Strategies for fitting a large model via AWS EC2 services

[mike-lawrence]

@lnix1

Am I understanding you right that ‘brms’ has not implemented supported yet and so 12 of my cores are going relatively unused (i.e. only one core/two threads are being used per chain)? If that is right, then the best way to improve speed would be to use make_stancode and then modify the resulting stan file to make use of reduce_sum and to utilize the extra compute power?

Yes, your understanding is correct on both points.

@lnix1

Also, thank you for sharing the link to ezStan. The watch_stan() function seems incredibly useful!

Just be careful; there’s a weird bug in RStan (and hence ezStan) on Linux such that failing to constantly watch the csv file as watch_stan() does causes the forked processes to self-terminate. This reminds me I should post a call for help in figuring this one out.

EDIT: FYI it seems RStan 2.21 might have fixed the weird bug I mentioned above; I can’t seem to replicate it anymore.