i have trained a small bayesian neural network by splitting a dataset in the usual training and test dataset. Now i also have an analytical model whose parameters needs to be estimated by mcmc methods on the same dataset. i would like to compare the quality of the predictions. should i split the dataset into test and training set for the analytical model too? i don't see the reason to have a test dataset for a model whose parameters are found through Monte Carlo methods, but then how should i compare the quality of the predictions? should i use the entire dataset or split it?