I’ve found that when using a model with many Cholesky factors, at least one chain will fail in RStan with:
Error in unserialize(socklist[[n]]) : error reading from connection
Error in serialize(data, node$con, xdr = FALSE) :
error writing to connection
Here’s a small model that fails:
parameters{
cholesky_factor_corr[2] y_Lcorr[20];
}
model{
for(c in 1:20)
y_Lcorr[c] ~ lkj_corr_cholesky(1);
}
out = stan("CholRed.stan")
But this doesn’t occur with fewer cholesky parameters, e.g.:
parameters{
cholesky_factor_corr[2] y_Lcorr[10];
}
model{
for(c in 1:10)
y_Lcorr[c] ~ lkj_corr_cholesky(1);
}