Type: Package
Title: Generalised Additive Models in 'greta' using 'mgcv'
Version: 0.2.0
Description: A 'greta' (Golding (2019) <doi:10.21105/joss.01601>) module that lets you use 'mgcv' smoother functions and formula syntax to define smooth terms for use in a 'greta' model. You can then define your own likelihood to complete the model, and fit it by Markov Chain Monte Carlo (MCMC).
License: Apache License (≥ 2)
Imports: cli, mgcv, rlang, stats
Depends: greta (≥ 0.5.0), R (≥ 4.1.0)
Suggests: knitr, rmarkdown, spelling, testthat (≥ 3.0.0)
Encoding: UTF-8
RoxygenNote: 7.3.2
Config/testthat/edition: 3
URL: https://github.com/greta-dev/greta.gam, https://greta-dev.github.io/greta.gam/
BugReports: https://github.com/greta-dev/greta.gam/issues
VignetteBuilder: knitr
Language: en-GB
NeedsCompilation: no
Packaged: 2024-12-17 00:00:37 UTC; nick
Author: Nick Golding ORCID iD [aut, cph], Nicholas Tierney ORCID iD [aut, cre], David L Miller [aut]
Maintainer: Nicholas Tierney <nicholas.tierney@gmail.com>
Repository: CRAN
Date/Publication: 2024-12-17 10:40:02 UTC

greta.gam: generalised additive models in greta using mgcv

Description

greta.gam is a module for greta that lets you use mgcv's smoother functions and formula syntax to define smooth terms for use in a greta model. You can then define your own likelihood to complete the model, and fit it by MCMC.

Author(s)

Maintainer: Nicholas Tierney nicholas.tierney@gmail.com (ORCID)

Authors:

See Also

Useful links:

evaluate smooths at new data

Description

Evaluate a set of smooths at new data locations

Usage

evaluate_smooths(x, newdata)

Arguments

x

a greta array created with greta.gam::smooths

newdata

a dataframe with the same column names and datatypes as that used to create x, with data at which to evaluate the smooths

Value

greta array

Author(s)

Nick Golding

Examples

## Not run: 
n <- 30
x <- runif(n, 0, 10)
f <- function(x) {
  sin(x * 2) + 1.6 * (x < 3) - 1.4 * (x > 7)
}
y <- f(x) + rnorm(n, 0, 0.3)
x_plot <- seq(0, 10, length.out = 200)

z <- smooths(~ s(x), data = data.frame(x = x))

distribution(y) <- normal(z, 0.3)

z_pred <- evaluate_smooths(z, newdata = data.frame(x = x_plot))

z_pred

## End(Not run)

greta array representations of mgcv smooth terms

Description

smooths translates the right hand side of a mgcv GAM formula into a corresponding Bayesian representation of smooth terms. This formula may include multiple combined smooths of different types, as well as fixed effect terms and intercepts. The resulting greta array representing the combined smooth can then be used in a greta model.

Usage

smooths(formula, data = list(), knots = NULL, sp = NULL, tol = 0)

Arguments

formula

a GAM formula representing the smooth terms, as in mgcv::gam(). Only the right hand side of the formula will be used.

data

a data frame or list containing the covariates required by the formula. These covariates cannot be greta arrays.

knots

an optional list containing user specified knot values to be used for basis construction, as in mgcv::gam() These knots cannot be greta arrays.

sp

an optional vector of smoothing parameters, two per smooth term in the model, in the same order as the formula. If sp = NULL, all smoothing parameters will be learned, otherwise all smoothing parameters must be specified by the user. The smoothing parameters may either be a numeric vector or a greta array (which could be a variable).

tol

a non-negative scalar numerical tolerance parameter. You can try increasing this if the model has numerical stability issues

Details

Only the right hand side of formula will be used to define the smooth terms. The user must complete the gam model by specifying the link and likelihood term in greta. A warning will be issued if the formula has a left hand side.

Note that by default, GAM formulas add an intercept term. If you have already specified an intercept for your greta model, you can remove the intercept from the smooth term by adding -1 as a term in your formula.

Like mgcv::jagam(), smooths translates a mgcv GAM formula into a Bayesian representation of the smooth terms, using the GAM smoothing penalty matrix as a multivariate normal prior to penalise model fitting. Unlike gam, smooths does not perform the integration required to penalise model fitting. The model must be fitted by MCMC to carry out this integration - it does not make sense to do maximum likelihood optimisation on a greta model that uses smooths.

Value

Object of class "greta_array".

Examples

## Not run: 
n <- 30
x <- runif(n, 0, 10)
f <- function(x) {
  sin(x * 2) + 1.6 * (x < 3) - 1.4 * (x > 7)
}
y <- f(x) + rnorm(n, 0, 0.3)
x_plot <- seq(0, 10, length.out = 200)

z <- smooths(~ s(x), data = data.frame(x = x))

distribution(y) <- normal(z, 0.3)

z_pred <- evaluate_smooths(z, newdata = data.frame(x = x_plot))

# build model
m <- model(z_pred)
draws <- mcmc(m, n_samples = 100)

plot(x, y, pch = 19, cex = 0.4, col = "red")
apply(draws[[1]], 1, lines, x = x_plot, col = "blue")
points(x, y, pch = 19, cex = 0.4, col = "red")

## End(Not run)