method-notes.md

Data sources

Insecticide resistance bioassay data

Vgsc allele frequency data

Potential Predictor variables

Modelling

Gaussian process stacked generalization ensemble modelling approach

The idea is to provide level 0 and level 1 approaches

Machine-learning models

Proposed alternative implementation

I propose that this work is first refactored into a targets pipeline. This provides numerous benefits:

• Clearly define the process for the model fitting:
  • data cleaning
  • model formula preparation
  • separating data into test/training sets
  • Fitting model on training data
  • Predicting model out to testing data
  • Combining testing data
• Model outputs are only run once
• Identifying pleasingly parallel components

Additionally, I think that there is some scope to use tidymodels instead of the separate approaches for bgam, xgboost, and randomForest. The benefit of this is that we can unify/simplify the following tasks:

• handling test/training data
• handling training/out of sample prediction
• building formulas for esoteric components of each model