Learners : Matchbox recommender : Command-line runners

Data splitter

The data splitter allows you to easily split the instance data into a training and a test set. It takes a number of parameters which control the splitting algorithm explained in the corresponding data mapping section.

Required parameters

• input-data - the dataset to split
• output-data-train - training part of the split dataset
• output-data-test - test part of the split dataset

Optional parameters

• training-users - fraction of training-only users; defaults to 0.5
• test-user-training-ratings - fraction of test user ratings for training; defaults to 0.25
• cold-users - fraction of cold (test-only) users; defaults to 0
• cold-items - minimum fraction of cold (test-only) items; defaults to 0
• ignored-users - fraction of users not included in either the training or the test set; defaults to 0
• ignored-items - fraction of items not included in either the training or the test set; defaults to 0
• remove-occasional-cold-items - remove occasionally produced cold items that would increase the fraction of cold items above the specified minimum

Example

Learner Recommender SplitData --input-data AllData.dat  
                              --output-data-train TrainingSet.dat   
                              --output-data-test TestSet.dat  
                              --ignored-users 0.8   
                              --training-users 0.1   
                              --remove-occasional-cold-items