[CNN] All variants should have the same dimension
It is difficult to design a model for different dimensions. So, I end up forcing the models to have the same dimensions after the first convolution block. This is difficult, of course, especially for the number of players, because I can't add zeros because it removes the link between the first player and the last player. I could repeat the rows, but then it would be difficult to express as a layer. The best thing to do would be to replicate it during the encoding phase, so that all variants would have 6 rows, 92 columns and 32 pages.