ci: Obtain confidence intervals for l1ADMM objects

In bdsegal/psplinesl1: P-splines with an l_1 penalty for repeated measures

Description

This function obtains Bayesian and Frequentist confidence bands for models fit with the ADMMl1 function

Usage

ci(model, alpha = 0.05, newData = NULL, lenOut = 100)

Arguments

model	output from l1ADMM with cv = FALSE
alpha	confidence level
newData	new dataset at which to get the predicted values and confidence intervals

Examples

library(psplinesl1)
data(simData)

# setup p-spline matrices
X <- list(ps(x = "x", data = simData, 
             norder = 2, k = 1, width = 0.05,
             center = TRUE))

# setup random effect matrices
rand <- re(x = "x", id = "id", data = simData,
           randomCurves = FALSE)

# run cross validation and view paths
cvOut <- cv(y = "y", id = "id", X = X, rand = rand,
                  K = 5,
                  pathLength = 20,
                  data = simData)
plot(cvOut)

# fit model with all data
a1 <- admm(y = "y", X, Z = rand$Z, S = rand$S,
            tau = cvOut$smoothOpt[1],
            lambda = cvOut$smoothOpt[2:(length(X)+1)],
            rho = min(5, max(cvOut$smoothOpt)),
            data = simData)

# get and plot fitted model with confidence bands
CI <- ci(model = a1, alpha = 0.05)
plot(CI)

# extract values from ci object for custom plotting
CIpoly <- data.frame(x = c(CI[[1]]$x, rev(CI[[1]]$x)), 
                     y = c(CI[[1]]$lower, 
                           rev(CI[[1]]$upper)))

ggplot(aes(x = x, y = y), data = newDat)+
  geom_polygon(data = CIpoly, fill = "grey")+
  geom_line(aes(x = CI[[1]]$x, y = CI[[1]]$smooth))

bdsegal/psplinesl1 documentation built on July 22, 2019, 11:38 a.m.