Fig. 5
From: Implicit-descriptor ligand-based virtual screening by means of collaborative filtering
Hyperparameter tuning for matrix factorization collaborative filtering. Illustration of the effects of the number of factors, f, and regularization parameter, \(\lambda\), on model performance. While model performance improves with more factors, it may lead to over fitting. The regularization parameter does not have a consistent performance, revealing that performance is not sensitive to \(\lambda\).The error bars represent the standard error of the mean across the \(\hbox {BEDROC}_{20}\) scores for each Target at the specified parameter value