Table 2.

Main dimensions of the three datasets used in our example.

Dataset	ntrain + nval	n test	m	S	v
short-clips	3600	270	2	12	73,221
stories	3737	291	4	10	73,165
short-films	3572	807	22	12	85,483

The dimensions include the number of training time samples n_train + n_val, the number of testing time samples n_test, the number of feature spaces m, the number of cross-validation splits s, and the number of voxels v. All datasets had a large number of voxels (v ≈ 10⁵). The largest number of feature spaces is m = 22 (see Appendix A.2 for a list of the 22 feature spaces).