SLIM Loss Function

[nasimsonboli]

Hi, I have a question about the implementation of SLIM which is according to this paper, Ning, Xia, and George Karypis. "Slim: Sparse linear methods for top-n recommender systems.". In the loss function the error is multiplied by 1/2, but it is not the same in Librec's implementation. Is there a specific reason that I'm missing?

The implementation in Librec: loss += errors + 0.5 * regL2Norm * coefficient * coefficient + regL1Norm * coefficient; the suggestion: loss += 0.5 * errors + 0.5 * regL2Norm * coefficient * coefficient + regL1Norm * coefficient;

Thanks, Nasim