“For hyperparameter tuning, random search is more effective than grid search because neural networks are often more sensitive to some parameters than others.”