Empirical errors
Closed this issue · 1 comments
mclaughlin6464 commented
It's well established that the errors produced by the emulator are not reliable. I need to implement a way to reliably return empirical uncertanties. Not even sure where to start tehre.
mclaughlin6464 commented
This was fairly easy to resolve and I've done just that. Now each emu carries a covariance matrix that is the average of the training errors. I was satisfied with tests that showed errors do not vary much with HOD parameters.
However, I will note that this may need a more nuanced change once cosmology is added.