Skip to content

Instantly share code, notes, and snippets.

Embed
What would you like to do?
Reproducible example for Loo error
## Load libraries
pacman::p_load(tidyverse,brms)
## Load data
data <- read_csv("data.csv")
## Define formulas
Model1_f <- bf(Outcome ~ Diagnosis + (1|ID))
Model2_f <- bf(Outcome ~ Diagnosis + (1|p|ID),
sigma ~ Diagnosis + (1|p|ID))
# Define priors
SkepticPrior1 <- c(
prior(normal(0,.05), class=Intercept),
prior(normal(0,.05), class=b),
prior(normal(0,.01), class=sd)
)
SkepticPrior2 <- c(
prior(normal(0,.05), class=Intercept),
prior(normal(0,.05), class=b),
prior(normal(0,.01), class=sd),
prior(lkj(5), class=cor),
prior(normal(0,.05), class=Intercept, dpar = sigma),
prior(normal(0,.05), class=b, dpar = sigma),
prior(normal(0,.01), class=sd, dpar = sigma)
)
# Run models
m1 <- brm(Model1_f,
data = data,
family = gaussian,
iter = 2000,
chains = 2,
cores = 2,
prior = SkepticPrior1,
sample_prior = T,
file = "Model1",
control= list(adapt_delta = 0.99,
max_treedepth=20))
m1 <- add_criterion(m1, criterion=c("loo"), reloo=F, file = "Model1")
m2 <- brm(Model2_f,
data = data,
family = gaussian,
iter = 2000,
chains = 2,
cores = 2,
prior = SkepticPrior1,
sample_prior = T,
file = "Model2",
control= list(adapt_delta = 0.99,
max_treedepth=20))
m2 <- add_criterion(m1, criterion=c("loo"), reloo=F, file = "Model2")
# Compare, when things stop working
w <- loo_model_weights(m1,m2)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment