set.seed(1) if (requireNamespace('ggplot2', quietly = TRUE)) { library(ggplot2) } testLongData = generateLongData( sizes = c(20L, 30L), fixed = Value ~ 1 + time, cluster = ~ 1 + time, random = ~ 1, id = 'id', data = data.frame(time = seq(0, 1, by = .1)), fixedCoefs = c(0, 0), clusterCoefs = cbind(c(-2, 1), c(2, -1)), randomScales = cbind(.1, .1), noiseScales = c(.1, .1), clusterNames = c('A', 'B'), shuffle = TRUE ) %>% .[, .(id, time, Value, Class)] %>% .[, Constant := 1] %>% .[, Cluster := Class] %>% .[, id := factor(id)] %>% .[] # set up capture functions to test for wrong column name handling response = function(...) stop('response column mishandling') cluster = function(...) stop('cluster column mishandling') id = function(...) stop('id column mishandling') Time = function(...) stop('Time name evaluation') Value = function(...) stop('Value name evaluation') Assessment = function(...) stop('Assessment name evaluation') Cluster = function(...) stop('Cluster name evaluation') Traj = function(...) stop('Traj name evaluation')