predict_nls: Average predictions from several (non)linear models based on...

In nlraa: Nonlinear Regression for Agricultural Applications

Average predictions from several (non)linear models based on IC weights

Description

Computes weights based on AIC, AICc, or BIC and it generates weighted predictions by the relative value of the IC values

predict function for objects of class gam

Usage

predict_nls(
  ...,
  criteria = c("AIC", "AICc", "BIC"),
  interval = c("none", "confidence", "prediction"),
  level = 0.95,
  nsim = 1000,
  resid.type = c("none", "resample", "normal", "wild"),
  newdata = NULL,
  weights
)

predict2_gam(
  ...,
  criteria = c("AIC", "AICc", "BIC"),
  interval = c("none", "confidence", "prediction"),
  level = 0.95,
  nsim = 1000,
  resid.type = c("none", "resample", "normal", "wild"),
  newdata = NULL,
  weights
)

Arguments

...	‘nls’ or ‘lm’ objects (‘glm’ and ‘gam’ objects inherit ‘lm’).
criteria	either ‘AIC’, ‘AICc’ or ‘BIC’.
interval	either ‘none’, ‘confidence’ or ‘prediction’.
level	probability level for the interval (default 0.95)
nsim	number of simulations to perform for intervals. Default 1000.
resid.type	either ‘none’, “resample”, “normal” or “wild”.
newdata	new data frame for predictions
weights	vector of weights of the same length as the number of models. It should sum up to one and it will override the information-criteria based weights. The weights should match the order of the models.

Value

numeric vector of the same length as the fitted object when interval is equal to ‘none’. Otherwise, a data.frame with columns named (for a 0.95 level) ‘Estimate’, ‘Est.Error’, ‘Q2.5’ and ‘Q97.5’

Note

all the objects should be fitted to the same data. Weights are based on the chosen IC value (exp(-0.5 * delta IC)). For models of class gam there is very limited support.

See Also

predict.lm, predict.nls, predict.gam, simulate_nls, simulate_gam

Examples

## Example
require(ggplot2)
require(mgcv)
data(barley, package = "nlraa")

fm.L <- lm(yield ~ NF, data = barley)
fm.Q <- lm(yield ~ NF + I(NF^2), data = barley)
fm.A <- nls(yield ~ SSasymp(NF, Asym, R0, lrc), data = barley)
fm.LP <- nls(yield ~ SSlinp(NF, a, b, xs), data = barley)
fm.QP <- nls(yield ~ SSquadp3(NF, a, b, c), data = barley)
fm.BL <- nls(yield ~ SSblin(NF, a, b, xs, c), data = barley)
fm.G <- gam(yield ~ s(NF, k = 6), data = barley)

## Print the table with weights
IC_tab(fm.L, fm.Q, fm.A, fm.LP, fm.QP, fm.BL, fm.G)

## Each model prediction is weighted according to their AIC values
prd <- predict_nls(fm.L, fm.Q, fm.A, fm.LP, fm.QP, fm.BL, fm.G)

ggplot(data = barley, aes(x = NF, y = yield)) + 
  geom_point() + 
  geom_line(aes(y = fitted(fm.L), color = "Linear")) +
  geom_line(aes(y = fitted(fm.Q), color = "Quadratic")) +
  geom_line(aes(y = fitted(fm.A), color = "Asymptotic")) +  
  geom_line(aes(y = fitted(fm.LP), color = "Linear-plateau")) + 
  geom_line(aes(y = fitted(fm.QP), color = "Quadratic-plateau")) + 
  geom_line(aes(y = fitted(fm.BL), color = "Bi-linear")) + 
  geom_line(aes(y = fitted(fm.G), color = "GAM")) + 
  geom_line(aes(y = prd, color = "Avg. Model"), size = 1.2)

nlraa documentation built on July 9, 2023, 6:08 p.m.