Open
Description
Idea:
Ensembles of small models (ESM) have been shown to be one option for estimating habitat suitability for species with few observations. They essentially work by fitting multiple models with 1-n subsets of predictors, which are then evaluated and averaged together.
To implement ESMs in ibis, there is however a necessity to retain some kind of model (not only the ensemble prediction).
Proposal:
- Add option
add_control_esm()for fitting ESMs. This works with linear models only for now (to ensure that coefficients can be reused). - Implement within
train()through repeated calls. Calculate an ensemble of the prediction as well as of the coefficients. Refit a GLM model with the specified coefficients. - Ensure data is recorded in classes and options alike.
- Add unit test for simple check.
References:
- Breiner et al. 2018 https://besjournals.onlinelibrary.wiley.com/doi/full/10.1111/2041-210X.12957