fit_sp_model: fit_sp_model

In maxikellerbauer/stAirPol: Spatio-temporal Modelling of Air Pollution Data

Description

Fitting a spatio-temporal Bayesian model. This function is a wrapper for spT.Gibbs. For details on the fitting procedure see spT.Gibbs. This function makes it very esay to use spT.Gibbs.

Usage

fit_sp_model(data, model = "GPP", knots_method = "grid",
  training_set = NULL, knots_plot = FALSE, knots = NULL,
  knots_seed = NULL, knots_count = 5, ...)

Arguments

data	Modelling data.frame which contains information about the covariables and the target variable
model	the spatio-temporal models to be fitted, current choices are: "GP", "truncatedGP", "AR", "GPP", and "truncatedGPP", with the first one as the default.
knots_method	Only used if the model choice is GPP" or "truncatedGPP", knots_method specifies the distribution method of the knots. Default is a random distribution with respect to the location of the sensor. With knots_method = 'grid' the knots will be arranged as a grid.
training_set	an object generated by get_test_and_training_set, if those an object is delivered, the data argument will be reduced to the in training_set specified training sensors.
knots_plot	Should be knots be plotted, Boolean, with FALSE as default
knots	matrix of knots locations
knots_seed	Random seed for the knots sampling
knots_count	How much knots should be used?
...	additional arguments for spT.Gibbs

Value

An object of the class spT, see spT.Gibbs for more details

See Also

spT.Gibbs, spT.initials, spT.geodist, fit_subintervalls, predict.stAirPol.model, predict_split

Examples

data("mini_dataset")
#' remove outliers
mini_dataset <- clean_model_data(mini_dataset)
#' simple formula
formula <-  value ~ rainhist + windhist + trafficvol

#' fit three kind of models and compair the PMCC
model.gp <- fit_sp_model(data = mini_dataset, formula = formula, model = 'GP')
model.ar <- fit_sp_model(data = mini_dataset, formula = formula, model = 'AR')
model.gpp <- fit_sp_model(data = mini_dataset, formula = formula, model = 'GPP')
print(model.gp$PMCC)
print(model.ar$PMCC)
print(model.gpp$PMCC)

#' different options for GPP model
model.gpp <- fit_sp_model(data = mini_dataset, formula = formula, model = 'GPP',
                          knots_method = 'random', knots_plot = TRUE,
                          knots_count = 20, knots_seed = 2202)

#' prioris
prio <- spT.priors(model = "GPP",
                   inv.var.prior = Gamm(0.5, 1),
                   beta.prior = Norm(0, 10^2)
)
model.gpp <- fit_sp_model(data = mini_dataset, formula = formula, model = 'GPP',
                          priors = prio)
#' see ?spTimer::spT.Gibbs for more options like scale.transform, spatial.decay


plot(model.gpp) #' trace plot of the MCMC-Iterations
summary(model.gpp) #' summary
#' perform the modelfit only on a trainingsset with 75% of the values
training_set <- get_test_and_training_set(mini_dataset, sampel_size = 0.75,
                                          random.seed = 220292)
model.gp <- fit_sp_model(data = mini_dataset, formula = formula,
                            model = 'GP', training_set = training_set)

maxikellerbauer/stAirPol documentation built on May 3, 2019, 3:16 p.m.