Learners : Matchbox recommender : Command-line runners

Trainer

Training is performed using the Train argument to Learner Recommender. It takes as input a training dataset and outputs a serialized trained model, which can be loaded later for making predictions. Training takes in a number of arguments, explained in the Setting up a recommender section. There is more detail on the training procedure in the Training section. Features are not normalized by the algorithm. You will need to do feature encoding and normalization beforehand.

Required parameters

• training-data - training dataset
• trained-model - trained model file

Optional parameters

• traits - number of traits (defaults to 4)
• iterations - number of inference iterations (defaults to 20)
• batches - number of batches to split the training data into (defaults to 1)
• use-user-features - use user features in the model (defaults to False)
• use-item-features - use item features in the model (defaults to False)

Example

Learner Recommender Train --training-data TrainingSet.dat  
                          --trained-model TrainedMatchbox.bin   
                          --traits 5  
                          --iterations 30  
                          --batches 4  
                          --use-user-features   
                          --use-item-features