Notebooks with data preprocessing steps. This includes processing, filtering and matching steps for the hydroclimatic, geospatial and agronomic data.

This repository contains a tweaked MixedModels.jl accommodating random effect matrices to support GAMMs defined as mixed models. The key add-on is multiplication support for random effect matrices with arbitrary real numbers.

Contains the files required to run a full gammJ procedure. The notebook gammJ_v9_example.ipynb fits a GAMM by leveraging model generation component by gamm4, and then executing them with trainGAM.jl (which leverages the modified MixedModels.jl) and postprocessGAM.jl to compute the psudo data covariance matrix.

Highlights the gamm4 performance bottleneck by fitting multiple models with an increasing amount of random intercept factor levels.

Points to the performance enhanced gamm4 code in notebook format. This code has only been tested for our use case and does not necessarily work for other model confgurations such as different linking functions.

Points to all gammJ execution notebooks for the individual models per milk performance variable (Milk Yield as well as ECM) and breed.

Points to experimental notebooks where a piecewise linear approach from econometrics with pooled regressions is used. These results are biased because of the unbalanced nature of our data.

Points to the LaTex source of the report.