- For simplicity, the visualization is given for five subjects (S1–S5). The dashed lines are added to denote that the visualization is limited to a single iteration of the outer loop, visualizing the tuning procedure for left-out test subject S1. For this single iteration of the outer loop, subject 1 (S1) is left-out as a true holdout set. The remaining subjects (S2–S5) are utilized to optimize the network parameters in the inner loop. For each hyperparameter set, the inner loop computes the prediction accuracy by iteratively using each inner loop subject as a holdout validation set. The hyperparameter set that results in the highest average accuracy on the inner loop subjects is utilized to train a model on all subjects of the inner loop (S2–25). This trained model is utilized to compute the metrics and explanations of the left-out test subject (S1). This process is repeated for all subjects
