ordinalClassificationDTC

The src directory contains the necessary scripts for the complete workflow, from data projection to 1 dimension to the generation of the graphs representing aggregated suborders. The datasets on which to perform linear ordinal classification must be located in nonProjected.

The whole process can be performed by executing mainOCC-DTC.R. The procedure includes the following steps:

1. Retreival of the datasets from the nonProjected directory and transform it into linear. For each class combination new univariate data is defined.
2. Generation of the fold lists: a new directory is created (foldLists) in which a set of 10x10 cross-validation partitions is produced.
3. A further directory, thresholdMaps, is created with the resulting trained support vector machines.
4. In predictionMaps the predicted outcomes, based on the defined threshold, are saved for each class combination.
5. The folder subcascades will contain all the retreieved class sequences organised by class combination and order size, whereas subcascades_threshold contains data with cascades filtered by a class-wise sensitivity >= threshold for each class.
6. In the directory simplified_threshold the ordinally arranged sequences are merged to keep the longer ones which include shorter orders.
7. Finally, the aggregated graphs, representing total orders or potential alternative suborders, are saved in pdf.