Constructs Gaussian process term metadata

Takes a Gaussian process term as defined by gp() and returns the metadata required to wire the term into a Stan model. Like construct_arima(), this does not modify the data or produce design matrix columns; the Gaussian process enters the linear predictor through a Hilbert-space reduced-rank approximation (see inst/stan/functions/gaussian_process.stan).

Usage

construct_gp(gp, data)

Arguments

gp

A Gaussian process term as defined by gp().

data

A data.frame of observations used to define the term. Must contain the time and (if specified) grouping variable.

Value

A list with the following elements:

• time, by, kernel, gp_type, nu, d, basis_prop, boundary_scale: passed through from the gp() term.

• T: number of distinct time points in the integrated series.

• G: number of groups (1 if by is unspecified).

• M: number of basis functions, ceiling(basis_prop * (T - d)).

• PHI: the (T - d) x M basis matrix. For d >= 1 the basis is built on the T - d free values that are integrated d times in Stan; the first d values of the realisation are anchored to zero.

• time_idx, group_idx: per-observation lookup indices.

• time_vals, group_levels: lookup vectors so the indices can be inverted.

• name: a label for the term, suitable as a parameter prefix.

See also

Functions used to help convert formulas into model designs ar(), arima(), arima_terms(), arma(), as_string_formula(), construct_arima(), construct_re(), construct_rw(), enw_formula(), enw_manual_formula(), gp(), gp_terms(), ma(), parse_formula(), re(), remove_arima_terms(), remove_gp_terms(), remove_rw_terms(), rw(), rw_terms(), split_formula_to_terms()

Examples

data <- enw_example("preproc")$metareference[[1]]
epinowcast:::construct_gp(gp(week), data)
#> $time
#> [1] "week"
#> 
#> $by
#> NULL
#> 
#> $kernel
#> [1] "matern32"
#> 
#> $gp_type
#> [1] 2
#> 
#> $nu
#> [1] 1.5
#> 
#> $d
#> [1] 0
#> 
#> $basis_prop
#> [1] 0.2
#> 
#> $boundary_scale
#> [1] 1.5
#> 
#> $T
#> [1] 6
#> 
#> $G
#> [1] 1
#> 
#> $M
#> [1] 2
#> 
#> $PHI
#>           [,1]       [,2]
#> [1,] 0.4082483  0.7071068
#> [2,] 0.6605596  0.7765344
#> [3,] 0.7986542  0.3320991
#> [4,] 0.7986542 -0.3320991
#> [5,] 0.6605596 -0.7765344
#> [6,] 0.4082483 -0.7071068
#> 
#> $time_idx
#>  [1] 1 1 1 1 1 1 1 2 2 2 2 2 2 2 3 3 3 3 3 3 3 4 4 4 4 4 4 4 5 5 5 5 5 5 5 6 6 6
#> [39] 6 6
#> 
#> $group_idx
#>  [1] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
#> [39] 1 1
#> 
#> $time_vals
#> [1] 0 1 2 3 4 5
#> 
#> $group_levels
#> [1] "1"
#> 
#> $name
#> [1] "gp__week"
#> 
epinowcast:::construct_gp(gp(week, day_of_week, kernel = "se"), data)
#> $time
#> [1] "week"
#> 
#> $by
#> [1] "day_of_week"
#> 
#> $kernel
#> [1] "se"
#> 
#> $gp_type
#> [1] 0
#> 
#> $nu
#> [1] 1.5
#> 
#> $d
#> [1] 0
#> 
#> $basis_prop
#> [1] 0.2
#> 
#> $boundary_scale
#> [1] 1.5
#> 
#> $T
#> [1] 6
#> 
#> $G
#> [1] 7
#> 
#> $M
#> [1] 2
#> 
#> $PHI
#>           [,1]       [,2]
#> [1,] 0.4082483  0.7071068
#> [2,] 0.6605596  0.7765344
#> [3,] 0.7986542  0.3320991
#> [4,] 0.7986542 -0.3320991
#> [5,] 0.6605596 -0.7765344
#> [6,] 0.4082483 -0.7071068
#> 
#> $time_idx
#>  [1] 1 1 1 1 1 1 1 2 2 2 2 2 2 2 3 3 3 3 3 3 3 4 4 4 4 4 4 4 5 5 5 5 5 5 5 6 6 6
#> [39] 6 6
#> 
#> $group_idx
#>  [1] 7 5 1 3 4 2 6 7 5 1 3 4 2 6 7 5 1 3 4 2 6 7 5 1 3 4 2 6 7 5 1 3 4 2 6 7 5 1
#> [39] 3 4
#> 
#> $time_vals
#> [1] 0 1 2 3 4 5
#> 
#> $group_levels
#> [1] "Friday"    "Monday"    "Saturday"  "Sunday"    "Thursday"  "Tuesday"  
#> [7] "Wednesday"
#> 
#> $name
#> [1] "gp__week__day_of_week"
#>