Specifying Optimal Within-Subject Residual Variance-Covariance Structure In Latent Growth Model By Borrowing Power From Machine Learning