Combining additive regression and convolutional neural networks for classifying circulation patterns associated with droughts

Implements semi-structured distributional regression for the classification of circulation patterns associated with droughts.

Structure

The main purpose and most important functions of each subsection are sketched in the following.

Preprocessing and EDA

1. preprocessing.R

• Does data preprocessing, creates both y and data for deepregression, creates class weigths
• Called from: models.R, models_evaluation.R, EDA.R

2. EDA.R

• Does EDA for circulation patterns of interest

Modelling

1. nested_resampling.R

• Contains functions for nested resampling (rep_ho(), nested_resampling_final())
• rep_ho():
  • Implements repreated hold-out using a (train-validation-test split)
  • Returns list of either: (training history per split, list(predictions, confusion matrix, predicted_classes), list(predictions, confusion matrix, predicted_classes)) or (training history per split, list(predictions, confusion matrix, predicted_classes))
  • Called by: nested_resampling_final.R
• nested_resampling_final():
  • Implements repeated hold outsplitting
  • Returns list of return values from rep_ho()

2. models.R

• Initializes various deepregression objects and does nested resampling (nested_resampling_final()) for them
• Saves return values from nested_resampling_final() into .RDS-files

Evaluation

1. performance_evaluation_functions.R

• Contains functions for automated performance evaluation (e.g. create avg. Confusion matrices)
• Saves .tex/.png for confusion matrices/plots generated
• Called from: models_evaluation.R

2. models_evaluation.R

• Reads RDS-files per model and does performance evaluation for performance_evaluation_functions.R for each

Result data

• Model folders: Contain outputs form performance_evaluation_functions.R
• .RDS files: Contain return values from nested_resampling_final() per model