Posterior Predictive for new unseen players in baseball example using partially_pooled model


I was playing around with pyro baseball example and I was wondering if during train_test splitting, I left out some players from the training data. Then, how would I go about making predictions on these new unseen players during testing given some information about them (e.g., lets say we have At-Bats and Hits available for 1 player who was not included in the training data, but now how do we make season prediction on them)?