The transformer function takes a bamlss.frame object and transforms all smooth.constructs into a random effects representation. Note that this is only possible for smooth terms with a single smoothing variance. The function is based on function smooth2random.

trans_random(x)
randomize(x)

Arguments

x

Object returned from function bamlss.frame.

Details

The decomposition is achieved by a spectral decomposition of the penalty and design matrix by finding a basis of the null space of the penalty matrix. This feature is used, e.g., for the JAGS sampler function. For more details see also jagam.

Value

A transformed bamlss.frame. To each smooth.construct model term an element named "Xf", the fixed effects design matrix, and an element "rand$Xr", the random effects design matrix, is added. In addition, for re-transforming parameters elements

"trans.U" and "trans.D" are supplied. See also function

smooth2random.

References

Fahrmeir L, Kneib T, Lang S, Marx B (2013). Regression - Models, Methods and Applications. Springer-Verlag, Berlin. ISBN 978-3-642-34332-2.

Wood S.N. (2006). Generalized Additive Models: An Introduction with R. Chapman and Hall/CRC.

Examples

## Simulate data.
d <- GAMart()

## Create a "bamlss.frame".
bf <- bamlss.frame(num ~ s(x1) + s(x2) + s(x3) + s(lon,lat), data = d)

## Structure of the "s(x1)" smooth.construct.
str(bf$x$mu$smooth.construct[["s(x1)"]])
#> List of 28
#>  $ term          : chr "x1"
#>  $ bs.dim        : num 10
#>  $ fixed         : logi FALSE
#>  $ dim           : int 1
#>  $ p.order       : num 0
#>  $ by            : chr "NA"
#>  $ label         : chr "s(x1)"
#>  $ xt            :List of 2
#>   ..$ center: logi TRUE
#>   ..$ before: logi TRUE
#>  $ id            : NULL
#>  $ sp            : NULL
#>  $ drop.null     : num 0
#>  $ X             : num [1:500, 1:9] -0.8941 -1.1407 -0.1109 -1.1451 -0.0447 ...
#>  $ S             :List of 1
#>   ..$ : num [1:9, 1:9] 7.73 -1.23 3.914 -0.833 -5.113 ...
#>  $ UZ            : num [1:502, 1:10] 1.052 0.958 0.931 0.928 0.92 ...
#>  $ Xu            : num [1:500, 1] -0.505 -0.494 -0.49 -0.49 -0.489 ...
#>  $ df            : num 9
#>  $ shift         : num [1(1d)] 0.508
#>  $ rank          : num 8
#>  $ null.space.dim: num 1
#>  $ plot.me       : logi TRUE
#>  $ side.constrain: logi TRUE
#>  $ repara        : logi TRUE
#>  $ S.scale       : num 22.3
#>  $ X.dim         : int 9
#>  $ fit.fun       :function (X, b, expand = TRUE, no.sparse.setup = FALSE)  
#>   ..- attr(*, ".internal")= logi TRUE
#>  $ prior         :function (parameters)  
#>   ..- attr(*, "var_prior")= chr "ig"
#>  $ grad          :function (score = NULL, parameters, full = TRUE)  
#>  $ hess          :function (score = NULL, parameters, full = FALSE)  
#>  - attr(*, "class")= chr [1:2] "tprs.smooth" "mgcv.smooth"
#>  - attr(*, "qrc")=List of 4
#>   ..$ qr   : num [1:10, 1] -1.96 -0.135 0.376 0.162 -0.389 ...
#>   ..$ rank : int 1
#>   ..$ qraux: num 1.48
#>   ..$ pivot: int 1
#>   ..- attr(*, "class")= chr "qr"
#>  - attr(*, "nCons")= int 1

## Transform.
bf <- randomize(bf)

## New structure adding fixed
## and random effect matrices.
str(bf$x$mu$smooth.construct[["s(x1)"]])
#> List of 32
#>  $ term          : chr "x1"
#>  $ bs.dim        : num 10
#>  $ fixed         : logi FALSE
#>  $ dim           : int 1
#>  $ p.order       : num 0
#>  $ by            : chr "NA"
#>  $ label         : chr "s(x1)"
#>  $ xt            :List of 2
#>   ..$ center: logi TRUE
#>   ..$ before: logi TRUE
#>  $ id            : NULL
#>  $ sp            : NULL
#>  $ drop.null     : num 0
#>  $ X             : num [1:500, 1:9] -0.8941 -1.1407 -0.1109 -1.1451 -0.0447 ...
#>  $ S             :List of 1
#>   ..$ : num [1:9, 1:9] 7.73 -1.23 3.914 -0.833 -5.113 ...
#>  $ UZ            : num [1:502, 1:10] 1.052 0.958 0.931 0.928 0.92 ...
#>  $ Xu            : num [1:500, 1] -0.505 -0.494 -0.49 -0.49 -0.489 ...
#>  $ df            : num 9
#>  $ shift         : num [1(1d)] 0.508
#>  $ rank          : num 8
#>  $ null.space.dim: num 1
#>  $ plot.me       : logi TRUE
#>  $ side.constrain: logi TRUE
#>  $ repara        : logi TRUE
#>  $ S.scale       : num 22.3
#>  $ X.dim         : int 9
#>  $ fit.fun       :function (X, b, expand = TRUE, no.sparse.setup = FALSE)  
#>   ..- attr(*, ".internal")= logi TRUE
#>  $ prior         :function (parameters)  
#>   ..- attr(*, "var_prior")= chr "ig"
#>  $ grad          :function (score = NULL, parameters, full = TRUE)  
#>  $ hess          :function (score = NULL, parameters, full = FALSE)  
#>  $ Xf            : num [1:500, 1] -0.6793 -1.2461 -0.0775 -1.7152 -0.0343 ...
#>  $ rand          :List of 1
#>   ..$ Xr: num [1:500, 1:8] 0.03749 0.05814 0.00187 0.08334 0.003 ...
#>   .. ..- attr(*, "s.label")= chr "s(x1)"
#>  $ trans.D       : num [1:9] 0.0563 0.0782 0.0812 0.1263 0.1379 ...
#>  $ trans.U       : num [1:9, 1:9] 0.0173 -0.0115 -0.0219 0.0166 0.0362 ...
#>  - attr(*, "class")= chr [1:2] "tprs.smooth" "mgcv.smooth"
#>  - attr(*, "qrc")=List of 4
#>   ..$ qr   : num [1:10, 1] -1.96 -0.135 0.376 0.162 -0.389 ...
#>   ..$ rank : int 1
#>   ..$ qraux: num 1.48
#>   ..$ pivot: int 1
#>   ..- attr(*, "class")= chr "qr"
#>  - attr(*, "nCons")= int 1