search

pmahon3 / UWO-MOTUS-Analysis

0 stars 0 forks source link

Parameters of signal strength distribution should vary by day ( and other parameter variation questions ) #4

Closed sjbonner closed 3 years ago

sjbonner commented 4 years ago

Data simulation and model currently assume that the mean and variance of the signal strength are identical in the three periods for each day. We should allow these to vary according to some distribution.

pmahon3 commented 4 years ago

Sounds good, I adopted the debug stategy of trying to extend the minimal variation required to cover delta.prime.

pmahon3 commented 3 years ago

@sjbonner piggybacking this issue for the more general question of: how much variation is the right amount?

I've extended some minimal variation to the Population level model from the validated individual model

I'm debating how much variation to add and how much is too much. I think varying signal strength by bird by day is realistic, however should we vary the variance parameters for the signal strength by bird by day? I have similar concerns about varying the variance parameters of individual bird properties that already vary per bird, per day.

Is it ok for hyperparameters to grow exponentially as 2^n for normal distributions which take two arguments?

pmahon3 commented 3 years ago

Regarding variation of signal strength by day, could there be further variation on this hyperparameter with respect to this distribution, although I don't know if it's desirable or not. Right now the mean signal strength, muY[i,j,p] varies per bird per day, but the means from which it is derived, muMuY[p] does not vary per bird. Is this something we want to include or would that be too much? You can see proposed changes here

pmahon3 commented 3 years ago

this is done here https://github.com/pmahon3/UWO-MOTUS-Analysis/commit/297681fcee9292d5526ffaef3982b5d40a86a0d2