Learning optimal objective values for MILP