I just realized that using the ratings (for blocks) as I did there isn’t going to work with how I defined blocks. Block indices are only 0,1,2, but each user gave an individual rating for each block leading to n_users*n_blocks ratings. If it was a block-level predictor, there would only be 3 values, right? I could make an index for each block (0-89), but then I wouldn’t be able to compare each first, second and third block across users.