search

harvard-ufds / saeczi

Small Area Estimation for Continuous Zero Inflated data
https://harvard-ufds.github.io/saeczi/
Other
4 stars 2 forks source link

todo #14

Closed joshyam-k closed 8 months ago

joshyam-k commented 9 months ago

Even though working with N_j x B matrices (where j indexes a domain) helps with memory issues, these objects can often still be too big.

joshyam-k commented 9 months ago

Part of the problem is that since I'm looping through j, the N_j x B matrices all simultaneously exist in memory.

If this part of the procedure:

https://github.com/harvard-ufds/saeczi/blob/b25705002e9ea21f440a4b073256f292a308182e/src/helpers.cpp#L29-L38

happens in a separate c++ function call for each j, will each N_j x B matrix just exist on the stack for the duration of that separate function call? That would mean that only one N_j x B matrix would exist on the stack at a time.