Skip to contents

Tidy an mvgam object's parameter posteriors

Source: R/tidier_methods.R
tidy.mvgam.Rd

Get parameters' posterior statistics, implementing the generic tidy from the package broom.

Usage

# S3 method for mvgam
tidy(x, probs = c(0.025, 0.5, 0.975), ...)

Arguments

x

An object of class mvgam.

probs

The desired probability levels of the parameters' posteriors. Defaults to c(0.025, 0.5, 0.975), i.e. 2.5%, 50%, and 97.5%.

...

Unused, included for generic consistency only.

Value

A tibble containing:

  • "parameter": The parameter in question.

  • "type": The component of the model that the parameter belongs to (see details).

  • "mean": The posterior mean.

  • "sd": The posterior standard deviation.

  • percentile(s): Any percentiles of interest from these posteriors.

Details

The parameters are categorized by the column "type". For instance, the intercept of the observation model (i.e. the "formula" arg to mvgam()) has the "type" "observation_beta". The possible "type"s are:

  • observation_family_extra_param: any extra parameters for your observation model, e.g. sigma for a gaussian observation model. These parameters are not directly derived from the latent trend components (continuing the gaussian example, contrast to mu).

  • observation_beta: betas from your observation model, excluding any smooths. If your formula was y ~ x1 + s(x2, bs='cr'), then your intercept and x1's beta would be categorized as this.

  • random_effect_group_level: Group-level random effects parameters, i.e. the mean and sd of the distribution from which the specific random intercepts/slopes are considered to be drawn from.

  • random_effect_beta: betas for the individual random intercepts/slopes.

  • trend_model_param: parameters from your trend_model.

  • trend_beta: analog of "observation_beta", but for any trend_formula.

  • trend_random_effect_group_level: analog of "random_effect_group_level", but for any trend_formula.

  • trend_random_effect_beta: analog of "random_effect_beta", but for any trend_formula.

Additionally, GP terms can be incorporated in several ways, leading to different "type"s (or absence!):

  • s(bs = "gp"): No parameters returned.

  • gp() in formula: "type" of "observation_param".

  • gp() in trend_formula: "type" of "trend_formula_param".

  • GP() in trend_model: "type" of "trend_model_param".

See also

Other tidiers: augment.mvgam()

Examples

if (FALSE) {
set.seed(0)
simdat <- sim_mvgam(T = 100,
                    n_series = 3,
                    trend_model = AR(),
                    prop_trend = 0.75,
                    family = gaussian())
simdat$data_train$x = rnorm(nrow(simdat$data_train))
simdat$data_train$year_fac = factor(simdat$data_train$year)

mod <- mvgam(y ~ - 1 + s(time, by = series, bs = 'cr', k = 20) + x,
             trend_formula = ~ s(year_fac, bs = 're') - 1,
             trend_model = AR(cor = TRUE),
             family = gaussian(),
             data = simdat$data_train,
             silent = 2)

tidy(mod, probs = c(0.2, 0.5, 0.8))
}