drawn in the background?

cutpoint

Either NULL, or an integer giving the observation number used to define a holdout sample. Prediction errors occurring after the cutpoint will be true out of sample errors. If NULL then all prediction errors are "in sample". See the discussion in bsts.prediction.errors.

Value

Invisibly returns the matrix of cumulative one-step ahead prediction errors (the lines in the top panel of the plot). Each row in the matrix corresponds to a model in model.list.

Examples

# NOT RUN {
  data(AirPassengers)
  y <- log(AirPassengers)
  ss <- AddLocalLinearTrend(list(), y)
  trend.only <- bsts(y, ss, niter = 250)

  ss <- AddSeasonal(ss, y, nseasons = 12)
  trend.and.seasonal <- bsts(y, ss, niter = 250)

  CompareBstsModels(list(trend = trend.only,
                         "trend and seasonal" = trend.and.seasonal))

  CompareBstsModels(list(trend = trend.only,
                         "trend and seasonal" = trend.and.seasonal),
                          cutpoint = 100)

# }