Loo 2.0 with very large stanfit crashes R

[bgoodri]

@Dalton

I mention this because in a similar thread on this topic, it was suggested that the function method of calculating loo is more memory efficient, but every example I’ve seen relies on taking the data and the draws of the parameters to calculate log_lik in R outside of Stan. I don’t want to do that as the list of matrices of parameters will be almost as large as the matrix of log_lik values, plus I already calculated log_lik and went to the trouble of storing it in memory.

This whole “calculate log_lik in generated quantities thing” is for situations where RAM is not an issue. The foo.function method is the more general way of doing it, which is used by rstanarm and brms.

You can write a function that takes data_i, creates a new empty stanfit object that is instantiated with only the i-th observation worth of data, call the constrain_parameters function for each of the S draws, and return that vector.