search

skiptoniam / ecomix

ecomix is a package to implement model based species level (Species Archetype Models) or site level (Regions of Common Profile) grouping of community data.
https://skiptoniam.github.io/ecomix/
7 stars 2 forks source link

Species-level predictions species_mix #36

Closed fhui28 closed 2 months ago

fhui28 commented 2 months ago

More a conceptual question than anything (since I am starting to work on the approximate SAMs idea!): I noticed when performing uncertainty quantification (using the bootstrap) for species-level predictions in species_mix, the posterior probabilities are conditioned on i.e., the tau does not change with bootstrapped datasets e.g., tau is a supplied argument in the internal function sam_internal_pred_species.

Should we condition on them, or should we (also) recompute tau for each new bootstrapped dataset since they are, after all, also functions of the parameters. I do not know, but would be curious to hear from you and potentially @Scott-Foster and @pdog-extreme as well...although perhaps you answered this already in another paper?

Cheers.

skiptoniam commented 2 months ago

I think that was on my to do list. I think I originally left tau as the taus from the initial fit because they don't change that much. But I think it's a good idea and we I can look to implementing that.

fhui28 commented 2 months ago

Fair enough...yeah it does not surprise me that it does not change much give that SAMs tend to classify pretty "hard", but sounds like a plan!