This project contains the source code used to evaluate the performance of EMF algorithm.
The source code has been structured has follows:
- experiments package contains the main classes to execute all the experiments.
- mf package contains the matrix factorization algorithms implementation not included into CF4J library.
- qualityMeasures package contains MAE and MSE implementation to evaluate the quality of the predictions performed.
The following experiments has been encoded:
- GenerateBinaryFiles. This class splits MovieLens and FilmTrust datasets into training and test sets. Results are stored in a CF4J's binary files. All datasets can be found into datasets directory.
- DatasetInfo. This class prints the main properties of the datasets.
- HyperparametersOptimization. This class performs a grid search to tune the hyper-parameters of the baselines that we are comparing with EMF.
- GeneticProgrammingOptimization. This class performs a genetic programming based optimization in order to discover the most accurate function to perform predictions using Matrix Factorization based Collaboartive Filtering.
- BaselineComparison. This class compares the functions returned by GeneticProgrammingOptimization experiment with several baselines.
The same code has been used to perform experiments within MovieLens and FilmTrust datasets. To switch between both datasets comment/uncomment the corresponding lines.
In lib directory contains the jar version of the libraries required by this project.