examples/basics.R

In kidzik/fcomplete: Trajectory estimation for sparsely observed longitudinal data

library("devtools")
install(".")
library("fcomplete")
library("ggplot2")
library("mclust")

# Load simulated 'toy' data
data = read.table("examples/simulated.Rda")

# Print the data
head(data)

# Run our model
model = fregression(Y ~ time|id, data, bins = 41, K = 3)

# Make some diagnostics plots
idx = 1:5
fcomplete:::plot.fcomplete(model$Y[idx,], model$fit[idx,], NULL,
           filename="pred-mean", title = "True curves & observations")

# Some output elements (basically the SVD decomposition)
model$v # principal components (elements from K + 1 are meaningless, I'll remove them)
model$u # loadings on the first K principal components

# 'patients' plotted on their 2D progression plane (third pattern skipped)
plot(model$u[,1:2])

# 'Progression' trends
matplot(t(model$v[1:3,]),t='l')

## Clustering
fit <- Mclust(model$u)
plot(fit) # plot results
summary(fit) # display the best model

# Get cluster means in functional space
matplot( t(t(fit$parameters$mean) %*% model$v[1:3,]), t='l')