data-sampling

Code for tuning undersampling and oversampling

This folder contains the code for tuning the undersampling and oversampling factors for the row-based methods.

This tuning is a grid search over various undersampling and oversampling ratios. The grid search was designed to be run on a computing cluster using slurm job arrays; consequently, only 1 point in the grid is evaluated by the main functions:

• evalSamplingGridRowFeatureMethod for the feature-based methods
• evalSamplingGridRowDataMethod for the data-based methods (1D CNN)

Function descriptions:

• createDataSamplingGrid: creates and saves the data structure for the grid search.
• evalSamplingGridRow{Feature,Data}Method: performs training and validation for 1 grid point.
• rowDataSampling: performs the undersampling and oversampling.
• writeGridSearchResultsToTxtFile: writes results to a text file for easier viewing
• selectBestSamplingParams: determines which grid point had the highest validtion score and saves those parameters