Preference Learning Toolbox

Loading data

This toolbox can train models from a dataset encoded in two files: the object file and the rank file. Two valid datasets must be loaded before continuing to the next steps.

The object file contains the features that will serve as potential inputs to the models trained. Each line in the file must contain the feature values for one data sample (object) separated by a given character (comma by default). Each object must be identified by an integer ID specified as first feature. Optionally, the first line of the file may contain a textual label for each feature (first one is expected to be 'ID').
The rank file contains the known relations among the data samples in the object file. Each of these relations is specified as an order among a subset of objects and it is encoded as a list of object IDs from the most preferred/highest ranked object to the least preferred/lowest ranked. The object IDs are separated by a given character (comma by default).

For an example dataset, the user is referred to the synthetic datasets available online.