Data for Mathematical Copilots: Better Ways of Presenting Proofs for Machine Learning