predict_gam {tidygam}

Get predictions from a GAM model

Description

Return predictions from a GAM model generated with mgcv. The output can be plotted with plot().

Usage

predict_gam(
  model,
  length_out = 10,
  values = NULL,
  series = NULL,
  exclude_terms = NULL,
  ci_z = 1.96,
  tran_fun = NULL,
  separate = NULL,
  sep_by = "\\."
)

Arguments

model

A gam or bam model object.

length_out

An integer indicating how many values to use along the numeric variables for predicting the response (the default is 10).

values

User supplied values for specific variables as a named list.

series

A string specifying the variable that corresponds to the series to be plotted on the $x$-axis. If a string is given, the other numeric variables in the model are set to their mean value, unless specific values are given in values. If a character vector of two strings is given, the two variables will be taken as the elements of a tensor product smooth. This allows the user to plot 2D raster plots.

exclude_terms

Terms to be excluded from the prediction. Term names should be given as they appear in the model summary (for example, "s(x0,x1)").

ci_z

The z-value for calculating the CIs (the default is 1.96 for 95 percent CI).

tran_fun

Function to use for transforming the predicted values and CIs.

separate

Names list of factor interaction variables to be separated.

sep_by

Character to separate by (the default is ⁠\\.⁠).

Value

A tibble with predictions.

Examples

library(mgcv)
set.seed(10)

sim_data_1 <- gamSim(1, n = 200, scale = 2)
model <- gam(y ~ x0 + s(I(x1^2)) + s(x2) + offset(x3), data = sim_data_1)
predict_gam(model)
predict_gam(model, values = list(x0 = mean(sim_data_1$x0)))
predict_gam(model, series = "x2")
predict_gam(model, exclude_terms = "s(I(x1^2))")

# By-variables
sim_data_2 <- gamSim(4)
model_2 <- gam(y ~ s(x2, by = fac) + s(x0), data = sim_data_2)
predict_gam(model_2)

# Poisson data
sim_data_3 <- sim_data_2
sim_data_3$y <- round(sim_data_2$y) + 20
model_3 <- gam(y ~ s(x2, by = fac), data = sim_data_3, family = poisson)
predict_gam(model_3, length_out = 50)
predict_gam(model_3, length_out = 50, tran_fun = exp)

# Bivariate smooths
model_4 <- gam(y ~ te(x1, x2), data = sim_data_1)
predict_gam(model_4)